Fix profile loading with ConfigKit history

.gitignore

@@ -2,6 +2,8 @@
 .codex
 AGENTS.md
 
+.DS_Store
+
 # ---> Go
 # If you prefer the allow list template instead of the deny list, see community template:
 # https://github.com/github/gitignore/blob/main/community/Golang/Go.AllowList.gitignore
@@ -22,6 +24,9 @@ AGENTS.md
 # Dependency directories (remove the comment below to include it)
 # vendor/
 
+# Go cache
+.gocache
+
 # Go workspace file
 go.work
 go.work.sum

.woodpecker/release.yml

@@ -2,39 +2,26 @@ when:
   - event: tag
 
 steps:
-  - name: build-release-assets
-    image: golang:1.25
+  validate-release:
+    image: golang:1.26.7
+    commands:
+      - ./scripts/check-release-candidate.sh "$CI_COMMIT_TAG"
+
+  build-release-assets:
+    image: golang:1.26.7
+    depends_on:
+      - validate-release
     commands:
       - |
         set -eu
+        case "$PWD" in
+          /*) ;;
+          *) echo "release workspace must have an absolute path" >&2; exit 1 ;;
+        esac
+        ./scripts/build-release-assets.sh "$CI_COMMIT_TAG" "$PWD/dist"
 
-        version="$CI_COMMIT_TAG"
-        dist="dist"
-        pkg="gitea.maximumdirect.net/eric/narratio/cmd/narratio"
-
-        rm -rf "$dist"
-        mkdir -p "$dist"
-
-        build_binary() {
-          goos="$1"
-          goarch="$2"
-          suffix="$3"
-          output="$dist/narratio-$version-$goos-$goarch$suffix"
-
-          CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" \
-            go build -trimpath -ldflags "-s -w -X gitea.maximumdirect.net/eric/narratio/internal/buildinfo.Version=$version" \
-            -o "$output" "$pkg"
-        }
-
-        build_binary linux amd64 ""
-        build_binary linux arm64 ""
-        build_binary darwin amd64 ""
-        build_binary darwin arm64 ""
-        build_binary windows amd64 ".exe"
-        build_binary windows arm64 ".exe"
-
-  - name: publish-release
-    image: woodpeckerci/plugin-release
+  publish-release:
+    image: woodpeckerci/plugin-release:0.3.1
     depends_on:
       - build-release-assets
     settings:
@@ -42,6 +29,8 @@ steps:
         from_secret: GITEA_RELEASE_TOKEN
       files:
         - dist/narratio-*
+      title: Narratio ${CI_COMMIT_TAG}
+      note: docs/releases/${CI_COMMIT_TAG}.md
       checksum: sha256
       checksum-file: SHA256SUMS
       checksum-flatten: true

.woodpecker/shuffle.yml

@@ -0,0 +1,8 @@
+when:
+  - event: cron
+
+steps:
+  shuffled-race-tests:
+    image: golang:1.26
+    commands:
+      - go test -race -shuffle=on -count=3 ./...

.woodpecker/verify.yml

@@ -0,0 +1,47 @@
+when:
+  - event: [push, pull_request]
+
+steps:
+  tests:
+    image: golang:1.26
+    commands:
+      - go test ./...
+
+  race-tests:
+    image: golang:1.26
+    depends_on: tests
+    commands:
+      - go test -race ./...
+
+  static-analysis:
+    image: golang:1.26
+    depends_on: tests
+    commands:
+      - go vet ./...
+
+  build:
+    image: golang:1.26
+    depends_on: tests
+    commands:
+      - go build ./...
+
+  documentation-and-examples:
+    image: golang:1.26
+    depends_on: tests
+    commands:
+      - go test ./internal/doccheck
+      - go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+
+  cross-build:
+    image: golang:1.26
+    depends_on: [race-tests, static-analysis, build, documentation-and-examples]
+    commands:
+      - |
+        set -eu
+        output_dir="$(mktemp -d)"
+        trap 'rm -rf "$output_dir"' EXIT
+        for target in linux/amd64 linux/arm64 darwin/arm64; do
+          goos="${target%/*}"
+          goarch="${target#*/}"
+          CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" go build -o "$output_dir/narratio-$goos-$goarch" ./cmd/narratio
+        done

README.md

@@ -1,279 +1,42 @@
 # narratio
 
-`narratio` is a Go orchestration application for processing D&D session audio into transcripts and generated artifacts.
+Narratio is a stage-driven Go orchestrator for turning D&D session audio into
+polished transcripts, validated Notarius extraction lanes, and generated
+artifacts.
 
-## Current Implementation
+It runs a deterministic workflow with manifest-driven continuation, remote
+publish, and restore support.
 
-Implemented now:
-
-- strict config loading/validation (`pipeline.yml` and `session.yml`)
-- local workspace/session layout, locking, and manifest persistence
-- resumable stage control (`run`, `plan`, `resume`, `run-stage`, `status`)
-- real `prepare`, `transcribe`, `merge`, `polish`, `normalize`, `trim`, and `analyze` stages
-- real WhisperX, Seriatim, and Audita adapters
-- real Scriptorium subprocess adapter
-- optional Scriptorium render diagnostics (`render_debug`)
-
-Not implemented yet:
-
-- `archive` stage behavior
-- `notify` stage behavior
-- additional analyze artifacts beyond `session_recap`
-- generic DAG orchestration
-
-## Config Files
-
-Narratio expects two YAML files:
-
-- `pipeline.yml`: pipeline/workspace settings
-- `session.yml`: per-session settings
-
-YAML decoding is strict (`KnownFields(true)`), so unknown fields fail fast.
-
-## Canonical Stage Order
-
-1. `prepare`
-2. `transcribe`
-3. `merge`
-4. `polish`
-5. `normalize`
-6. `trim`
-7. `analyze`
-8. `archive`
-9. `notify`
-
-## Transcript Tiers
-
-- `transcripts/merged.json`: canonical deterministic merged transcript from Seriatim merge
-- `transcripts/processed.json`: full raw Audita-polished transcript output
-- `transcripts/normalized.json`: Seriatim-normalized transcript from the normalize stage
-- `transcripts/trimmed.json`: gameplay-only normalized polished transcript from trim stage
-
-## Normalize Configuration
-
-`pipeline.normalize` is optional. When omitted, Narratio defaults to:
-
-- `output_path: transcripts/normalized.json`
-- `output_schema: seriatim-intermediate`
-- `report: true`
-
-Allowed `normalize.output_schema` values:
-
-- `seriatim-minimal`
-- `seriatim-intermediate`
-- `seriatim-full`
-
-`normalize.output_path` is treated as session-workdir-relative when not absolute.
-
-Normalize stage behavior summary:
-
-- normalize runs after `polish` and before `trim`
-- normalize resolves `transcripts/processed.json`
-- normalize runs Seriatim `normalize` to produce `transcripts/normalized.json`
-- normalize diagnostics are written to:
-  - `artifacts/seriatim.normalize.report.json` (when enabled)
-  - `logs/seriatim.normalize.stdout.log`
-  - `logs/seriatim.normalize.stderr.log`
-  - `config/seriatim.normalize.generated.yml`
-
-## Trim Configuration
-
-`pipeline.trim` is optional. If omitted, no trim config is loaded. If `trim.enabled` is omitted, it defaults to `false`.
-
-When `trim.enabled: true`:
-
-- `trim.output_path` is required
-- `trim.bounds.prompt_id` is required
-- `trim.bounds.transcript_input_name` is required
-- `trim.bounds.output_path` is required
-- `trim.bounds.timeout` must be a valid Go duration when provided
-- `trim.bounds.render_debug: true` requires `trim.bounds.render_output_path`
-- `trim.bounds.profile_id` may be empty to use the prompt default profile
-
-Trim paths are treated as session-workdir-relative when not absolute.
-
-Example trim config:
-
-```yaml
-trim:
-  enabled: true
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
+```sh
+narratio run 2026-04-04
 ```
 
-Trim behavior summary:
+This requires resolvable `pipeline.yml`, `campaign.yml`, and concrete
+`session.yml` files or their explicit command-line alternatives.
 
-- trim discovers and validates `transcripts/normalized.json`
-- trim uses Scriptorium bounds (`dnd_session.bounds` by example config) to produce `artifacts/session_bounds.json`
-- bounds IDs are validated against the same normalized transcript ID space that Seriatim trim will consume
-- trim converts bounds to Seriatim keep selector (for example `10-868`) and runs Seriatim trim
-- if trim is disabled, Narratio copies normalized transcript to trimmed transcript and records `trim_action=copy_disabled`
+## Documentation
 
-Trim outputs and diagnostics:
+- [CLI reference](docs/cli.md) — commands, arguments, flags, and invocation
+  behavior.
+- [Configuration](docs/config.md) — discovery, fields, defaults, and
+  validation.
+- [Operations](docs/operations.md) — runtime workflow, state, publishing,
+  recovery, and cleanup.
+- [Troubleshooting](docs/troubleshooting.md) — symptom-driven diagnosis and
+  safe remedies.
+- [Integration contracts](docs/integrations/) — external tools, formats, and
+  compatibility expectations.
+- [Maintained examples](examples/README.md) — complete copyable configuration
+  and input files, including the production/testing split bundle.
 
-- `artifacts/session_bounds.json`
-- `transcripts/trimmed.json`
-- `logs/scriptorium.bounds.stdout.log`
-- `logs/scriptorium.bounds.stderr.log`
-- `config/scriptorium.bounds.generated.yml`
-- `logs/seriatim.trim.stdout.log`
-- `logs/seriatim.trim.stderr.log`
-- `config/seriatim.trim.generated.yml`
-- optional bounds render-debug outputs:
-  - `artifacts/session_bounds.render.json`
-  - `logs/scriptorium.bounds.render.stdout.log`
-  - `logs/scriptorium.bounds.render.stderr.log`
-  - `config/scriptorium.bounds.render.generated.yml`
+## Maintainer Documentation
 
-Render-debug files are diagnostics and are not treated as canonical stage output artifact refs.
-
-## Scriptorium Configuration
-
-`pipeline.scriptorium` is optional. When present, Narratio validates and uses it for analyze-stage artifact generation.
-
-Key points:
-
-- `scriptorium.binary` is required when section is present
-- `scriptorium.config_path` is optional
-- `scriptorium.timeout` defaults to `10m` when omitted
-- `scriptorium.render_debug` enables render diagnostics globally
-- artifacts are configured under `scriptorium.artifacts` (map shape supports multiple artifacts)
-- enabled artifacts require `prompt_id` and `output_path`
-- artifact `render_debug` may override global render setting
-- `vars` currently support boolean and string values
-
-Example `session_recap` artifact definition:
-
-```yaml
-scriptorium:
-  binary: "scriptorium"
-  config_path: "/etc/scriptorium/config.yml"
-  timeout: "10m"
-  render_debug: false
-
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: "dnd.session_recap"
-      profile_id: "local-quality" # optional
-      output_path: "artifacts/session_recap.md"
-      timeout: "10m"
-      # render_debug: true # optional per-artifact override
-
-      inputs:
-        transcript:
-          source: "trimmed_transcript"
-          required: true
-
-        previous_recap:
-          source: "previous_session_artifact"
-          artifact: "session_recap"
-          path: "" # optional; set when available
-          required: false
-
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: "session_recap"
-```
-
-Prompt IDs and profile IDs are configuration values. They are not hardcoded in analyze-stage logic.
-
-Do not put secrets in `pipeline.yml`. If API-key behavior is configured, use env var names only.
-
-## Scriptorium Runtime Behavior
-
-Narratio integrates with Scriptorium through the public CLI subprocess contract:
-
-- generation: `scriptorium run`
-- diagnostics/testing: `scriptorium render --format json` when `render_debug` is enabled
-
-For the initial implementation, only `session_recap` generation is supported.
-
-Analyze-stage session recap behavior:
-
-- available transcript input sources for configured artifacts: `processed_transcript`, `normalized_transcript`, `trimmed_transcript`
-- session recap should use gameplay-only transcript input (`source: trimmed_transcript`)
-- Narratio resolves `trimmed_transcript` from trim manifest output (`transcript_trimmed`) or fallback `transcripts/trimmed.json`
-- Narratio resolves `normalized_transcript` from normalize manifest output (`transcript_normalized`) or fallback `transcripts/normalized.json`
-- missing trimmed transcript fails clearly and advises running trim stage first
-- `normalized_transcript` is the preferred full-transcript source for future table/meta-analysis artifacts
-- `processed_transcript` remains supported for advanced/debug use cases
-- optionally includes `previous_recap` when configured and resolvable
-- omits optional previous recap when unavailable
-- fails if required inputs are missing
-- validates output file exists and is non-empty
-
-Expected session output paths:
-
-- `artifacts/session_recap.md`
-- `logs/scriptorium.session_recap.stdout.log`
-- `logs/scriptorium.session_recap.stderr.log`
-- `config/scriptorium.session_recap.generated.yml`
-- `artifacts/session_recap.render.json` when render diagnostics are enabled
-
-## Examples
-
-Starter files:
-
-- `examples/pipeline.minimal.yml`
-- `examples/session.minimal.yml`
-- `examples/speakers.yml`
-
-## Commands
-
-Run tests:
-
-```bash
-go test ./...
-```
-
-Plan a run:
-
-```bash
-go run ./cmd/narratio plan --config examples/pipeline.minimal.yml --session examples/session.minimal.yml
-```
-
-Run full pipeline:
-
-```bash
-go run ./cmd/narratio run --config examples/pipeline.minimal.yml --session examples/session.minimal.yml
-```
-
-Run analyze only:
-
-```bash
-go run ./cmd/narratio run-stage --config examples/pipeline.minimal.yml --session examples/session.minimal.yml analyze
-```
-
-## Operational Note
-
-Checksum-based stale detection is not implemented yet.
-
-If prepared inputs or prompt/runtime config change, rerun the appropriate upstream stages before relying on downstream artifacts.
-
-Examples:
-
-- glossary/autocorrect/speaker-context changes: rerun at least `merge`, `polish`, `normalize`, `trim`, and `analyze`
-- trim bounds prompt/profile/config changes: rerun at least `normalize`, `trim`, and `analyze`
-- session recap prompt/profile/input-source changes: rerun `analyze`
-
-## Roadmap
-
-Near-term roadmap:
-
-- extend analyze to additional configured artifacts
-- support workflows where later artifacts consume earlier generated artifacts
-- keep orchestration explicit without a generic DAG engine
-- implement archive and notify backends
+- [Development guide](docs/development.md) — first-read orientation and
+  task-specific reading routes.
+- [Internal overview](docs/internal/overview.md) — implemented component map.
+- [Architecture](docs/policy/architecture.md) — normative boundaries and
+  invariants.
+- [Documentation policy](docs/policy/documentation.md) — canonical ownership
+  and maintenance rules.
+- [Testing policy](docs/policy/testing.md) — test value, boundaries, and
+  sufficiency.

architecture.md

@@ -1,316 +0,0 @@
-# Narratio Architecture
-
-## 1. Purpose
-
-`narratio` is a Go orchestrator for D&D session processing. It runs a stage-based local pipeline from audio input through transcript processing and artifact generation, with manifest-based skip/force/resume behavior.
-
-Narratio integrates with Scriptorium through the **public CLI** (`scriptorium run` and `scriptorium render`) via synchronous subprocess execution.
-
-## 2. Current Status
-
-Implemented:
-
-- strict `pipeline.yml` + `session.yml` loading with strict YAML field checking (`KnownFields(true)`)
-- local workspace/session layout, lock file handling, artifact path helpers, checksums, and atomic writes
-- manifest store and stage status transitions for resumable runs
-- real `prepare`, `transcribe`, `merge`, and `polish` stages
-- real WhisperX HTTP adapter
-- real Seriatim subprocess adapter
-- real Audita subprocess adapter
-- real Scriptorium subprocess adapter
-- real `normalize` stage producing `transcripts/normalized.json`
-- real `trim` stage producing `transcripts/trimmed.json`
-- real `analyze` stage for initial `session_recap` generation
-- optional Scriptorium render diagnostics (`render_debug`) before production run
-
-Still placeholder/future:
-
-- `archive` stage behavior
-- `notify` stage behavior
-- additional Scriptorium artifact types beyond `session_recap`
-- artifact-to-artifact workflows beyond the initial single-artifact implementation
-- generic stale detection based on input/config checksums
-
-## 3. Pipeline and Stage Boundaries
-
-Canonical stage order:
-
-1. `prepare`
-2. `transcribe`
-3. `merge`
-4. `polish`
-5. `normalize`
-6. `trim`
-7. `analyze`
-8. `archive`
-9. `notify`
-
-Boundary rules:
-
-- orchestration logic lives in `internal/app`
-- stage business logic lives in `internal/stage`
-- external-tool CLI construction lives in adapter packages
-- Scriptorium CLI details stay in `internal/adapters/scriptorium`
-
-## 4. Scriptorium Integration Model
-
-Integration mode:
-
-- public CLI subprocesses only (no Scriptorium internal Go packages, no HTTP API)
-- production generation uses `scriptorium run`
-- diagnostics/testing render uses `scriptorium render --format json`
-
-Run invocation shape used by adapter:
-
-```bash
-scriptorium run --prompt <prompt_id> --input name=path --out <output_path>
-```
-
-Optional flags passed when configured:
-
-- `--config <path>`
-- `--profile <profile_id>`
-- repeated `--var name=value`
-- repeated `--input name=path`
-- `--timeout <duration>`
-- `--api-key-env <ENV_NAME>` when configured
-
-Render invocation shape used by adapter:
-
-```bash
-scriptorium render --prompt <prompt_id> --input name=path --format json --out <render_output_path>
-```
-
-Adapter behavior:
-
-- always passes `--out`
-- captures stdout/stderr separately
-- writes generated invocation metadata YAML (redacted, no secrets)
-- treats exit code `0` as success
-- treats exit code `1` as failure
-- treats exit code `2` as failure with `validation_failed=true` and preserves output metadata when available
-- validates successful output files exist and are non-empty
-- does not treat non-empty stderr as failure by itself
-
-## 5. Configuration Contract
-
-`pipeline.scriptorium` is optional. Existing pipelines without Scriptorium continue to work.
-
-`pipeline.trim` is optional. Existing pipelines without trim config continue to work.
-
-`pipeline.normalize` is optional. Existing pipelines without normalize config continue to work.
-
-When `pipeline.normalize` is omitted, defaults are applied:
-
-- `output_path: transcripts/normalized.json`
-- `output_schema: seriatim-intermediate`
-- `report: true`
-
-When `pipeline.normalize` is present:
-
-- `output_path` must be non-empty
-- `output_schema` must be one of `seriatim-minimal`, `seriatim-intermediate`, or `seriatim-full`
-- relative `output_path` values are session-workdir-relative paths
-- Seriatim binary settings still come from `pipeline.seriatim`
-
-When `pipeline.trim` is present:
-
-- `enabled` is optional and defaults to `false` when omitted
-- relative `output_path`, `bounds.output_path`, and `bounds.render_output_path` values are session-workdir-relative paths
-- do not store secrets in trim config values
-
-When `pipeline.trim.enabled: true`:
-
-- `output_path` is required and non-empty
-- `bounds.prompt_id` is required and non-empty
-- `bounds.transcript_input_name` is required and non-empty
-- `bounds.output_path` is required and non-empty
-- `bounds.timeout` must parse as a Go duration when provided
-- `bounds.render_debug: true` requires non-empty `bounds.render_output_path`
-- `bounds.profile_id` may be empty to use the prompt default profile
-- prompt IDs are config values, not hardcoded stage logic
-
-When `pipeline.scriptorium` is present:
-
-- `binary` is required and non-empty
-- `config_path` is optional; when provided it must be non-empty
-- `timeout` is optional; when provided it must parse as a Go duration
-- default `timeout` is `10m`
-- unknown YAML fields fail strict decode
-
-Artifacts are configured as a map under `pipeline.scriptorium.artifacts` so multiple artifacts are possible in the config shape.
-
-For each artifact definition:
-
-- `enabled: true` requires non-empty `prompt_id`
-- `enabled: true` requires non-empty `output_path`
-- `timeout` must parse as Go duration when present
-- optional per-artifact `render_debug` may override global `scriptorium.render_debug`
-- `inputs` are named and each input requires non-empty `source`
-- inputs may be optional (`required: false`)
-- `vars` values currently support `string` and `bool`
-
-Prompt IDs and profile IDs are configuration values, not hardcoded stage logic.
-
-Trim config shape:
-
-```yaml
-trim:
-  enabled: true
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
-```
-
-## 6. Transcript Tiers
-
-Narratio currently produces and uses four transcript tiers:
-
-- `transcripts/merged.json`: canonical deterministic merged transcript from Seriatim merge
-- `transcripts/processed.json`: full raw Audita-polished transcript output (includes pre/post-game content)
-- `transcripts/normalized.json`: normalized transcript generated by Seriatim normalize
-- `transcripts/trimmed.json`: gameplay-only normalized polished transcript from trim stage
-
-Trim reads `transcripts/normalized.json`, validates bounds IDs against that same transcript ID space, and writes `transcripts/trimmed.json`.
-
-## 7. Normalize Stage (Current Implementation)
-
-Normalize stage behavior:
-
-- stage order position: after `polish` and before `trim`
-- discovers processed transcript from manifest polish outputs (`transcript_processed`) when present, else `work/<session_id>/transcripts/processed.json`
-- validates processed transcript JSON shape (`segments` array required)
-- runs Seriatim `normalize` to produce normalized transcript
-- validates normalized transcript JSON shape (`segments` array required)
-- validates normalize report JSON when enabled
-
-Expected normalize outputs and diagnostics:
-
-- `transcripts/normalized.json`
-- `artifacts/seriatim.normalize.report.json` (when normalize report is enabled)
-- `logs/seriatim.normalize.stdout.log`
-- `logs/seriatim.normalize.stderr.log`
-- `config/seriatim.normalize.generated.yml`
-
-## 8. Trim Stage (Current Implementation)
-
-Trim stage behavior:
-
-- stage order position: after `normalize` and before `analyze`
-- discovers normalized transcript from manifest normalize outputs (`transcript_normalized`) when present, else `work/<session_id>/transcripts/normalized.json`
-- validates normalized transcript JSON shape (`segments` array required)
-- when `trim.enabled: false` (or trim config omitted), deterministically copies normalized transcript to `transcripts/trimmed.json` and records `trim_action=copy_disabled`
-- when `trim.enabled: true`:
-  - runs Scriptorium bounds prompt using configured `trim.bounds.prompt_id`
-  - writes bounds output to configured path (typically `artifacts/session_bounds.json`)
-  - parses and validates bounds output against the same normalized transcript being trimmed
-  - converts bounds range to Seriatim keep selector (for example `10-868`)
-  - runs Seriatim `trim` to produce `transcripts/trimmed.json`
-  - supports no-trim bounds actions (`none`/`copy`) by copying normalized transcript unchanged
-- validates trimmed transcript JSON shape (`segments` array required)
-
-Expected trim outputs and diagnostics:
-
-- `artifacts/session_bounds.json`
-- `transcripts/trimmed.json`
-- `logs/scriptorium.bounds.stdout.log`
-- `logs/scriptorium.bounds.stderr.log`
-- `config/scriptorium.bounds.generated.yml`
-- `logs/seriatim.trim.stdout.log`
-- `logs/seriatim.trim.stderr.log`
-- `config/seriatim.trim.generated.yml`
-- optional bounds render-debug outputs when enabled:
-  - `artifacts/session_bounds.render.json`
-  - `logs/scriptorium.bounds.render.stdout.log`
-  - `logs/scriptorium.bounds.render.stderr.log`
-  - `config/scriptorium.bounds.render.generated.yml`
-
-Render-debug files are diagnostics. They are recorded in stage metadata/log/config refs and are not treated as canonical stage output artifact refs.
-
-## 9. Analyze Stage (Current Implementation)
-
-The current real analyze implementation supports only `scriptorium.artifacts.session_recap`.
-
-Behavior:
-
-- if `pipeline.scriptorium` is missing, analyze returns a skipped result with metadata
-- if no Scriptorium artifacts are enabled, analyze returns a skipped result with metadata
-- if enabled artifacts exist but `session_recap` is not enabled, analyze fails clearly
-- available transcript input sources for configured artifacts: `processed_transcript`, `normalized_transcript`, `trimmed_transcript`
-- `session_recap` should use `trimmed_transcript` input (`transcripts/trimmed.json`) for in-universe recap generation
-- `trimmed_transcript` input is resolved from manifest (`trim` output kind `transcript_trimmed`) when available, otherwise fallback path `work/<session_id>/transcripts/trimmed.json`
-- `normalized_transcript` input is resolved from manifest (`normalize` output kind `transcript_normalized`) when available, otherwise fallback path `work/<session_id>/transcripts/normalized.json`
-- `processed_transcript` input is resolved from manifest (`polish` output kind `transcript_processed`) when available, otherwise fallback path `work/<session_id>/transcripts/processed.json`
-- `normalized_transcript` is the preferred full-transcript source for future table/meta-analysis artifacts
-- `processed_transcript` remains available for advanced/debug use cases
-- transcript inputs are validated as JSON with top-level `segments` array
-- configured inputs are resolved by source
-- optional `previous_recap` is omitted when unavailable
-- required `previous_recap` fails before invocation when unavailable
-- vars are built from config + session metadata
-- `render_debug` controls pre-run `scriptorium render` diagnostics
-- render failure stops stage before production run
-- render output is validated as JSON
-- production call uses Scriptorium adapter `RunArtifact`
-- successful run output must exist and be non-empty
-- missing `trimmed_transcript` input for configured `trimmed_transcript` source fails clearly with guidance to run trim stage first
-- manifest records output refs, logs, generated config paths, and non-secret provenance metadata
-
-## 10. Session Recap Paths
-
-Current expected paths for `session_recap`:
-
-- artifact output: `artifacts/session_recap.md`
-- run stdout log: `logs/scriptorium.session_recap.stdout.log`
-- run stderr log: `logs/scriptorium.session_recap.stderr.log`
-- run generated invocation/config: `config/scriptorium.session_recap.generated.yml`
-- render output (when enabled): `artifacts/session_recap.render.json`
-- render stdout log: `logs/scriptorium.session_recap.render.stdout.log`
-- render stderr log: `logs/scriptorium.session_recap.render.stderr.log`
-- render generated invocation/config: `config/scriptorium.session_recap.render.generated.yml`
-
-## 11. Security and Privacy
-
-- do not store secrets in pipeline YAML, generated invocation YAML, logs, or manifest metadata
-- if API-key integration is configured, pass env var names only (never raw key values)
-- avoid logging transcript content or rendered prompt content by default
-- treat generated artifacts and logs as potentially sensitive session material
-
-## 12. Operational Caveat (Pre-Stale-Detection)
-
-Checksum-based stale detection is not implemented yet.
-
-If prepared inputs or prompt/runtime configuration change (for example glossary files, prompt IDs, profile IDs, or relevant pipeline settings), rerun the appropriate prior stages to refresh downstream artifacts.
-
-Examples:
-
-- glossary or autocorrect changes usually require rerunning at least `merge`, `polish`, `normalize`, `trim`, and `analyze`
-- trim prompt/profile changes require rerunning at least `normalize`, `trim`, and `analyze`
-- session recap prompt/profile/input-source changes require rerunning `analyze`
-
-## 13. Roadmap
-
-Planned next steps:
-
-- extend analyze beyond `session_recap` to additional configured artifacts
-- support artifact inputs that consume prior generated artifacts
-- keep this composable without adding a generic DAG engine in the near term
-- implement real `archive` backend behavior
-- implement real `notify` backend behavior
-- add checksum-based stale detection and stale transitions
-
-Architectural invariants remain:
-
-- strict config decoding/validation
-- manifest-driven run control
-- clear stage/adapter separation
-- configuration-driven prompt/profile/input/vars/output mapping
-- Scriptorium integration through public CLI subprocess contract

docs/cli.md

@@ -0,0 +1,472 @@
+# CLI Reference
+
+## Shortest Useful Command
+
+```bash
+narratio run 2026-04-04
+```
+
+This runs the canonical full pipeline for session `2026-04-04`.
+
+## Command Overview
+
+Top-level commands:
+
+- `version`: print the Narratio build version.
+- `run <session_id>`: run all or one contiguous range of the canonical stage order.
+- `regenerate-artifacts <session_id>`: force-run extraction through analysis.
+- `run-stage <stage> <session_id>`: run one stage.
+- `analyze <session_id>`: force-run analyze.
+- `publish <session_id>`: force-run publish.
+- `clean <session_id>` or `clean --all`: remove local work/spool state.
+- `session <subcommand>`: session helper commands.
+- `config <subcommand>`: validate, display, source-trace, catalog, or compare resolved pipeline configuration.
+
+Session subcommands:
+
+- `session init <session_id>`
+- `session plan <session_id>`
+- `session validate <session_id>`
+- `session status <session_id>`
+- `session restore <session_id>`
+- `session artifacts <session_id>`
+- `session locks <session_id>`
+- `session locks add <session_id> <source>`
+- `session locks remove <session_id> <source>`
+
+## Common Config Flags
+
+Most session-aware commands accept:
+
+- `--config <pipeline.yml>`
+- `--campaign <id>`
+- `--campaign-file <campaign.yml>`
+- `--session <session.yml>`
+- `--session-id <session_id>`
+- `--previous-session-id <session_id>`
+- `--profile <name>`
+- `--secrets-dir <directory>`
+
+Rules:
+
+- `--campaign` and `--campaign-file` are mutually exclusive.
+- `--session` is not used by `session init`.
+- if both positional `<session_id>` and `--session-id` are provided, values must match.
+- `--previous-session-id` is a strict expectation: the selected session file
+  must contain the same `previous_session_id`.
+- `--profile` selects `profiles/<name>.yml` relative to the pipeline root. It
+  may be supplied once; omitted, empty, and the exact name `default` select the
+  base configuration, while an unknown named profile fails. The same selection
+  applies to all common-flag commands, including `regenerate-artifacts`.
+- `--secrets-dir` selects ConfigKit credential-file bootstrap explicitly. A
+  relative value uses caller CWD. Without the flag, a present `secrets/`
+  directory beside the resolved root pipeline is used; an absent one is skipped.
+- `clean --all` cannot be combined with campaign/session selectors.
+- notification delivery is currently limited to the configured `noop` mode; see
+  the [configuration reference](./config.md#notifications).
+
+## Session ID Input Rules
+
+Session-aware commands accept one of these forms:
+
+- positional session ID: `... <session_id>`
+- compatibility flag: `... --session-id <session_id>`
+
+When both are present, command parsing requires an exact match.
+
+Commands with additional positionals keep their command-specific order:
+
+- `run-stage <stage> <session_id>` or `run-stage <stage> --session-id <session_id>`
+- `session locks add <session_id> <source>` or `session locks add --session-id <session_id> <source>`
+- `session locks remove <session_id> <source>` or `session locks remove --session-id <session_id> <source>`
+
+## Command Reference
+
+### `config validate`, `config show`, `config sources`, `config artifacts`, and `config diff`
+
+```bash
+narratio config validate [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config show [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config sources [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config artifacts [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config diff <left-profile> <right-profile> [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--secrets-dir <directory>]
+```
+
+These commands resolve the selected profile, defaults, ordinary paths, and—if
+a campaign is selected—the campaign-owned party. They neither discover or load
+a session nor create a workspace, manifest, run, lock, adapter, or remote
+connection. Like every pipeline-loading command, they bootstrap the selected or
+conventional secrets directory before resolving configuration; `config show`
+never includes bootstrap reports or secret values in its YAML output.
+
+Campaign selection is optional for a pipeline without party-driven artifact
+families. A pipeline with `scriptorium.artifact_families` needs a selected or
+configured default campaign so Narratio can expand its concrete artifacts and
+publish rules. `--campaign` and `--campaign-file` remain mutually exclusive.
+Session, range, force, and artifact-execution flags are not accepted.
+
+`config validate` writes a concise root-path, selected-profile (or `none`), and
+effective-digest summary after successful complete validation. `config show`
+writes one deterministic, secret-free YAML document containing defaulted and
+expanded concrete configuration. It omits ConfigKit directives, artifact
+family declarations, and runtime provenance.
+
+`config sources` reports the same fully validated resolution without printing
+effective values. Its header identifies the root, ordered imports, selected
+profile and overlay, selected campaign, party source, and digest. The
+remaining tab-separated records are sorted as `path`, `role`, and `source`.
+Roles distinguish root, import, profile, centralized default, campaign, party,
+and generated family ownership. A generated party member has
+one family record and one party record at the same logical path. The output
+never reads or prints secret values.
+
+`config artifacts` prints the concrete catalog used by analysis, ordered by
+artifact key. Its tab-separated header is `key`, `source`, `definitions`,
+`enabled`, `output`, `dependencies`, `family`, `character`, `family_source`, and
+`party_source`. Definitions and effective dependencies are sorted comma-separated
+lists, and control characters in fields are escaped. Family-member rows identify
+their family, character, defining family file, and party file. An empty catalog
+prints only the header. Ordinary artifacts can be inspected without a campaign;
+party-driven families retain the campaign requirement. See
+[Scriptorium Artifact Entries](./config.md#scriptorium-artifact-entries) for the
+configuration layout.
+
+`config diff` resolves both supplied profile names independently and compares
+their fully resolved, secret-free effective mappings. It does
+not accept `--profile`; the two positional names must be distinct, declared
+profiles. When party-driven families are present, both profiles must resolve to
+the same selected campaign and party. Use `--campaign-file` if profile-specific
+campaign configuration would otherwise select different files.
+
+Equal profiles print `no differences`. Otherwise, sorted tab-separated records
+use one of these forms, with compact JSON values:
+
+```text
+added    <path>    <right-value>
+removed  <path>    <left-value>
+changed  <path>    <left-value>    <right-value>
+```
+
+Mappings are flattened to their logical field paths; lists remain one atomic
+value. The command compares defaulted concrete artifacts and publish rules, not
+profile names, source-file layout, or formatting. It succeeds when differences
+are found, making it suitable for review and migration checks.
+
+### `version`
+
+```bash
+narratio version
+```
+
+Official release binaries report their exact Git tag. Binaries built directly
+from source without release linker metadata report `dev`.
+
+### `run`
+
+```bash
+narratio run <session_id> [--from <stage>] [--through <stage>] [--force] [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Behavior:
+
+- evaluates one inclusive contiguous range of the canonical stage order;
+- defaults an omitted `--from` to `prepare` and an omitted `--through` to
+  `cleanup`, so omitting both retains full-pipeline behavior;
+- rejects unknown endpoints and a `--from` endpoint after `--through`;
+- runs `render` before `extract`; an omitted or disabled Notarius
+  configuration records an explicit `notarius_disabled` self-skip;
+- skips already-succeeded stages unless `--force` is set or a stage-specific
+  resume check finds its durable result obsolete;
+- applies `--force` only to stages in the selected range;
+- rejects repeated `--from`, `--through`, or `--force` options, including
+  `--name=value` spellings;
+- continues interrupted or partially completed sessions by running non-succeeded stages;
+- writes the session manifest as the sole durable progress ledger. Each
+  invocation also creates its run-local directory and returns an immediate
+  command summary; neither creates another durable progress record.
+- reports the resolved profile (or `none`) and effective configuration digest.
+
+When `--artifacts` is present, the selected range must contain `analyze` or
+`publish`. Either consumer is sufficient, including a one-stage range.
+
+### `regenerate-artifacts`
+
+```bash
+narratio regenerate-artifacts <session_id> [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Exactly equivalent to:
+
+```bash
+narratio run <session_id> --force --from extract --through analyze [caller options]
+```
+
+The command always reruns extraction. Analysis rebuilds the selected configured
+artifacts and any prerequisites required by those targets; without
+`--artifacts`, it uses the normal default analysis selection. Publish and notify
+never run. Common session/configuration options and repeatable artifact values
+pass through unchanged.
+
+Because the expansion owns `--force`, `--from`, and `--through`, callers cannot
+supply those options. The shared `run` parser reports them as duplicate
+singleton flags. The alias has no private execution options or behavior, and
+runtime diagnostics may identify the operation as `run`.
+
+### `run-stage`
+
+```bash
+narratio run-stage <stage> <session_id> [--force] [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Valid stage names:
+
+- `prepare`
+- `transcribe`
+- `merge`
+- `polish`
+- `normalize`
+- `trim`
+- `render`
+- `extract`
+- `analyze`
+- `publish`
+- `notify`
+- `cleanup`
+
+Rules:
+
+- `--artifacts` is accepted only for `analyze` and `publish` stage targets.
+
+### `analyze`
+
+```bash
+narratio analyze <session_id> [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Equivalent to:
+
+```bash
+narratio run-stage analyze <session_id> --force [...common config flags]
+```
+
+### `publish`
+
+```bash
+narratio publish <session_id> [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Equivalent to:
+
+```bash
+narratio run-stage publish <session_id> --force [...common config flags]
+```
+
+This command selects only `publish`; it does not invoke cleanup. Use a normal
+full run to include configured cleanup, or `run-stage cleanup` to retry cleanup
+from valid publication evidence.
+
+### `clean`
+
+```bash
+narratio clean <session_id> [--dry-run] [--clear-cache] [...common config flags]
+narratio clean --all [--dry-run] [--clear-cache] [--config <pipeline.yml>]
+```
+
+Behavior:
+
+- session mode removes the selected session's local work and spool state;
+- `--all` removes all local session work and spool state;
+- cache remains unless `--clear-cache` is provided.
+
+See [Operations: Cleanup](./operations.md#cleanup) for manual and pipeline-stage
+deletion scope.
+
+### `session plan`
+
+```bash
+narratio session plan <session_id> [--from <stage>] [--through <stage>] [--force] [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Uses the same inclusive bounds, endpoint validation, force scope, and artifact
+selection contract as `run`. It validates config and prints run/skip decisions
+for selected stages only without creating the local workdir or changing the
+manifest. Resume-capable selected stages are checked against durable evidence.
+The output includes the resolved profile (or `none`) and effective configuration
+digest without writing provenance or any manifest state.
+For `analyze`, the preview also lists explicit targets, prerequisite-only work,
+execution order, and reusable current artifacts with concise reasons. These
+artifact decisions come from the same reconciliation and work planner used by
+execution; the preview does not predict output identities.
+
+### `session validate`
+
+```bash
+narratio session validate <session_id> [...common config flags]
+```
+
+Read-only preflight checks for config validity, required inputs, audio mode,
+previous-session requirements, publish outputs, and independent remote locks.
+
+### `session status`
+
+```bash
+narratio session status <session_id> [...common config flags]
+```
+
+Prints local manifest state and, when storage is available, status for the
+pointer-selected remote commit and its declared published outputs. Current
+publish metadata reports `published`, `carried_locked`, `locked_unpublished`,
+`optional_missing`, and `unselected`; carried output lines include the run
+derived from their exact declared key, the key itself, and its declared
+destination. A separately displayed `dest` is the destination requested by the
+current configuration and may differ while an older declaration is locked.
+
+### `session init`
+
+```bash
+narratio session init <session_id> --output ./session.yml [options]
+narratio session init <session_id> --remote [options]
+```
+
+Required target selection:
+
+- exactly one of:
+  - `--output <path>`
+  - `--remote`
+
+Options:
+
+- `--config <pipeline.yml>`
+- `--campaign <id>` or `--campaign-file <campaign.yml>`
+- `--secrets-dir <directory>`
+- `--previous-session-id <id>`
+- `--date <YYYY-MM-DD>`
+- `--title <text>`
+- `--audio-dir <path>`
+- `--audio-s3-prefix <prefix>`
+- `--force`
+
+Rules:
+
+- `--audio-dir` and `--audio-s3-prefix` are mutually exclusive.
+- if campaign `session_template_file` is configured, `session init` renders it.
+- generated session YAML must be concrete (no unresolved `{{ ... }}` placeholders).
+
+### `session restore`
+
+```bash
+narratio session restore <session_id> [--dry-run] [--force] [--include-audio] [...common config flags]
+```
+
+Behavior:
+
+- discovers committed remote current state;
+- plans local restores;
+- writes an execution report;
+- blocks unresolved conflicts. `--force` permits replacement only of eligible
+  regular files.
+
+See [Operations: Restore Workflow](./operations.md#restore-workflow) for the
+default restore scope, report location, and conflict-handling workflow.
+
+### `session artifacts`
+
+```bash
+narratio session artifacts <session_id> [--remote] [...common config flags]
+```
+
+Lists effective built-in, configured Scriptorium, and configured extraction
+sources; reports planned, available, unavailable, and published state without
+reading payload bodies; and includes publish rules, lock state, and optional
+remote published-state availability. Remote inspection uses exact commit
+declarations and displays the five current publish outcomes when recorded.
+
+### `session locks`
+
+```bash
+narratio session locks <session_id> [...common config flags]
+narratio session locks add <session_id> <source> [--reason <text>] [--force] [...common config flags]
+narratio session locks remove <session_id> <source> [...common config flags]
+```
+
+Behavior:
+
+- list mode validates and reports independent remote locks sorted by source;
+- add creates one lock and reports an existing lock unless `--force` is used;
+- `--force` performs one generation-matched replacement and reports a
+  concurrent change for operator retry; and
+- remove is idempotent and reports when the source was not locked.
+
+See [Operations: Publish Locks](./operations.md#publish-locks) for lock storage
+and publication behavior.
+
+## `--artifacts` Selection Rules
+
+An artifact-family key selects all of its concrete character members. A
+concrete generated key selects only that member; mixed family and concrete
+selection is deduplicated and executed as concrete keys. The resulting plan
+and command output identify both the concrete key and, where applicable, its
+family and character ID.
+
+- accepted on `run`, `session plan`, `run-stage`, `analyze`, and `publish`;
+- repeatable and comma-separated values are combined, surrounding whitespace
+  is removed, and duplicate names are collapsed;
+- names must exist in `pipeline.scriptorium.artifacts`;
+- empty entries are invalid;
+- on `run-stage`, only `analyze` and `publish` accept the option.
+
+Effects:
+
+- selects explicit analyze targets; required configured prerequisites may be
+  reused or rebuilt before them;
+- filters publish rules that source `narratio.artifact.<name>`;
+- does not filter built-in transcript/bounds or explicitly configured
+  `narratio.extraction.<name>` publish sources; and
+- does not select or filter Notarius lanes.
+
+## Common Workflows
+
+Run full pipeline:
+
+```bash
+narratio run 2026-04-04
+```
+
+Dry-run restore plan:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Generate a concrete session file from template/default structure:
+
+```bash
+narratio session init 2026-04-04 --output ./session.yml --date 2026-04-04 --title "Session 12"
+```
+
+Force publish only:
+
+```bash
+narratio publish 2026-04-04
+```
+
+Regenerate post-transcript artifacts without publishing:
+
+```bash
+narratio regenerate-artifacts 2026-04-04 --artifacts session_recap,player_handout
+```
+
+## Output And Exit Behavior
+
+- Successful commands write their result or summary to standard output and
+  exit with status `0`.
+- Command failures and invalid invocations write an error to standard error and
+  exit with status `1`.
+- An unknown top-level command also prints the top-level usage summary to
+  standard error.
+- `session restore --help` prints its command-specific usage and exits with
+  status `0`.
+
+Output is intended for operator inspection. Narratio does not currently offer
+a machine-readable CLI output mode; durable machine-readable state is recorded
+in manifests and reports described in [Operations](./operations.md).

docs/config.md

@@ -0,0 +1,579 @@
+# Configuration Reference
+
+## Purpose
+
+Narratio resolves three YAML documents:
+
+- `pipeline.yml`: pipeline/runtime settings
+- `campaign.yml`: campaign identity and stable input defaults
+- `session.yml`: session identity, metadata, and audio source selection
+
+## Discovery and Selection
+
+### `pipeline.yml`
+
+When `--config` is omitted, search order is:
+
+1. `/usr/local/etc/narratio/pipeline.yml`
+2. `/etc/narratio/pipeline.yml`
+
+### `campaign.yml`
+
+Selection rules:
+
+- if `--campaign-file` is set, use that path;
+- else if `--campaign <id>` is set, use `{pipeline.campaigns.root}/{id}/campaign.yml`;
+- else use `{pipeline.campaigns.root}/{pipeline.campaigns.default_campaign_id}/campaign.yml`.
+
+### `session.yml`
+
+When `--session` is omitted, local search order is:
+
+1. `/usr/local/etc/narratio/session.yml`
+2. `/etc/narratio/session.yml`
+
+If local session discovery fails and a `session_id` is known, Narratio attempts remote session loading from:
+
+- `{root_prefix}/campaigns/{campaign}/sessions/{session_id}/session.yml`
+
+using configured object storage.
+
+The downloaded remote session file is command-scoped: Narratio removes it after
+the command finishes and records only the remote object provenance alongside
+the durable copied session input.
+
+### Read-only effective pipeline inspection
+
+`narratio config validate`, `narratio config show`, `narratio config sources`,
+and `narratio config artifacts` use the same `--config`, `--campaign`,
+`--campaign-file`, and
+`--profile` selection rules as pipeline commands, but do not select, discover,
+or load a session. They resolve configured typed credential references as part
+of the ordinary ConfigKit load but expose only their environment-variable names;
+they do not create runtime state.
+`narratio config diff <left-profile> <right-profile>` uses the same pipeline and
+campaign selectors, resolves each named profile with an independent ConfigKit
+load, and does not accept a separate `--profile` flag.
+
+Campaign selection is optional only when the resolved pipeline has no
+`scriptorium.artifact_families`. When families are declared, Narratio selects a
+campaign through an explicit flag or `pipeline.campaigns.default_campaign_id`,
+then parses the campaign-owned party and expands concrete artifacts and any
+family publish rules before validation. `config validate` prints the resulting
+root, profile, and effective digest. `config show` emits the normalized
+effective pipeline YAML, with defaults and concrete expansion included but
+ConfigKit directives and family declarations omitted. This review projection is
+not promised to be a reloadable authoring document. `config sources` prints a stable
+source projection instead of effective values: root/import/profile/default
+ownership plus campaign/party and generated-family records. Canonical derived
+players trace to the party. `config artifacts` lists the concrete artifact map,
+effective dependencies, and defining sources. The [CLI reference](cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
+owns command syntax and output conventions.
+
+`config diff` compares normalized field values rather than YAML text or source
+ownership. It emits sorted `added`, `removed`, and `changed` records, uses
+compact deterministic JSON values, treats lists atomically, and reports `no
+differences` when the complete effective configurations are equal. Concrete
+family members and generated publish rules participate after expansion; moving
+an equal value between eligible root/import sources does not create a
+difference.
+
+### Migrating to the maintained bundle
+
+Use the [production/testing bundle](../examples/production-testing/pipeline.yml)
+as the complete copyable migration reference. Keep production settings in the
+base, split them into explicit additive imports, and place testing differences
+in the fixed `profiles/testing.yml` overlay. Put ordinary artifacts and families
+in the recommended one-definition files shown by that bundle. Convert campaign
+rosters to [canonical party input](integrations/party.md), then express character
+work as families. Inspect the result with `config validate`, `config show`,
+`config sources`, and `config artifacts`; use `config diff default testing` to
+review the overlay before running a session. Unversioned party documents are
+rejected.
+
+### Identity segments
+
+Campaign IDs (`campaign_id` and `default_campaign_id`), session IDs, previous
+session IDs, and Narratio run IDs are opaque portable segments. They must use
+only ASCII letters, digits, `.`, `_`, and `-`; empty values, `.`/`..`, path
+separators, drive forms, whitespace, control characters, and non-ASCII text are
+rejected. Narratio does not trim or rewrite these values. Existing manifests or
+remote state with an unsafe legacy identity must be migrated before use.
+
+## Validation and Merge Rules
+
+- ConfigKit performs strict typed YAML decoding and accepts exactly one document;
+  unknown fields, duplicate keys, and trailing documents fail load.
+- A pipeline file may explicitly import additive YAML fragments through the
+  root-only `configkit.imports` list. Imported files contribute fields to one
+  logical pipeline document; they do not override fields supplied by the root
+  or another import.
+- An omitted profile, an empty profile option, and the exact name `default`
+  select the base configuration. Any other name selects
+  `profiles/<name>.yml` relative to the pipeline root.
+- Campaign, session, and canonical-party documents are strict single-source
+  ConfigKit schemas. They do not enable imports, profiles, or `configkit`
+  directives.
+- Configured timeout and retry-delay durations must be positive. An omitted
+  artifact timeout continues to inherit its configured Scriptorium timeout.
+- Session files must be concrete; unresolved `{{ ... }}` placeholders fail load.
+- Session and party inputs inspected before typed loading are limited to 8 MiB.
+- Pipeline defaults are applied before validation.
+- Defaults apply only to absent fields. Explicit empty strings, zero values,
+  `false`, and `null` retain ConfigKit's typed meaning and are then checked by
+  Narratio domain validation; they are not silently replaced by defaults.
+  Explicit null optional stage sections stay absent, while required sections
+  fail validation.
+- Campaign and session identities must agree.
+- Required stable files (`speakers_file`, `autocorrect_file`, `glossary_file`,
+  `party_file`) are campaign-owned; the first three may be overridden by the
+  session. The optional `spell_catalog_file` also inherits a campaign value
+  when omitted by the session.
+- `party_file` is parsed when pipeline and campaign configuration are combined.
+  It must be a versioned [canonical party](integrations/party.md), is
+  campaign-owned and derives the players input internally. `players_file` and
+  session `party_file` are unsupported configuration keys.
+- Exactly one audio mode must be configured in session input:
+  - local (`audio_dir` or `audio_files`), or
+  - S3 (`audio_s3.prefix`).
+
+### Pipeline imports and profiles
+
+Large pipeline configurations may be split into explicitly named fragments and
+may use fixed-path profile overlays:
+
+```yaml
+configkit:
+  imports:
+    - path: config/storage.yml
+    - path: config/integrations.yaml
+
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+```
+
+The maintained [production/testing bundle](../examples/production-testing/pipeline.yml)
+is a complete copyable example of this structure, including canonical-party
+artifact families.
+
+Imports are resolved relative to the directory containing the root pipeline
+file and are loaded in declaration order. Narratio does not scan directories or
+infer fragments. Each import must be a confined regular `.yml` or `.yaml` file:
+absolute paths, traversal, symlinks, directories, duplicate files, and an
+import of the root pipeline itself are rejected. Imported files cannot declare
+another `configkit` control section.
+
+Base merging is additive. A map may be extended by multiple files when every
+leaf is distinct, but a scalar, list, or map/list/scalar kind cannot be claimed
+more than once, even when the repeated values are identical. Conflict errors
+name the full field path and every source that claimed it. Narratio additionally
+rejects an artifact or artifact-family key declared by more than one base file,
+even when those files contribute disjoint children.
+
+Profile names are case-sensitive and use ConfigKit's safe fixed-path naming
+rules. Only the selected overlay is opened; a malformed unselected overlay does
+not affect a base or different-profile load. Maps merge recursively, overlay
+scalars replace base scalars, and overlay lists replace base lists completely.
+Explicit `false`, zero, empty strings, empty containers, and null values remain
+explicit inputs and are not refilled by defaults. Profiles cannot inherit,
+stack, import files, or declare another ConfigKit control section.
+
+During ConfigKit loading, Narratio applies fresh typed defaults, recorded
+entry-level defaults and normalization, read-only file-local validation, and
+ordinary path resolution before computing a deterministic effective
+configuration digest. The digest represents the normalized, secret-free
+runtime pipeline mapping; it excludes ConfigKit directives, source
+provenance, profile identity, and raw environment secret values. Equivalent
+effective mappings therefore have the same digest regardless of how fields are
+split among the root and imports.
+
+An imported field has the same meaning it would have in a monolithic root
+pipeline. In particular, ordinary relative pipeline paths continue to resolve
+from the root pipeline directory, not from the importing fragment's directory.
+
+## Minimal Working Configuration
+
+`pipeline.yml`
+
+```yaml
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+whisperx:
+  transcribe_url: https://transcription.example.com/transcribe
+```
+
+`campaign.yml`
+
+```yaml
+campaign_id: sample-campaign
+inputs:
+  speakers_file: ./speakers.yml
+  autocorrect_file: ./autocorrect.yml
+  glossary_file: ./glossary.yml
+  party_file: ./party.yml
+```
+
+`session.yml` (local audio)
+
+```yaml
+session_id: 2026-05-03
+inputs:
+  audio_dir: ./audio
+```
+
+## Secrets Handling
+
+- Do not place raw secrets in YAML.
+- Credential fields contain environment-variable names, represented internally
+  as ConfigKit typed secret references (for example
+  `pipeline.audita.llm_api_key: AUDITA_LLM_API_KEY`). A configured reference
+  must resolve to a nonempty environment value during load.
+- Every command that loads a pipeline accepts `--secrets-dir <path>`. When the
+  flag is omitted, Narratio uses a present `secrets/` directory beside the
+  resolved root `pipeline.yml`; an absent conventional directory is ignored.
+  Explicit relative paths are resolved from the caller's working directory,
+  and a missing or empty explicit selection is an error.
+- Each immediate regular file whose name is a valid environment-variable name
+  supplies one value. ConfigKit removes trailing CR/LF, permits a symlink as the
+  selected directory, rejects symlink entries, and limits input to 1 MiB per
+  file, 8 MiB total, and 1024 entries.
+- An existing process environment value takes precedence over a credential file.
+- On POSIX, operators must provision the directory with no group/other access
+  (normally `0700`) and files with no group/other access (normally `0600`);
+  ConfigKit validates type and bounds but does not enforce permission modes.
+- Bootstrap happens once per command, before any configuration load or adapter
+  construction. Safe reports contain filenames, paths, and statuses, never values.
+
+## Publish Configuration Summary
+
+Publish rules live under `pipeline.publish`.
+
+```yaml
+publish:
+  enabled: true
+  upload_run: true
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+```
+
+Rules:
+
+- `outputs[].source` is required.
+- `outputs[].dest` may be omitted when derivable from source.
+- extraction sources require an explicit `outputs[].dest` and publish only when
+  a rule names that source; the Notarius index and complete bundle are not
+  publish sources.
+- `outputs[].required` defaults to `true`.
+- publish locks are independent remote session objects managed with
+  `narratio session locks`; they are not pipeline configuration fields.
+
+## Full Schema
+
+### Pipeline
+
+| Field | Type | Required | Default / Rule |
+| --- | --- | --- | --- |
+| `configkit.imports[].path` | string | No | explicit additive pipeline fragment relative to the root pipeline directory; `.yml` or `.yaml` regular files only |
+| `pipeline.workspace.root` | string | No | `/var/lib/narratio` |
+| `pipeline.workspace.cleanup_after_publish` | bool | No | `false`; selects the published run's workspace target for final cleanup and requires effective `pipeline.publish.enabled=true` and `pipeline.publish.upload_run=true` |
+| `pipeline.campaigns.root` | string | No | `/usr/local/share/narratio/campaigns` |
+| `pipeline.campaigns.default_campaign_id` | string | No | empty |
+| `pipeline.storage.backend` | string | No | `local`; supported values are `local` and `s3` (case-insensitive) |
+| `pipeline.storage.s3.bucket` | string | Conditional | required when backend is `s3` and S3 session-audio or publish upload is enabled |
+| `pipeline.storage.s3.root_prefix` | string | No | `dnd` |
+| `pipeline.storage.s3.region` | string | No | empty |
+| `pipeline.storage.s3.endpoint` | string | No | empty |
+| `pipeline.storage.s3.force_path_style` | bool | No | `false` |
+| `pipeline.storage.s3.access_key_id` | secret reference | Conditional | unset; must be paired with `secret_access_key` |
+| `pipeline.storage.s3.secret_access_key` | secret reference | Conditional | unset; must be paired with `access_key_id` |
+| `pipeline.spool.root` | string | No | `/var/spool/narratio` |
+| `pipeline.spool.delete_audio_after_publish` | bool | No | `false`; selects the published run's spool-audio target for final cleanup and requires effective `pipeline.publish.enabled=true` and `pipeline.publish.upload_run=true` |
+| `pipeline.cache.root` | string | No | `/var/cache/narratio` |
+| `pipeline.cache.s3_audio` | bool | No | `true` |
+| `pipeline.publish.enabled` | bool | No | `true` |
+| `pipeline.publish.upload_run` | bool | No | `true` |
+| `pipeline.publish.outputs[]` | list | No | defaults to final trimmed JSON plus final and final-trimmed Markdown outputs |
+| `pipeline.publish.outputs[].source` | string | Yes (per rule) | must reference built-in or configured artifact source |
+| `pipeline.publish.outputs[].dest` | string | Conditional | derived if omitted and source supports derivation |
+| `pipeline.publish.outputs[].required` | bool | No | `true` |
+| `pipeline.whisperx.transcribe_url` | string | Yes | absolute `http` or `https` URL |
+| `pipeline.whisperx.language` | string | No | `en` |
+| `pipeline.whisperx.timeout` | duration | No | `30m` |
+| `pipeline.whisperx.retries` | int | No | `3` |
+| `pipeline.whisperx.retry_delay` | duration | No | `2s` |
+| `pipeline.whisperx.concurrency` | int | No | `2` |
+| `pipeline.seriatim.binary` | string | No | `seriatim` |
+| `pipeline.seriatim.timeout` | duration | No | `10m` |
+| `pipeline.seriatim.output_schema` | string | No | `seriatim-intermediate` |
+| `pipeline.seriatim.coalesce_gap` | float | No | `3.0` |
+| `pipeline.seriatim.report` | bool | No | `true` |
+| `pipeline.seriatim.env.overlap_word_run_gap` | float | No | unset |
+| `pipeline.seriatim.env.overlap_word_run_reorder_window` | float | No | unset |
+| `pipeline.seriatim.env.backchannel_max_duration` | float | No | unset |
+| `pipeline.seriatim.env.filler_max_duration` | float | No | unset |
+| `pipeline.audita.binary` | string | No | `audita` |
+| `pipeline.audita.timeout` | duration | No | `3h` |
+| `pipeline.audita.llm_api_key` | secret reference | No | unset |
+| `pipeline.audita.modules[]` | list[string] | No | empty |
+| `pipeline.audita.base_url` | string | No | empty |
+| `pipeline.audita.model` | string | No | empty |
+| `pipeline.audita.total_llm_concurrency` | int | No | unset |
+| `pipeline.audita.proposal_llm_concurrency` | int | No | unset |
+| `pipeline.audita.validation_model` | string | No | empty |
+| `pipeline.audita.validation_llm_concurrency` | int | No | unset |
+| `pipeline.audita.transcript_description` | string | No | empty |
+| `pipeline.audita.config_path` | string | No | empty |
+| `pipeline.audita.output_schema` | string | No | empty |
+| `pipeline.audita.work_dir_retention` | string | No | empty |
+| `pipeline.audita.report` | bool | No | `true` |
+| `pipeline.normalize.output_path` | string | No | `transcripts/final.json` |
+| `pipeline.normalize.output_schema` | string | No | `seriatim-intermediate` |
+| `pipeline.normalize.report` | bool | No | `true` |
+| `pipeline.trim.enabled` | bool | No | `true` |
+| `pipeline.trim.output_path` | string | No | `transcripts/final.trimmed.json` |
+| `pipeline.trim.bounds.prompt_id` | string | No | `dnd.session_bounds` |
+| `pipeline.trim.bounds.profile_id` | string | No | empty |
+| `pipeline.trim.bounds.transcript_input_name` | string | No | `transcript` |
+| `pipeline.trim.bounds.output_path` | string | No | `artifacts/session_bounds.json` |
+| `pipeline.trim.bounds.timeout` | duration | No | `10m` |
+| `pipeline.trim.bounds.render_debug` | bool | No | `false` |
+| `pipeline.trim.bounds.render_output_path` | string | Conditional | required when `render_debug` is true |
+| `pipeline.trim.seriatim.report` | bool | No | `false` |
+| `pipeline.notarius.enabled` | bool | No | `false` |
+| `pipeline.notarius.binary` | string | No | `notarius` |
+| `pipeline.notarius.config_path` | string | Conditional | required when enabled; relative paths resolve from the pipeline file directory |
+| `pipeline.notarius.pipeline_id` | string | Conditional | required when enabled |
+| `pipeline.notarius.timeout` | duration | No | `3h`; must be positive |
+| `pipeline.notarius.working_directory` | string | No | directory containing resolved `config_path`; relative paths resolve from the pipeline file directory |
+| `pipeline.notarius.references` | map[string]string | No | empty; maps normalized Notarius selectors to supported prepared Narratio source IDs; maximum 256 entries |
+| `pipeline.notarius.outputs` | map | Conditional | at least one entry when enabled |
+| `pipeline.render.enabled` | bool | No | `true` |
+| `pipeline.render.format` | string | No | `markdown` (only supported value) |
+| `pipeline.render.title` | string | No | empty (falls back to `session.title` when set) |
+| `pipeline.render.include_timestamps` | bool | No | `true` |
+| `pipeline.render.include_segment_ids` | bool | No | `true` |
+| `pipeline.render.include_metadata` | bool | No | `false` |
+| `pipeline.scriptorium.binary` | string | No | `scriptorium` |
+| `pipeline.scriptorium.config_path` | string | No | empty |
+| `pipeline.scriptorium.timeout` | duration | No | `10m` |
+| `pipeline.scriptorium.render_debug` | bool | No | `false` |
+| `pipeline.scriptorium.artifacts` | map | No | empty |
+| `pipeline.scriptorium.artifact_families` | map | No | empty; expands one ordinary artifact per canonical party character |
+| `pipeline.notification.mode` | string | No | `noop`; the only supported notification mode until a provider is implemented |
+
+### Notarius Reference Bindings
+
+`pipeline.notarius.references` maps a Notarius CLI selector to a prepared
+Narratio source, not to a filesystem path:
+
+```yaml
+notarius:
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+```
+
+Supported sources are `narratio.input.party`, `narratio.input.players`,
+`narratio.input.glossary`, and `narratio.input.spell_catalog`. Each map entry is
+required by its presence: omit a binding when the selected Notarius pipeline
+does not need it. A spell-catalog binding additionally requires an effective
+campaign or session `spell_catalog_file`.
+
+Selectors accept Notarius's `slot`, `chunk.slot`, `lane.slot`,
+`lane.extract.slot`, `lane.merge.slot`, and `lane.normalize.slot` forms.
+Narratio trims whitespace around
+selectors and their dot-separated components, rejects empty components and
+`=`, rejects duplicate normalized selectors, and limits the map to 256 entries.
+It validates only selector structure and the prepared source vocabulary;
+Notarius owns target-slot declarations and media compatibility.
+
+Before extraction, Narratio resolves every binding from the current prepared
+session manifest and streams it into a verified invocation-local snapshot whose
+absolute path is passed to Notarius. Missing, unsafe, empty,
+changed-during-copy, or checksum-inconsistent prepared evidence fails with
+guidance to force `prepare`. Bindings are sorted by normalized selector and are
+part of extraction fingerprint and resume identity. See the
+[Notarius integration contract](./integrations/notarius.md) for the subprocess
+boundary and the [complete example](../examples/pipeline.full.annotated.yml)
+for a copyable configuration.
+
+### Notarius Output Entries
+
+For each `pipeline.notarius.outputs.<name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `lane_id` | string | Yes | unique Notarius lane ID |
+| `media_type` | string | Yes | exact accepted descriptor media type |
+| `schema_id` | string | Yes | exact accepted descriptor schema ID |
+| `schema_version` | string | Yes | exact accepted descriptor schema version |
+| `module_key` | string | No | exact accepted module key when set |
+
+Output names must match `^[a-z][a-z0-9_]*$` and become selectable sources named
+`narratio.extraction.<name>`. Lane IDs must be unique. Every declared output is
+required from a successful Notarius result; a missing, rejected, duplicate, or
+contract-incompatible lane fails extraction. See the
+[complete maintained example](../examples/pipeline.full.annotated.yml) for the
+current ten-lane D&D mapping and the [Notarius contract](./integrations/notarius.md)
+for compatibility ownership.
+
+### Scriptorium Artifact Entries
+
+For each `pipeline.scriptorium.artifacts.<name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `enabled` | bool | No | `false` if omitted |
+| `depends_on[]` | list[string] | No | explicit ordering-only edges; must reference configured artifact keys, may not self-reference, and join inferred input edges in one acyclic graph |
+| `render_debug` | bool | No | per-artifact override |
+| `prompt_id` | string | Conditional | required when artifact is enabled |
+| `profile_id` | string | No | empty |
+| `output_path` | string | Conditional | required when enabled; also required when referenced by publish/output/input rules |
+| `timeout` | duration | No | artifact override |
+| `inputs` | map | No | input key names must be non-empty |
+| `vars` | map | No | values must be string or bool; `session_id` is reserved and overwritten by Narratio |
+
+Narratio adds `session_id=narratio-session-<session_id>` to every Scriptorium request for sticky upstream LLM routing. If an artifact config sets `vars.session_id`, Narratio replaces that value before invoking Scriptorium. Use a different variable name if a prompt needs the raw Narratio session ID as content.
+
+Without `--artifacts`, analyze executes enabled configured artifacts. With an
+explicit `--artifacts` list, the exact named configured artifacts are the
+one-invocation targets even if their `enabled` values are false. Analyze closes
+those targets over the effective dependency graph: a current prerequisite is reused, while a
+stale, missing, or failed prerequisite is rebuilt before its dependent.
+Unrelated artifacts are not executed. Named targets and any prerequisite that
+may require rebuilding must therefore have valid executable fields. This
+override affects analyze planning only; publish uses the list only to filter
+configured `narratio.artifact.<name>` output rules.
+
+For each artifact input `pipeline.scriptorium.artifacts.<name>.inputs.<input_name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `source` | string | Yes | built-in runtime source, prepared input source, `narratio.extraction.<name>`, `narratio.artifact.<name>`, or `narratio.previous_session.artifact.<name>` |
+| `required` | bool | No | optional input requirement |
+
+`artifact` and `path` are obsolete and rejected by strict configuration
+loading. Use the canonical `source` identifier to select the input; Narratio
+does not provide adapter-specific input passthrough fields.
+
+Each `narratio.artifact.<name>` input source automatically contributes an
+effective dependency on that configured artifact. Narratio unions these inferred
+edges with explicit `depends_on` entries, then trims, deduplicates, and sorts the
+result during final configuration resolution. Authors use `depends_on` only for
+ordering constraints that are not already expressed by an artifact input.
+Use [`narratio config artifacts`](./cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
+to inspect keys, definition sources, outputs, and effective dependencies without
+loading a session or creating runtime state.
+
+### Scriptorium Artifact Families
+
+`pipeline.scriptorium.artifact_families` declares a shared artifact template
+for every canonical campaign character. Configuration resolution expands each
+family into ordinary `pipeline.scriptorium.artifacts` entries before analyze
+planning or Scriptorium invocation. Party parsing rejects unsupported document
+schemas before family expansion.
+
+For each `pipeline.scriptorium.artifact_families.<name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `enabled`, `prompt_id`, `profile_id`, `timeout`, `render_debug`, `depends_on`, `inputs`, `vars` | ordinary artifact fields | No | copied to each generated artifact under the corresponding ordinary rules |
+| `for_each` | string | Yes | exactly `party.characters` |
+| `output_path_pattern` | string | Yes | safe path beneath `artifacts/` with exactly one `{character_id}` token and no other brace syntax |
+| `member_vars` | map | No | maps an ordinary Scriptorium variable name to a supported canonical character selector |
+| `member_dependencies` | list | No | unique family keys; each generated member depends on the corresponding generated member of each listed family |
+| `publish` | map | No | typed family publish policy (`enabled`, `required`, `dest_pattern`) expanded into concrete publish outputs when enabled |
+
+Generated keys are `<family>_<character_id>` and generated output paths must
+not collide with explicit artifacts or another generated artifact. Families
+expand even when disabled; normal analyze selection still omits disabled
+artifacts unless they are explicitly selected by their concrete key.
+
+Supported `member_vars` selectors are `character_id`, `player.name`,
+`character.name`, `character.class_summary`, and `character.alias_summary`.
+Their resolved values are strings. A member variable may not reuse a static
+`vars` name; `session_id` remains owned and overwritten by Narratio as for any
+other Scriptorium artifact.
+
+Within a family only, an input source may use
+`narratio.member_artifact.<family>`. Resolution rewrites the source to the
+corresponding ordinary `narratio.artifact.<family>_<character_id>` source and
+infers the concrete member dependency. Use `member_dependencies` only for
+additional same-character ordering constraints that are not already inputs.
+This syntax is rejected in explicit artifacts and never reaches runtime stages
+or Scriptorium.
+
+### Notifications
+
+Narratio currently supports only `notification.mode: noop`, which is also the
+default when the section is omitted. The notify stage performs no delivery in
+this mode. Backend, recipient, timeout, and other provider settings are
+rejected by strict configuration loading until Narratio has a provider
+integration.
+
+### Campaign
+
+| Field | Type | Required | Notes |
+| --- | --- | --- | --- |
+| `campaign_id` | string | Yes | canonical opaque campaign identity |
+| `session_template_file` | string | No | used by `session init` when set |
+| `inputs.speakers_file` | string | Yes | stable input default |
+| `inputs.autocorrect_file` | string | Yes | stable input default |
+| `inputs.glossary_file` | string | Yes | stable input default |
+| `inputs.party_file` | string | Yes | stable campaign party source; relative paths resolve from `campaign.yml` |
+| `inputs.spell_catalog_file` | string | No | optional spell-catalog overlay default; required when a Notarius reference selects `narratio.input.spell_catalog` |
+
+### Session
+
+| Field | Type | Required in session file | Notes |
+| --- | --- | --- | --- |
+| `session_id` | string | Yes | opaque identity; must match CLI session target when provided |
+| `previous_session_id` | string | No | opaque identity; must not equal `session_id` |
+| `campaign` | string | No | opaque identity; filled from `campaign_id` during resolve if omitted |
+| `date` | string | No | metadata |
+| `title` | string | No | metadata |
+| `inputs.speakers_file` | string | No | overrides campaign stable input |
+| `inputs.autocorrect_file` | string | No | overrides campaign stable input |
+| `inputs.glossary_file` | string | No | overrides campaign stable input |
+| `inputs.spell_catalog_file` | string | No | overrides the optional campaign spell catalog; empty or omitted inherits the campaign value |
+| `inputs.audio_dir` | string | Conditional | local audio mode |
+| `inputs.audio_files[]` | list[string] | Conditional | local audio mode |
+| `inputs.audio_s3.prefix` | string | Conditional | S3 audio mode |
+
+Audio rules:
+
+- configure local mode (`audio_dir` or `audio_files`) or S3 mode (`audio_s3.prefix`), not both.
+- `audio_s3` requires `pipeline.storage.backend: s3` and a configured S3 bucket.
+
+### Storage backend selection
+
+`local` is the default and disables remote object-store operations. Configure
+`s3` explicitly before supplying `storage.s3`; a populated S3 block does not
+select a backend on its own. Unknown backend names and an S3 block paired with
+`local` are rejected during configuration validation.
+
+### Previous-session expectation
+
+`previous_session_id` is optional in a session file. When a command supplies
+`--previous-session-id`, however, the session file must contain the same value;
+an omitted or different value is rejected before the command performs work.
+
+## Maintained Examples
+
+See the [maintained examples index](../examples/README.md) for complete pipeline,
+campaign, session, template, and input fixtures. Keep complete copyable files
+there rather than duplicating them in this reference.

docs/development.md

@@ -0,0 +1,64 @@
+# Development
+
+This is the first-read landing page for people and LLM coding agents working on
+Narratio. It provides a concise repository orientation and routes each kind of
+change to its canonical documentation.
+
+Narratio is a stage-driven Go orchestrator for turning D&D session audio into
+polished transcripts and generated artifacts. Start with the
+[README](../README.md) for product context,
+[Architecture](policy/architecture.md) for normative system boundaries, and the
+[Internal Overview](internal/overview.md) for implemented component ownership.
+
+## Go Prerequisite
+
+Maintained development and verification require Go 1.26. Use the official Go
+1.26 toolchain before running repository commands.
+
+## What To Read
+
+| When working on | Read | Why |
+| --- | --- | --- |
+| Finding the package or component that owns current behavior | [Internal Overview](internal/overview.md) | It is the implemented component inventory and routes to focused internal documents. |
+| Application shape, boundaries, dependency direction, runtime invariants, safety properties, or dependencies | [Architecture](policy/architecture.md) | It defines the intended system shape, ownership, and non-goals. |
+| Any documentation addition or revision | [Documentation Policy](policy/documentation.md) | It defines canonical owners, audiences, current-behavior rules, and maintenance requirements. |
+| Adding, changing, reviewing, rewriting, or deleting tests | [Testing Policy](policy/testing.md) | It defines risk-based sufficiency, durable test boundaries, test-double guidance, and test lifecycle decisions. |
+| CLI composition or command behavior | [Internal Overview](internal/overview.md) and [CLI Reference](cli.md) | The overview routes to command ownership; the reference owns public syntax and invocation behavior. |
+| Configuration loading, resolution, or user-visible configuration | [Internal Overview](internal/overview.md) and [Configuration](config.md) | The overview routes to implementation ownership; the reference owns fields, defaults, discovery, and validation. |
+| Session workflow, status, restore, cleanup, or object storage | [Restore Internals](internal/command-restore.md), [Workspace Internals](internal/workspace.md), [Storage Internals](internal/storage.md), [Operations](operations.md), and [Troubleshooting](troubleshooting.md) | These separate implementation mechanics, operator procedures, and symptom-driven recovery. |
+| Pipeline sequencing or the behavior of a stage | [Internal Overview](internal/overview.md) and its focused stage documents | The overview owns the implemented stage inventory and routes to each stage contract. |
+| Adapters or external tool contracts | [Adapter Internals](internal/adapters.md) and [Integration Contracts](integrations/README.md) | The internal guide owns adapter composition and mechanics; integration documents own external formats and protocols. |
+| Manifests, artifacts, workspace paths, or publish behavior | [Manifest Internals](internal/manifest.md), [Artifact Internals](internal/artifacts.md), [Workspace Internals](internal/workspace.md), [Publish Internals](internal/stage-publish.md), and [Operations](operations.md) | These separate implementation state and resolution from operator-visible layout and lifecycle. |
+| Maintained configuration or input examples | [Configuration](config.md) and [Examples](../examples/README.md) | The reference owns field meanings; the examples directory owns complete copyable files. |
+| Preparing, validating, or publishing a release | [Release Procedure](release.md) | The maintainer procedure owns version selection, candidate validation, guarded tag publication, and optional later CI inspection. |
+| Proposed or unimplemented behavior | `docs/roadmap/` | Future work belongs only in roadmap documentation until implemented. |
+
+For an existing subsystem, also inspect its focused tests and package-level
+contracts before changing behavior.
+
+## Validation
+
+Use focused package tests while iterating. Every pull request and push runs the
+following repository-wide checks before it can be accepted:
+
+```sh
+go test ./...
+go test -race ./...
+go vet ./...
+go build ./...
+go test ./internal/doccheck
+go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+```
+
+The documentation check verifies local Markdown links and the dependency graph
+of the Woodpecker workflows. The configuration check loads every maintained
+pipeline and session example. Tag CI reuses this validation path before its
+asynchronous asset publication; the maintainer release boundary is documented
+in the [Release Procedure](release.md).
+
+Woodpecker also runs `go test -race -shuffle=on -count=3 ./...` on its scheduled
+job to expose ordering and repeatability defects. Current runners cross-compile
+the supported Linux amd64 and arm64 targets and the best-effort macOS arm64
+target. The macOS cross-build establishes compilation only, not native runtime
+evidence. Add a native check only when an official runner label and successful
+native-run evidence are available.

docs/integrations/README.md

@@ -0,0 +1,36 @@
+# Integrations Index
+
+## Audience
+
+Operators, developers, and coding agents who need to understand Narratio's
+externally observable integration boundaries.
+
+## Scope
+
+`docs/integrations/` is the canonical reference for protocols, invocation and
+data contracts, logical outputs, and compatibility behavior at external tool
+boundaries.
+
+These documents describe what Narratio sends or invokes, what it accepts in
+return, and how failures are surfaced. Internal composition and stage mechanics
+belong in [the adapter implementation guide](../internal/adapters.md) and the
+focused stage documents.
+
+## Integration Contracts
+
+- [Audita](./audita.md): transcript polishing (`audita process`).
+- [Notarius](./notarius.md): complete pipeline execution and safe JSON bundle
+  discovery (`notarius run`).
+- [Party](./party.md): canonical campaign roster input.
+- [Seriatim](./seriatim.md): merge, normalize, trim, and render operations.
+- [Scriptorium](./scriptorium.md): artifact generation and debug rendering
+  (`scriptorium run|render`).
+- [WhisperX](./whisperx.md): speaker-audio transcription over HTTP.
+
+## Related Canonical Docs
+
+- [Configuration](../config.md): operator-facing configuration reference.
+- [Adapter implementation](../internal/adapters.md): shared adapter boundary and
+  runner wiring.
+- [Internal documentation](../internal/overview.md): stage-specific integration
+  usage and component ownership.

docs/integrations/audita.md

@@ -1,147 +1,69 @@
-# Audita
+# Integration: Audita
 
-Audita is a framework-first transcript correction application. The public `audita` package provides:
+## Purpose
+Define the Audita adapter contract used by the `polish` stage.
 
-- deterministic transcript normalization
-- token-batched module orchestration
-- concrete `glossary`, `homophones`, `spoken_word`, and `grammar` modules built on reusable proposal / validator contracts
-- structured run reporting and work-dir diagnostics
+## External Boundary
 
-The previous working implementation has been preserved as `audita_prototype` inside this repository. Its full regression suite lives under `tests/audita_prototype`.
+Narratio invokes `audita process` as a subprocess for each polish operation.
+The configured timeout and parent cancellation bound the invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
 
-## Development
+## Request Contract
+`PolishRequest` carries:
+- required transcript/glossary/output/work-dir paths;
+- optional report path (required when report mode is enabled);
+- generated config and stdout/stderr log paths;
+- optional per-invocation module override.
 
-This project is set up for `uv`.
+The constructed runner owns static Audita settings: binary, timeout,
+credentials, default modules, model and endpoint settings, validation and output
+settings, report mode, and concurrency. The `polish` stage supplies only
+invocation-specific paths and may override modules for that invocation.
 
-```sh
-uv sync --extra dev
-uv run pytest
-```
+## Result Contract
+`PolishResult` returns:
+- processed transcript path;
+- optional report path;
+- work dir and generated-config/log paths;
+- exit code, duration, binary provenance;
+- adapter metadata map.
 
-## Usage
+## Validation and Failure Semantics
+Construction fails for invalid static config values, including:
+- empty binary;
+- non-positive timeout;
+- invalid base URL;
+- invalid output schema;
+- invalid work-dir retention value;
+- invalid concurrency values.
 
-Process a transcript with the current framework implementation:
+Run fails for:
+- missing required request paths;
+- subprocess execution failure;
+- invalid processed transcript JSON (`segments` array required);
+- invalid report JSON when reporting is enabled.
 
-```sh
-uv run audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
+Processed transcript JSON is limited to 64 MiB and optional report JSON to 16
+MiB. Both must be regular files without symlinked path components.
 
-The framework currently runs this default module sequence:
+Failure results still include output/log/config/exit metadata for diagnostics.
 
-1. `glossary`
-2. `homophones`
-3. `glossary`
-4. `spoken_word`
-5. `grammar`
+## Deterministic Behavior
+- CLI args are built from runner config + request in a fixed order.
+- A configured typed `llm_api_key` reference is resolved during pipeline load;
+  the captured value is supplied only as the child's `AUDITA_LLM_API_KEY`
+  override and is never re-read from the parent environment.
+- Generated invocation YAML (`audita.generated.v1`) is emitted when requested.
+- Manifest writes are stage-owned; adapter itself is stateless.
 
-Resolved run instance names are auto-numbered for repeats, so the default report pipeline is:
+## Configuration
 
-1. `glossary_1`
-2. `homophones`
-3. `glossary_2`
-4. `spoken_word`
-5. `grammar`
+Operator-selected values are defined under `pipeline.audita.*` in the
+[configuration reference](../config.md#pipeline).
 
-The default module sequence is fully implemented today:
+Maintained example with Audita config:
 
-- `glossary` proposes glossary-supported acoustic corrections
-- `homophones` proposes conservative homophone and mistranscription corrections
-- `spoken_word` proposes conservative dysfluency cleanup
-- `grammar` proposes punctuation, capitalization, and spacing cleanup only
-
-To run a custom module sequence, pass `--modules`:
-
-```sh
-uv run audita process transcript.json --glossary glossary.yaml --modules grammar --output corrected.json
-```
-
-To also write a structured JSON report:
-
-```sh
-uv run audita process transcript.json --glossary glossary.yaml --output corrected.json --report-json report.json
-```
-
-From a checked-out repository, you can also use the root launcher:
-
-```sh
-./audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
-
-For a system-wide command, install the source tree under `/usr/local/src/audita`, sync dependencies there, and symlink the root launcher into your `PATH`:
-
-```sh
-cd /usr/local/src/audita
-uv sync --extra dev
-ln -s /usr/local/src/audita/audita /usr/local/bin/audita
-audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
-
-Without `--output`, Audita writes the corrected transcript JSON to stdout and progress logs to stderr.
-`--report-json` writes a separate machine-readable run report and never mixes report data into stdout.
-
-Useful configuration can be supplied by CLI flag or environment variable. CLI flags take precedence over environment variables. Normal runs now require LLM API credentials, because the `glossary`, `homophones`, `spoken_word`, and `grammar` modules make real LLM calls. `AUDITA_LLM_API_KEY` and `--llm-api-key` are the preferred provider-neutral credential surfaces, while `OPENROUTER_API_KEY` remains supported as a backward-compatible fallback.
-
-| Environment variable | CLI flag | Default | Purpose |
-| --- | --- | --- | --- |
-| `AUDITA_MODULES` | `--modules` | `glossary,homophones,glossary,spoken_word,grammar` | Comma-separated logical module keys to run; CLI overrides the environment value |
-| `AUDITA_LLM_API_KEY` | `--llm-api-key` | unset | Preferred provider-neutral LLM API credential; CLI overrides both environment-key variants |
-| `AUDITA_VALIDATION_LLM_API_KEY` | `--validation-llm-api-key` | unset | Validation-phase LLM API credential; defaults to the primary LLM API key |
-| `AUDITA_MODEL` | `--model` | `openrouter/google/gemma-4-31b-it` | LLM model name sent to the configured OpenAI-compatible endpoint |
-| `AUDITA_VALIDATION_MODEL` | `--validation-model` | unset | Validation-phase LLM model; defaults to `AUDITA_MODEL` |
-| `AUDITA_BASE_URL` | `--base-url` | `https://openrouter.ai/api/v1` | OpenAI-compatible API base URL |
-| `AUDITA_VALIDATION_BASE_URL` | `--validation-base-url` | unset | Validation-phase OpenAI-compatible API base URL; defaults to `AUDITA_BASE_URL` |
-| `AUDITA_LLM_TIMEOUT_SECONDS` | `--llm-timeout-seconds` | `600` | Per-request timeout in seconds for LLM calls to the configured OpenAI-compatible endpoint |
-| `AUDITA_VALIDATION_LLM_TIMEOUT_SECONDS` | `--validation-llm-timeout-seconds` | unset | Validation-phase per-request timeout in seconds; defaults to `AUDITA_LLM_TIMEOUT_SECONDS` |
-| `AUDITA_VALIDATION_MAX_PROMPT_TOKENS` | `--validation-max-prompt-tokens` | `2048` | Maximum estimated tokens per validation-phase LLM prompt batch |
-| `AUDITA_TARGET_SECTIONS` | `--target-sections` | unset | Exact number of contiguous proposal-stage transcript sections; errors if min/max token bounds cannot be satisfied |
-| `AUDITA_MAX_RETRIES` | `--max-retries` | `3` | Maximum Instructor retries for structured responses |
-| `AUDITA_VALIDATION_MAX_RETRIES` | `--validation-max-retries` | unset | Validation-phase structured-output retries; defaults to `AUDITA_MAX_RETRIES` |
-| `AUDITA_VALIDATION_LLM_CONCURRENCY` | `--validation-llm-concurrency` | unset | Validation-phase LLM concurrency; defaults to `AUDITA_LLM_CONCURRENCY` |
-| `AUDITA_MAX_SECTION_TOKENS` | `--max-section-tokens` | `8192` | Maximum estimated tokens per proposal-stage transcript section |
-| `AUDITA_MIN_SECTION_TOKENS` | `--min-section-tokens` | `2048` | Minimum estimated tokens per proposal-stage transcript section when balancing for concurrency |
-| `AUDITA_GLOSSARY_CONFIDENCE_THRESHOLD` | `--glossary-confidence-threshold` | `0.8` | Minimum confidence required for glossary proposals to survive validation |
-| `AUDITA_GRAMMAR_CONFIDENCE_THRESHOLD` | `--grammar-confidence-threshold` | `0.8` | Minimum confidence required for grammar proposals to survive validation |
-| `AUDITA_HOMOPHONES_CONFIDENCE_THRESHOLD` | `--homophones-confidence-threshold` | `0.8` | Minimum confidence required for homophone proposals to survive validation |
-| `AUDITA_SPOKEN_WORD_CONFIDENCE_THRESHOLD` | `--spoken-word-confidence-threshold` | `0.8` | Minimum confidence required for spoken-word proposals to survive validation |
-| `AUDITA_NORMALIZE_MAX_SEGMENT_GAP` | `--normalize-max-segment-gap` | `4.0` | Same-speaker gaps eligible for deterministic merging |
-| `AUDITA_NORMALIZE_ELLIPSIS_GAP` | `--normalize-ellipsis-gap` | `3.5` | Same-speaker gaps above this value are joined with ` ... ` |
-| `AUDITA_NORMALIZE_MAX_SEGMENT_DURATION` | `--normalize-max-segment-duration` | `60.0` | Maximum merged segment duration |
-| `AUDITA_NORMALIZE_MAX_SEGMENT_TOKENS` | `--normalize-max-segment-tokens` | `2048` | Maximum merged segment prompt payload size |
-| `AUDITA_WORK_DIR` | `--work-dir` | `/tmp/audita` | Per-run scratch diagnostics directory |
-| `AUDITA_WORK_DIR_RETENTION` | `--work-dir-retention` | `auto` | Whether to retain the per-run work directory: `auto`, `always`, or `never` |
-
-Set `AUDITA_MODULES=grammar` to run only the grammar module by default, or override it per command with `--modules`.
-
-Validation-phase LLM settings inherit from the primary `AUDITA_*` LLM settings by default. Set any of the `AUDITA_VALIDATION_*` values only when you want LLM-backed validators to use a different model, endpoint, credential, timeout, retry budget, or concurrency level.
-
-OpenRouter remains the default out of the box:
-
-```sh
-export AUDITA_LLM_API_KEY=your-openrouter-key
-audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
-
-You can point Audita at any OpenAI-compatible endpoint by changing `AUDITA_BASE_URL` and, if needed, `AUDITA_MODEL`. For example, a local vLLM server:
-
-```sh
-export AUDITA_LLM_API_KEY=local-dev-key
-export AUDITA_BASE_URL=http://localhost:8000/v1
-export AUDITA_MODEL=meta-llama/Llama-3.1-8B-Instruct
-audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
-
-Or the actual OpenAI API:
-
-```sh
-export AUDITA_LLM_API_KEY=your-openai-key
-export AUDITA_BASE_URL=https://api.openai.com/v1
-export AUDITA_MODEL=gpt-4.1-mini
-audita process transcript.json --glossary glossary.yaml --output corrected.json
-```
-
-`AUDITA_WORK_DIR` stores per-run diagnostics while processing. Under the default `AUDITA_WORK_DIR_RETENTION=auto`, clean successful runs are removed, while failed runs and successful runs with final skipped corrections are preserved. Use `always` to keep every run directory and `never` to remove successful run directories even when skips remain.
-Failed runs always preserve the run directory and include an authoritative `report.json` alongside normalization and prompt/response diagnostics.
-
-## Prototype Archive
-
-The archived prototype remains importable as `audita_prototype` and is still covered by its original regression suite. This is intentional: the new `audita` package is a framework-oriented rewrite, not a thin wrapper around the old code.
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/notarius.md

@@ -0,0 +1,132 @@
+# Notarius Integration Contract
+
+## Boundary
+
+Narratio uses Notarius as a subprocess to extract configured structured JSON
+lanes from the final trimmed Seriatim transcript. Narratio owns invocation,
+safe bundle discovery, lane selection, and its own artifact metadata. Notarius
+owns pipeline definitions, lane schemas, the receipt, and bundle formats.
+
+Canonical Notarius v0.6.0 references:
+
+- [CLI reference](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/cli.md)
+- [Subprocess consumer contract](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/subprocess.md)
+- [D&D pipeline and lane contracts](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/dnd-pipeline.md)
+- [Run-result receipt](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/run-result.md)
+- [JSON output bundle](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/json-output.md)
+- [D&D spell-catalog overlay](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/dnd-spell-catalog-overlays.md)
+
+The [complete Narratio example](../../examples/pipeline.full.annotated.yml)
+records the exact current constraints for all ten D&D lanes. Treat the linked
+Notarius documents as canonical when changing those values; Narratio does not
+duplicate the complete schemas.
+
+## Invocation
+
+When `pipeline.notarius.enabled` is true, Narratio resolves the executable,
+configuration path, input path, output directory, and working directory to
+absolute paths. Narratio requires the Notarius v0.6.0 CLI contract when
+references are configured and invokes each binding as a separate argument
+before `--json`:
+
+```text
+notarius run <pipeline_id> --config <config_path> --input <trimmed_json> --output-dir <staging_dir> [--reference <selector>=<verified_snapshot_path>]... --json
+```
+
+Reference paths are absolute invocation-local snapshots streamed from the
+manifest-verified canonical files prepared inside the current Narratio session
+workspace. Narratio verifies snapshot checksum and size before and after the
+subprocess, and passes only configured bindings, ordered lexically by normalized
+selector, as direct argument-vector entries without shell interpretation. A CLI
+binding takes precedence over a matching external path in Notarius
+configuration. Narratio never emits `--without-reference`.
+
+For canonical party configuration, the `party` binding is the unchanged,
+validated authored roster and the `players` binding is its generated
+projection. Both retain their established `narratio.input.party` and
+`narratio.input.players` source IDs, and both are resolved from the prepared
+manifest rather than from campaign configuration at extraction time.
+
+The maintained D&D boundary binds only the four campaign-owned external slots:
+
+```text
+notarius run dnd-session \
+  --config <absolute config path> \
+  --input <absolute trimmed transcript path> \
+  --output-dir <absolute staging directory> \
+  --reference glossary=<absolute verified glossary snapshot> \
+  --reference party=<absolute verified party snapshot> \
+  --reference players=<absolute verified players snapshot> \
+  --reference spell_catalog=<absolute verified spell catalog snapshot> \
+  --json
+```
+
+The spell-catalog binding is omitted when the campaign does not maintain that
+optional overlay. Registry, scene-description, combat-turn, and occurrence
+handoffs generated during the same Notarius run remain in Notarius pipeline
+composition and must not be emitted as CLI references. The linked CLI and D&D
+consumer documents own selector targeting, declared slots, media compatibility,
+and generated-handoff collision rules.
+
+Standard output is reserved for the JSON receipt. Standard error is captured
+separately as diagnostic output. Narratio applies the configured timeout and
+does not interpret stdout as a receipt unless the subprocess exits successfully.
+It does not pass a Narratio session ID or run `notarius config validate`
+automatically; the configured working directory and Narratio's minimal child
+environment apply to the subprocess.
+
+## Accepted Result
+
+Narratio's supported invocation baseline is Notarius v0.6.0. The accepted
+receipt remains `notarius.run-result.v2`; reference flags do not change the
+receipt or ten-lane output contract. The receipt
+must identify the configured pipeline, and its `index_file` must be exactly
+`index.json` beneath the reported bundle root. The production index must name
+the management files exactly as `manifest.json`, `rejected.json`,
+`warnings.json`, and `diagnostics.json`. All receipt, index, and lane paths must
+stay inside that bundle; symlinks and non-regular lane payloads are rejected.
+
+Supported receipt and index shapes tolerate unknown fields for forward
+compatibility, while required identity, validation, count, manifest,
+rejection, warning, diagnostic, and lane-list fields remain mandatory.
+Narratio applies bounded reads to the receipt, index, rejection, warning, and
+diagnostic documents. Warning and diagnostic envelopes, group counts,
+occurrence counts, truncation state, framework-owned origins, and
+receipt-to-bundle counts must be internally consistent. Optional chunk-map and
+evidence-context descriptors must carry their complete generic contract
+metadata when present.
+
+For every entry in `pipeline.notarius.outputs`, Narratio requires exactly one
+index descriptor with the configured lane ID, media type, schema ID, schema
+version, and, when configured, module key. Missing, duplicate, rejected, or
+incompatible required lanes fail extraction even if Notarius exited zero. A
+configured lane whose v2 validation summary is `rejected` or `incomplete` also
+fails extraction.
+Unconfigured lanes may remain in the preserved bundle but do not become
+selectable Narratio sources.
+
+Each accepted configured lane is registered as
+`narratio.extraction.<output_key>`. The bundle index is retained for audit and
+resume validation but is not selectable. Scriptorium and publish rules consume
+only explicitly named lane sources; `--artifacts` never selects Notarius lanes.
+
+## Failure And Compatibility Behavior
+
+- Startup and nonzero-exit errors fail extraction and retain captured diagnostics.
+- Invalid receipt JSON or an unsupported receipt schema fails before bundle use.
+- Unsafe or incompatible index data and required-lane rejection fail before the
+  staged bundle is promoted to durable storage.
+- Contract and external provenance metadata are preserved on lane artifact
+  records and through explicit publication.
+- Undeclared selectors, incompatible reference files, and external/generated
+  reference collisions are Notarius errors and fail extraction normally.
+
+Rejection, validation, warning, and diagnostic summaries retain bounded stable
+identity, category, origin, reason-code, status, and occurrence fields without
+copying free-form external messages into Narratio manifest metadata or reading
+lane payload bodies.
+
+Configuration fields and defaults are in [Configuration](../config.md).
+Operator paths, rerun procedures, and bundle retention are in
+[Operations](../operations.md). See [Troubleshooting](../troubleshooting.md)
+for failure recovery.

docs/integrations/party.md

@@ -0,0 +1,71 @@
+# Canonical Party Input
+
+`party.yml` is a campaign-owned roster input. It must be a single, versioned
+`narratio.party.v1` document when Narratio resolves a pipeline, campaign, and
+session together.
+
+```yaml
+schema_version: narratio.party.v1
+
+characters:
+  arannis:
+    player:
+      name: Eric
+    character:
+      name: Arannis
+      alias:
+        - Ari
+        - The Grey Owl
+      classes:
+        - name: wizard
+          level: 8
+```
+
+`characters` is a non-empty mapping. Each key is a stable character ID using
+the configured-artifact key grammar: a lowercase ASCII letter followed by zero
+or more lowercase ASCII letters, digits, or underscores. Character order is
+preserved where roster order matters.
+
+Every entry has `player.name`, `character.name`, and a non-empty
+`character.classes` list. Class entries require a non-empty `name` and may
+include a positive integer `level`. The optional, intentionally singular
+`character.alias` field is a list. Names, aliases, and class names must be
+non-empty, trimmed display strings without control characters. Character names
+and aliases must be unique across the full roster under Unicode-aware
+case-insensitive comparison; player names may repeat.
+
+The document has exactly one YAML document and accepts no unknown fields.
+`schema_version` is required: a missing, wrong, or malformed value is an error.
+
+## Migration
+
+Unversioned party inputs are rejected. Replace each legacy party document with
+the versioned structure above, move each roster entry under `characters`, and
+remove separate player and session-party settings: those keys are unsupported.
+
+The configuration rules and source-relative path behavior are defined in the
+[Configuration Reference](../config.md).
+
+## Derived players document
+
+During `prepare`, Narratio copies the canonical party source bytes unchanged
+to `inputs/party.yml` and writes this deterministic players-only projection to
+`inputs/players.yml`:
+
+```yaml
+schema_version: narratio.players.v1
+players:
+    - name: Eric
+      character:
+        id: arannis
+        name: Arannis
+        alias:
+            - Ari
+            - The Grey Owl
+```
+
+There is one entry per character, sorted by stable character ID. Repeated
+player names remain separate entries. The optional `alias` list retains its
+declared order and is omitted when empty. The projection carries no class
+data. Its prepared manifest record is marked `derived_from_party`; it is not a
+separate user-provided `players_file`.

docs/integrations/scriptorium.md

@@ -1,339 +1,71 @@
-# Narratio -> Scriptorium CLI Integration
+# Integration: Scriptorium
 
-## 1. Purpose
+## Purpose
+Define the Scriptorium adapter contract used by `analyze` and trim-bounds generation in `trim`.
 
-This document defines how Narratio should invoke Scriptorium through the **public CLI**.
+## External Boundary
 
-This is a **subprocess integration contract**, not an internal Go API contract.
-
-## 2. Assumptions
-
-- `scriptorium` is installed and available on `PATH`.
-- Scriptorium is configured with `config.yml`.
-- `config.yml` provides `prompt_dir`, `profile_dir`, and `schema_dir` as needed.
-- Prompt and profile libraries are already deployed for the environment.
-- Narratio provides prepared artifact files (for example polished transcript, glossary, previous recap, campaign notes).
-- Initial integration is synchronous subprocess execution.
-- Narratio remains the orchestrator.
-
-In normal operation, Narratio does not need to pass `--prompt-dir` and `--profile-dir` if they are supplied by Scriptorium config.
-
-Narratio may pass `--config <PATH>` when it must use a non-default Scriptorium config file.
-
-## 3. Core Commands Narratio May Call
-
-Primary commands for subprocess integration:
+Narratio invokes Scriptorium as a subprocess in these modes:
 
 - `scriptorium run`
 - `scriptorium render`
 
-For production generation, use `scriptorium run`.
-
-`scriptorium render` is for debugging, dry-runs, test assertions, and validating command construction without LLM execution.
-
-Note: `scriptorium serve` and HTTP API exist, but they are not the initial integration path.
-
-## 4. Command Selection Guidance
-
-- Use `run` to generate an output artifact.
-- Use `render` to inspect the prepared prompt and effective settings without calling the LLM.
-- Use `render --format json` when Narratio/tests need structured prepare output.
-
-## 5. Recommended `run` Invocation Shape
-
-Production shape:
-
-```bash
-scriptorium run \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --out <output-artifact-path>
-```
-
-Common optional additions:
-
-- `--config <path>`: use a specific Scriptorium config file.
-- `--profile <profile_id>`: override prompt default profile.
-- `--var name=value` (repeatable): small metadata values.
-- `--input name=path` (repeatable): additional named artifacts.
-- `--timeout <duration>`: per-run timeout override.
-- Runtime model override flags (`--llm-base-url`, `--model`, etc.) only for exceptional/operator-directed cases.
-
-## 6. Recommended `render` Invocation Shape
-
-Human-readable debug shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format text
-```
-
-Structured debug/test shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format json \
-  --out <render-debug-path>
-```
-
-`render` does **not** call the LLM, does **not** validate model output, and does **not** perform repair.
-
-## 7. Inputs
-
-- Pass inputs as repeated `--input name=path` flags.
-- `name` must match the Prompt Definition input name.
-- Prefer absolute paths, or paths relative to a working directory controlled by Narratio.
-- Pass Audita output as the primary transcript input.
-- Additional inputs may include glossary, previous recap, campaign notes, event logs, final state maps, or other prompt-specific artifacts.
-- Scriptorium reads input files directly; Narratio does not need to inline file content for CLI use.
-
-## 8. Variables
-
-Use repeated `--var name=value` for small metadata values.
-
-Typical examples:
-
-- `session_date`
-- `session_id`
-- `campaign_name`
-- `previous_session_id`
-- `output_kind`
-
-Large content belongs in input files, not `--var` values.
-
-## 9. Prompt IDs and Output Artifact Types
-
-Narratio should treat prompt IDs as configuration, not hardcoded business logic.
-
-Narratio config may map stage/output names to prompt IDs, for example:
-
-- session recap prompt
-- structured event extraction prompt
-- glossary suggestion prompt
-- player-facing summary prompt
-
-Prompt IDs used by Narratio should come from the deployed Scriptorium prompt library.
-
-## 10. Profiles
-
-- Prompts may declare `default_profile`.
-- Narratio may omit `--profile` to use prompt default profile.
-- Narratio may pass `--profile` to force profile selection.
-- This enables environment/profile selection like `local-fast`, `local-quality`, `frontier`, `batch`, or test profiles.
-- Profile names should generally be Narratio configuration values.
-
-## 11. Runtime Overrides
-
-Supported runtime override flags:
-
-- `--llm-base-url`
-- `--model`
-- `--api-key-env`
-- `--temperature`
-- `--max-tokens`
-- `--top-p`
-- `--timeout`
-
-Guidance:
-
-- Keep normal model/runtime settings in Execution Profiles.
-- Use runtime overrides only for explicit per-run exceptions, tests, or operator overrides.
-- Never pass raw API keys on the command line.
-- `--api-key-env` names an environment variable; Narratio must ensure that variable is set in subprocess environment.
-
-## 12. Config Behavior
-
-- Default config path: `/etc/scriptorium/config.yml`.
-- `--config <PATH>` overrides default path.
-- Missing default config is allowed by Scriptorium.
-- If `--config` is provided explicitly, the file must exist and be valid.
-- CLI flags override `config.yml`.
-- `config.yml` overrides built-in application defaults.
-
-Narratio can either:
-
-- rely on system default config path, or
-- carry an explicit config path and pass `--config`.
-
-## 13. Environment Handling
-
-Subprocess environment recommendations:
-
-- Pass through required API-key environment variables referenced by `api_key_env`.
-- Do not pass raw API keys as CLI arguments.
-- Avoid logging full environment dumps.
-- Capture stdout and stderr separately.
-- Use a controlled working directory.
-- Prefer absolute artifact paths.
-
-## 14. Output Handling
-
-For `scriptorium run`:
-
-- Use `--out` when Narratio needs durable artifact files.
-- Without `--out`, artifact content is written to stdout.
-- Preferred orchestration pattern: always use `--out`, then treat the file as stage output artifact.
-- Capture stderr for diagnostics.
-
-For `scriptorium render`:
-
-- Use `--out` to store render diagnostics.
-- Use `--format json` when tests need to inspect selected profile, effective runtime settings, input hashes, prompt hash, and rendered messages.
-
-## 15. Exit Status and Errors
-
-Current CLI behavior (verified from implementation/tests):
-
-- `0`: success.
-- `1`: runtime/parse/config/load/render/generation/IO error.
-- `2`: run completed but output validation failed (`ValidationFailed`).
-
-Additional details:
-
-- On `run`, output artifact write happens before exit code selection. If validation fails, artifact may still be written and exit code is `2`.
-- `stderr` carries both errors and normal run summary output; non-empty stderr alone does not imply failure.
-- `render` returns `0` on success and `1` on failures.
-
-Narratio should treat non-zero exit codes as failed stage execution, but may record generated artifact paths if a run exited `2` and output file exists.
-
-## 16. Recommended Narratio Integration Pattern
-
-1. Build CLI args from Narratio stage configuration.
-2. Use subprocess context cancellation/timeout.
-3. Pass absolute input paths.
-4. Pass `--out` to a session-scoped artifact path.
-5. Add `--var` metadata values.
-6. Optionally add `--config`.
-7. Optionally add `--profile`.
-8. Ensure required API-key env vars are present.
-9. Run subprocess synchronously.
-10. Capture stdout/stderr separately.
-11. On success, store output artifact path and invocation metadata in stage artifacts.
-12. On failure, store exit code and stderr diagnostics in stage status.
-
-## 17. Suggested Narratio Configuration Shape
-
-Illustrative `pipeline.yml` shape:
-
-```yaml
-scriptorium:
-  binary: scriptorium
-  config_path: /etc/scriptorium/config.yml
-  timeout: 10m
-  render_debug: false
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: dnd.session_recap
-      profile_id: local-quality   # optional
-      output_path: artifacts/session_recap.md
-      timeout: 10m
-      render_debug: false          # optional artifact override
-      inputs:
-        transcript:
-          source: trimmed_transcript
-          required: true
-        previous_recap:
-          source: previous_session_artifact
-          artifact: session_recap
-          path: ""                 # optional
-          required: false
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: session_recap
-```
-
-The key idea: map Narratio artifact names to prompt ID, optional profile, expected inputs, vars, and output destination.
-
-## 18. Testing Strategy for Narratio Integration
-
-- Use `scriptorium render --format json` to verify command construction without LLM calls.
-- Use dedicated test prompt/profile libraries for integration tests.
-- Use small fixture transcripts.
-- Verify missing-input failure behavior.
-- Verify prompt `default_profile` behavior.
-- Verify explicit `--profile` override behavior.
-- Verify `--config` behavior (default and explicit).
-- Verify output file creation when `--out` is used.
-- Verify stderr capture on failures.
-- Avoid real API keys in tests.
-
-## 19. Security and Privacy Notes
-
-- Never pass raw API keys on command line.
-- Do not log full rendered prompts by default; transcripts may contain sensitive content.
-- Avoid logging prompt content unless explicit debug mode is enabled.
-- Treat generated artifacts as potentially sensitive.
-- Use session-scoped, access-controlled output paths.
-- `api_key_env` names should come from environment management, not embedded secrets.
-
-## 20. Initial D&D Artifact Generation Examples
-
-These are examples only. Use prompt IDs from the deployed prompt library.
-
-Session recap:
-
-```bash
-scriptorium run \
-  --prompt dnd.session_recap \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input glossary=/work/session-42/glossary.yml \
-  --out /work/session-42/artifacts/session_recap.md
-```
-
-Structured events:
-
-```bash
-scriptorium run \
-  --prompt dnd.structured_events \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --out /work/session-42/artifacts/structured_events.json
-```
-
-Glossary suggestions:
-
-```bash
-scriptorium run \
-  --prompt dnd.glossary_suggestions \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input previous_recap=/work/session-41/artifacts/session_recap.md \
-  --out /work/session-42/artifacts/glossary_suggestions.md
-```
-
-Player-facing summary:
-
-```bash
-scriptorium run \
-  --prompt dnd.player_summary \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input structured_events=/work/session-42/artifacts/structured_events.json \
-  --out /work/session-42/artifacts/player_summary.md
-```
-
-## 21. Non-Goals
-
-Initial Narratio integration should not:
-
-- call Scriptorium internal Go packages
-- use HTTP API as the primary path
-- expect Scriptorium to read S3 refs directly
-- make Scriptorium responsible for Narratio stage state
-- make Scriptorium responsible for notification
-- require Scriptorium to understand D&D workflow semantics beyond prompt definitions
-
-## 22. Future Extension Notes
-
-Possible later extensions:
-
-- HTTP API integration
-- S3 artifact references if Scriptorium adds S3 reader support
-- richer render diagnostics and policy controls
-- token budgeting/prompt-size checks
-- batch execution if Scriptorium later adds batch support
+The request timeout and parent cancellation bound each invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
+
+## Request Contract
+Both request types carry:
+- binary/config/prompt/profile IDs;
+- input map and vars map;
+- output path;
+- timeout;
+- generated config + stdout/stderr log paths;
+- optional API-key env var name;
+- optional working directory.
+
+## Result Contract
+`ArtifactResult` returns:
+- output/log/generated-config paths;
+- exit code and duration;
+- command mode (`run` or `render`);
+- prompt/profile provenance;
+- `ValidationFailed` marker;
+- metadata map.
+
+## Validation and Failure Semantics
+Request validation fails for:
+- missing binary, prompt id, or output path;
+- non-positive timeout;
+- empty input/var names;
+- empty input path values;
+- missing required credential env var when `APIKeyEnv` is set.
+
+Run behavior:
+- subprocess errors propagate with context;
+- `run` exit code `2` is mapped to `ValidationFailed=true`;
+- successful subprocess still fails if output file is missing or empty.
+
+Each artifact result is limited to 64 MiB and must be a regular file without
+symlinked path components.
+
+Render behavior:
+- subprocess errors propagate;
+- output file must exist and be non-empty.
+
+## Deterministic Behavior
+- input and var maps are sorted into deterministic `--input` and `--var` CLI args.
+- stage wiring adds `session_id=narratio-session-<session_id>` to every Scriptorium request for sticky upstream routing, overriding any configured `vars.session_id`.
+- generated invocation YAML (`scriptorium.generated.v1`) is emitted when requested.
+- adapter is stateless and does not own artifact-selection policy.
+
+## Configuration
+
+Operator-selected values are defined under `pipeline.scriptorium.*`, including
+per-artifact settings under `pipeline.scriptorium.artifacts.*`, in the
+[configuration reference](../config.md#pipeline).
+
+Maintained examples with Scriptorium config:
+
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/seriatim.md

@@ -1,403 +1,62 @@
-# seriatim
+# Integration: Seriatim
 
-`seriatim` merges per-speaker WhisperX-style JSON transcripts into a single JSON transcript that preserves speaker identity and chronological order.
+## Purpose
+Define the Seriatim adapter contract used by `merge`, `normalize`, `trim`, and `render`.
 
-The current implementation supports the `merge` command. It reads one or more input JSON files, optionally maps each input file to a canonical speaker using `speakers.yml`, sorts all segments by timestamp, detects and resolves overlaps when word-level timing is available, assigns consecutive numeric `id` values, and writes a merged JSON artifact.
+## External Boundary
 
-## Usage
+Narratio invokes Seriatim as a subprocess in these modes:
 
-Run from source:
+- `seriatim merge`
+- `seriatim normalize`
+- `seriatim trim`
+- `seriatim render`
 
-```sh
-go run ./cmd/seriatim merge \
-  --input-file samples/raw/2026-04-19-Eric_Rakestraw.json \
-  --input-file samples/raw/2026-04-19-Mike_Brown.json \
-  --output-file merged.json
-```
+The configured timeout and parent cancellation bound each invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
 
-Optional report output:
+## Request/Result Contracts
+- `MergeRequest`/`MergeResult`: multi-input merge to base transcript, optional report.
+- `NormalizeRequest`/`NormalizeResult`: transcript normalization with explicit schema.
+- `TrimRequest`/`TrimResult`: transcript trimming with required keep selector.
+- `RenderRequest`/`RenderResult`: transcript-to-markdown rendering with explicit format and render booleans.
 
-```sh
-go run ./cmd/seriatim merge \
-  --input-file eric.json \
-  --input-file mike.json \
-  --output-file merged.json \
-  --report-file report.json
-```
+Results include output/log/config paths, timing, exit code, and metadata.
 
-## CLI
+## Validation and Failure Semantics
+Runner construction validates:
+- binary presence;
+- timeout > 0;
+- supported output schema (`seriatim-minimal|seriatim-intermediate|seriatim-full`);
+- non-negative coalesce gap.
 
-```text
-seriatim merge [flags]
-```
+Invocation fails on:
+- missing required request paths/inputs;
+- invalid normalize schema override;
+- unsupported render format;
+- subprocess failure;
+- invalid JSON outputs for merge/normalize/trim;
+- missing `segments` array for normalize/trim transcript outputs;
+- empty render output files.
 
-Global flags:
+When report paths are provided/enabled, report files must parse as JSON.
+Each Seriatim JSON or rendered-text result is limited to 64 MiB and must be a
+regular file without symlinked path components.
 
-| Flag | Description |
-| --- | --- |
-| `--help` | Show command help. |
-| `--version` | Show application version. Local builds default to `dev`; release builds inject the release version. |
+## Deterministic Behavior
+- argument ordering is deterministic per command construction.
+- merge env overrides are explicit (`SERIATIM_*`) and only emitted when configured.
+- generated invocation YAML (`seriatim.generated.v1`) is emitted when requested.
+- adapter does not write manifests or choose stage inputs.
 
-`merge` flags:
+## Configuration
 
-| Flag | Required | Default | Description |
-| --- | --- | --- | --- |
-| `--input-file` | Yes | none | Input transcript JSON file. Repeat once per speaker/input file. |
-| `--output-file` | Yes | none | Merged transcript JSON output path. |
-| `--report-file` | No | none | Optional report JSON output path. |
-| `--speakers` | No | none | Speaker map YAML file. When omitted, input file basenames are used as speaker labels. |
-| `--autocorrect` | No | none | Autocorrect rules YAML file. When omitted, the default `autocorrect` module leaves text unchanged. |
-| `--input-reader` | No | `json-files` | Input reader module. |
-| `--output-modules` | No | `json` | Comma-separated output modules. |
-| `--output-schema` | No | `seriatim-intermediate` | JSON output contract. Allowed values are `seriatim-minimal`, `seriatim-intermediate`, and `seriatim-full`. If omitted, the runtime default is used; consumers that depend on a specific shape should set this explicitly. |
-| `--preprocessing-modules` | No | `validate-raw,normalize-speakers,trim-text` | Comma-separated preprocessing modules, evaluated in order. |
-| `--postprocessing-modules` | No | `detect-overlaps,resolve-overlaps,backchannel,filler,resolve-danglers,coalesce,detect-overlaps,autocorrect,assign-ids,validate-output` | Comma-separated postprocessing modules, evaluated in order. |
-| `--coalesce-gap` | No | `3.0` | Maximum same-speaker gap in seconds for `coalesce`; also used as the `resolve-overlaps` context window. Must be a non-negative float. |
+Operator-selected values are defined under `pipeline.seriatim.*` and
+`pipeline.render.*` in the
+[configuration reference](../config.md#pipeline).
 
-Environment variables:
+Maintained examples with Seriatim config:
 
-| Environment Variable | Default | Description |
-| --- | --- | --- |
-| `SERIATIM_OUTPUT_SCHEMA` | `seriatim-intermediate` | Output schema used when `--output-schema` is not explicitly provided. Allowed values are `seriatim-minimal`, `seriatim-intermediate`, and `seriatim-full`. The CLI flag takes precedence. |
-| `SERIATIM_OVERLAP_WORD_RUN_GAP` | `1.0` | Maximum gap in seconds between adjacent timed words when `resolve-overlaps` builds word-run replacement segments. Must be a positive float. |
-| `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW` | `1.0` | Near-start window in seconds for ordering replacement word runs shortest-first. Must be a positive float. |
-| `SERIATIM_BACKCHANNEL_MAX_DURATION` | `2.0` | Maximum duration in seconds for `backchannel` classification. Must be a positive float. |
-| `SERIATIM_FILLER_MAX_DURATION` | `1.25` | Maximum duration in seconds for `filler` classification. Must be a positive float. |
-
-## Input JSON Format
-
-Each input file must be valid JSON with a top-level `segments` array. The current parser accepts the WhisperX segment subset needed for merging:
-
-```json
-{
-  "segments": [
-    {
-      "start": 1.25,
-      "end": 3.5,
-      "text": "Hello there.",
-      "words": [
-        {"word": "Hello", "start": 1.25, "end": 1.55, "score": 0.98},
-        {"word": "there.", "start": 1.7, "end": 2.0}
-      ]
-    }
-  ]
-}
-```
-
-Required segment fields:
-
-- `start`: number, must be `>= 0`.
-- `end`: number, must be `>= start`.
-- `text`: string.
-
-Optional word fields:
-
-- `words`: array of word timing objects.
-- `words[].word`: string.
-- `words[].start`: optional number, must be `>= 0` when present.
-- `words[].end`: optional number, must be `>= start` when present with `start`.
-- `words[].score`: optional number.
-- `words[].speaker`: optional raw speaker label string.
-
-Word-level timing is preserved internally for overlap resolution. If a word is missing `start` or `end`, seriatim keeps the word text, emits a warning in the optional report, and does not use that word as a timing anchor. Word timing is not emitted in the final JSON artifact.
-
-## Speaker Map Format
-
-`speakers.yml` maps input files to canonical speaker names using ordered substring rules:
-
-This file is optional. If `--speakers` is omitted, `seriatim` uses each input file basename as the segment speaker label.
-
-```yaml
-match:
-  - speaker: "Eric Rakestraw"
-    match:
-      - "Eric_Rakestraw"
-      - "Eric"
-
-  - speaker: "Mike Brown"
-    match:
-      - "Mike_Brown"
-      - "mb"
-```
-
-For each `--input-file`, `seriatim` takes the file basename and evaluates the rules in order. The first rule with a matching substring wins, and no later rules are evaluated.
-
-For example, this input:
-
-```text
-samples/raw/2026-04-19-Eric_Rakestraw.json
-```
-
-matches this rule because the basename contains `Eric_Rakestraw`:
-
-```yaml
-- speaker: "Eric Rakestraw"
-  match:
-    - "Eric_Rakestraw"
-```
-
-Important details:
-
-- Matching is against the input file basename, not the full path.
-- Matching is case-insensitive.
-- Rules are evaluated from first to last.
-- Each rule must have a non-empty `speaker`.
-- Each rule must have at least one non-empty `match` string.
-- Duplicate speaker names are invalid.
-- Every input file must match at least one rule or the command fails.
-
-Deprecated old format:
-
-```yaml
-inputs:
-  eric.json:
-    speaker: "Eric Rakestraw"
-```
-
-The old `inputs:` direct mapping format is no longer supported.
-
-## Output JSON Format
-
-`--output-modules json` controls the writer. `--output-schema` controls the JSON contract that writer serializes.
-
-The named schemas are stable public contracts. If a consumer depends on a specific shape, it should request that schema explicitly at runtime. The runtime default selection may change in a future release.
-
-The `seriatim-intermediate` schema is the current default selection when neither `--output-schema` nor `SERIATIM_OUTPUT_SCHEMA` is set. It stays close to the minimal schema, but adds optional `categories` on each segment:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "output_schema": "seriatim-intermediate"
-  },
-  "segments": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 3.5,
-      "speaker": "Eric Rakestraw",
-      "text": "Hello there.",
-      "categories": ["backchannel"]
-    }
-  ]
-}
-```
-
-The `seriatim-full` schema uses the full seriatim envelope:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "input_reader": "json-files",
-    "input_files": ["eric.json", "mike.json"],
-    "preprocessing_modules": ["validate-raw", "normalize-speakers", "trim-text"],
-    "postprocessing_modules": ["detect-overlaps", "resolve-overlaps", "backchannel", "filler", "resolve-danglers", "coalesce", "detect-overlaps", "autocorrect", "assign-ids", "validate-output"],
-    "output_modules": ["json"]
-  },
-  "segments": [
-    {
-      "id": 1,
-      "source": "eric.json",
-      "source_segment_index": 0,
-      "speaker": "Eric Rakestraw",
-      "start": 1.25,
-      "end": 3.5,
-      "text": "Hello there.",
-      "overlap_group_id": 1
-    },
-    {
-      "id": 2,
-      "source": "eric.json",
-      "source_ref": "word-run:1:1:1",
-      "derived_from": ["eric.json#0"],
-      "speaker": "Eric Rakestraw",
-      "start": 2.0,
-      "end": 2.5,
-      "text": "Resolved word run",
-      "categories": ["backchannel"]
-    }
-  ],
-  "overlap_groups": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 4.0,
-      "segments": ["eric.json#0", "mike.json#0"],
-      "speakers": ["Eric Rakestraw", "Mike Brown"],
-      "class": "unknown",
-      "resolution": "unresolved"
-    }
-  ]
-}
-```
-
-The `seriatim-minimal` schema emits minimal metadata and compact ordered segments:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "output_schema": "seriatim-minimal"
-  },
-  "segments": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 3.5,
-      "speaker": "Eric Rakestraw",
-      "text": "Hello there."
-    }
-  ]
-}
-```
-
-Minimal output intentionally omits categories, overlap groups, source/provenance fields, and pipeline configuration metadata.
-
-Intermediate output intentionally omits overlap groups and source/provenance fields, but keeps optional `categories` and minimal metadata.
-
-Segments are sorted deterministically by:
-
-```text
-(start, end, source, source_segment_index/source_ref, speaker)
-```
-
-Final segment IDs are assigned after sorting and start at `1`.
-
-The public Go output contract is available from:
-
-```go
-import "gitea.maximumdirect.net/eric/seriatim/schema"
-```
-
-The same package embeds machine-readable JSON Schemas in `schema/full-output.schema.json`, `schema/intermediate-output.schema.json`, and `schema/minimal-output.schema.json`. The default `validate-output` postprocessor validates the selected output shape and verifies final segment IDs are present, sequential, and start at `1`.
-
-## Overlap Detection
-
-The default postprocessing pipeline detects overlapping segment groups.
-
-Overlap behavior:
-
-- A strict timing overlap is required: `next.start < current_group_end`.
-- Segments that only touch at a boundary are not grouped.
-- Groups require at least two distinct speakers.
-- Transitive overlaps are grouped together.
-- Segments in detected groups receive `overlap_group_id`.
-- `overlap_groups[].segments` contains stable references in `source#source_segment_index` format.
-- `class` is currently `unknown`.
-- `resolution` is `unresolved` until `resolve-overlaps` replaces the group.
-
-## Overlap Resolution
-
-The default postprocessing pipeline runs `detect-overlaps`, then `resolve-overlaps`, then `backchannel`, then `filler`, then `resolve-danglers`, then `coalesce`, then a second `detect-overlaps` pass.
-
-For each detected overlap group, `resolve-overlaps` uses preserved WhisperX word timing to build smaller word-run replacement segments:
-
-- The resolution window expands the detected overlap group by `--coalesce-gap` seconds on both sides.
-- Nearby same-speaker context segments are included when they intersect the expanded window and their start or end is within `--coalesce-gap` of the original overlap boundary.
-- Once a segment is selected for replacement, all timed words from that segment participate in word-run construction; the window controls segment selection, not per-word clipping.
-- Context segments that are part of another detected overlap group are not pulled into the current group.
-- Untimed words are included in replacement text in original word order when nearby timed words create a replacement run.
-- Untimed words do not affect replacement segment start/end times or word-run gap splitting.
-- Words for the same speaker are merged into one run when the gap between adjacent words is no greater than `SERIATIM_OVERLAP_WORD_RUN_GAP`.
-- The default word-run gap is `1.0` seconds.
-- Set `SERIATIM_OVERLAP_WORD_RUN_GAP` to a positive number of seconds to override the default.
-- Near-start replacement word runs are reordered so shorter segments come first when adjacent starts are within `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW`.
-- The default word-run reorder window is `1.0` seconds.
-- Set `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW` to a positive number of seconds to override the default.
-- Replacement segment text is built by joining word text with single spaces.
-- Replacement segments include `source_ref` and `derived_from`.
-- Replacement segments omit `source_segment_index` because they are derived from one or more original segments.
-- Resolved overlap groups are removed before the second detection pass.
-- Replacement segments are left without `overlap_group_id` until the second detection pass annotates any remaining overlap.
-- If a speaker has no usable word timing in a group, that speaker's original segment is kept.
-- If no speakers in a group have usable word timing, the original group and annotations remain unchanged.
-
-## Backchannels
-
-The default pipeline runs `backchannel` before `coalesce`. It tags short acknowledgement segments with:
-
-```json
-"categories": ["backchannel"]
-```
-
-Backchannel matching is case-insensitive, ignores punctuation for matching and word-count purposes, trims surrounding whitespace, and requires a matching acknowledgement phrase, no more than three whitespace-delimited words, and duration no greater than `SERIATIM_BACKCHANNEL_MAX_DURATION` seconds. The default maximum duration is `2.0` seconds.
-
-## Fillers
-
-The default pipeline runs `filler` after `backchannel` and before `coalesce`. It tags short filler utterances with:
-
-```json
-"categories": ["filler"]
-```
-
-Filler matching is case-insensitive, ignores punctuation for matching and word-count purposes, trims surrounding whitespace, and requires only filler tokens such as `um`, `uh`, `er`, `erm`, `ah`, `eh`, `hmm`, `mm`, or repeated combinations of those tokens. Matching segments must contain no more than three whitespace-delimited words and have duration no greater than `SERIATIM_FILLER_MAX_DURATION` seconds. The default maximum duration is `1.25` seconds.
-
-## Dangler Resolution
-
-The default pipeline runs `resolve-danglers` before `coalesce` and before the second overlap detection pass. It repairs short derived fragments when they share provenance with a nearby segment:
-
-- Dangling-end fragments have no more than two words and end in punctuation.
-- Dangling-start fragments have no more than two words.
-- Matching uses same-speaker segments with any shared `derived_from` value.
-- Merged segments use `source_ref` values such as `resolve-danglers:1`, keep the target segment's transcript position, and union `derived_from`.
-
-## Coalescing
-
-The default pipeline runs `coalesce` after `resolve-danglers` and before the second overlap detection pass. It merges adjacent same-speaker segments in the transcript's current order when `next.start - current.end <= --coalesce-gap`.
-
-Coalesced segments use `source_ref` values such as `coalesce:1`, include `derived_from`, and omit `source_segment_index`.
-
-Different-speaker backchannel and filler segments do not block coalescing of surrounding same-speaker segments. Same-speaker backchannel and filler segments are merged normally when they are within `--coalesce-gap`. When same-speaker segments are coalesced, any `backchannel` or `filler` category from the merged inputs is dropped from the coalesced segment.
-
-## Autocorrect
-
-Autocorrect is included in the default postprocessing pipeline. If `--autocorrect` is omitted, the module leaves transcript text unchanged and records a skip event in the optional report.
-
-Enable corrections by passing `--autocorrect`:
-
-```sh
-go run ./cmd/seriatim merge \
-  --input-file input.json \
-  --autocorrect autocorrect.yml \
-  --output-file merged.json
-```
-
-`autocorrect.yml` format:
-
-```yaml
-autocorrect:
-  - target: "Hrank"
-    match:
-      - "hrank"
-      - "Frank"
-
-  - target: "Mike Brown"
-    match:
-      - "Mike Pat"
-```
-
-Matching behavior:
-
-- Matching is case-sensitive.
-- Matches apply only to whole tokens, not substrings inside larger words.
-- Punctuation and whitespace can surround a match.
-- Multi-word and hyphenated matches are supported.
-- Duplicate match strings are invalid, including duplicates across separate rules.
-
-## Current Limitations
-
-- Only JSON input is supported.
-- Overlap resolution depends on WhisperX word timing; groups without usable word timing remain unresolved.
-- Alternate output formats are not implemented yet.
-
-## Release Builds
-
-Local builds record version metadata as `dev`. Release builds should inject the release version with `ldflags`:
-
-```sh
-go build -ldflags "-X gitea.maximumdirect.net/eric/seriatim/internal/buildinfo.Version=v1.0.0" ./cmd/seriatim
-```
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/whisperx.md

@@ -0,0 +1,71 @@
+# Integration: WhisperX
+
+## Purpose
+
+WhisperX transcribes each prepared speaker audio file for Narratio's
+`transcribe` stage. Narratio uses an HTTP boundary and installs each successful
+response as that speaker's raw transcript JSON.
+
+## HTTP Boundary
+
+Narratio sends an HTTP `POST` to the configured transcription URL using
+`multipart/form-data` with:
+
+- `file`: the audio file, retaining its base filename; and
+- `language`: the configured language string.
+
+The server must return a `2xx` response whose body is valid JSON. Narratio does
+not currently require a more specific response schema at this boundary.
+
+The transcription URL must be an absolute `http` or `https` URL. The audio body
+is streamed through a fresh multipart writer for every attempt, so its memory
+use is bounded by the transport buffer rather than by the complete audio file.
+WhisperX response acquisition is capped at 10 MiB.
+
+## Request And Result Contract
+
+Each adapter request identifies a speaker, a readable audio file, and the
+destination for the raw transcript. The HTTP request carries the audio and
+language; the speaker identifier remains Narratio orchestration metadata.
+
+On success, Narratio atomically writes the response body to the requested
+destination. The adapter result reports that logical output together with the
+attempt count, final HTTP status when available, elapsed duration, and adapter
+identity metadata. A failed or invalid response is not installed as the
+transcript output.
+
+## Retry, Timeout, And Cancellation
+
+- The configured timeout applies independently to each HTTP attempt.
+- `retries` means additional attempts after the first.
+- HTTP `429`, HTTP `5xx`, attempt timeouts, and network errors are retryable.
+- Other HTTP `4xx` responses and explicit cancellation are not retryable.
+- Narratio waits the configured retry delay between attempts and aborts that
+  wait when the parent context is canceled.
+
+## Validation And Failure Semantics
+
+Client construction rejects a missing or non-HTTP(S) absolute transcription URL,
+a missing language, a non-positive timeout, negative retries, or a negative
+retry delay. A request fails before transmission when its audio or output path
+is missing.
+
+Non-`2xx` status, transport failure, response-size overflow, invalid JSON, or
+failure to install the output causes the transcription to fail. Errors include
+attempt context, and the result retains attempts, final status when available,
+and elapsed duration for diagnostics.
+
+## Determinism And Concurrency
+
+Each audio request has stable multipart field names, and successful bytes are
+installed atomically. The transcribe stage may process speaker files in
+parallel, bounded by the configured concurrency. It records results in stable
+speaker order after all work completes; any speaker failure fails the stage.
+
+## Related Canonical Docs
+
+- [Configuration](../config.md#pipeline) defines the operator-selected
+  WhisperX URL, language, timeouts, retry policy, and concurrency.
+- [Adapter implementation](../internal/adapters.md) describes internal wiring.
+- [Transcribe stage](../internal/stage-transcribe.md) describes stage mechanics,
+  durable artifacts, and manifests.

docs/internal/adapters.md

@@ -0,0 +1,113 @@
+# Internal: Adapters
+
+## Purpose
+
+Explain the adapter interfaces and production composition used by application
+and stage orchestration. Externally observable protocols and formats belong in
+the [integration contracts](../integrations/).
+
+## Adapter Boundaries
+
+Narratio stage logic depends on adapter interfaces, not transport-specific details.
+
+Primary adapters:
+
+- `whisperx.Client`
+- `seriatim.Runner`
+- `audita.Runner`
+- `scriptorium.Runner`
+- `notarius.Runner`
+- `storage.ObjectStore`
+- `notify.Sender`
+
+## Ownership
+
+Adapters own:
+
+- HTTP/subprocess/SDK argument and transport details.
+- Backend-specific request/response mapping.
+
+Adapters do not own:
+
+- stage ordering/skip/force logic;
+- manifest transitions;
+- canonical path policy.
+
+## Default Wiring
+
+The application composes default adapters per invocation when they were not
+injected through the stage environment, and only when the selected execution
+plan needs them:
+
+- WhisperX HTTP client for `transcribe`.
+- Seriatim subprocess runner for `merge`, `normalize`, `trim`, or `render`.
+- Audita subprocess runner for `polish`.
+- Scriptorium subprocess runner for `trim` or `analyze`.
+- Notarius subprocess runner for `extract` when extraction is enabled.
+- Noop notifier (`notify.NoopSender`) for `notify`.
+- Object store only when required by selected stages/config.
+
+Remote publish locks are loaded only for a selected, enabled publish that
+uploads a run. Shared session lifecycle setup still applies to every selected
+range, but an unselected integration is neither initialized nor validated by
+application-owned composition. Constructor, clock, randomness, remote-lock,
+and post-publish deletion boundaries are scoped to the invocation; stages still
+receive their collaborators through the stage environment. Each selected stage
+retains its own fail-fast configuration and input validation.
+
+`session plan` is outside production adapter composition. It performs
+resume validation and models selected transitions against cloned manifest
+state without constructing or invoking stage-execution adapters. The shared
+command configuration loader may still use object storage to retrieve a missing
+remote session file before planning begins.
+
+Notarius is composed only when extraction is enabled; the extract stage owns
+prepared reference resolution, receipt, bundle, and configured-lane policy.
+The adapter validates the ordered selector/absolute-path pairs and is the sole
+owner of serializing them as repeated `--reference` arguments before `--json`.
+
+Application-owned object-store construction loads configured filesystem
+secrets before invoking the invocation's storage factory. Production commands
+use the concrete storage adapter constructor; focused tests can supply an
+in-memory backend without changing package state.
+
+## Failure Semantics
+
+- Constructor errors fail stage execution setup early.
+- Runtime adapter errors propagate to stage code and then manifest failure handling.
+- Subprocess adapters persist stage logs/generated configs through stage-managed paths.
+- Shared subprocess execution keeps private lifecycle mechanics separate from
+  Narratio's environment, diagnostics, redaction, and error policy. It starts
+  an owned process group on Linux and macOS. Every terminal path disposes of
+  that owned tree before returning. After a natural leader exit, the runner checks
+  for remaining group members and uses bounded graceful then forceful termination.
+  Cancellation, deadlines, and
+  diagnostic limits use the same terminal disposal path without losing their
+  original result classification. Child environments contain only the execution
+  baseline and adapter-specified values; configured credentials are resolved
+  once by ConfigKit and passed explicitly as sensitive values. Adapters do not
+  re-read their named environment references. Stdout and stderr are redacted while streaming into separate
+  8 MiB diagnostic captures; a bounded wait closes a stream retained by a
+  departed leader's descendant. Unsupported platforms reject owned command
+  execution.
+
+## Implementation And Tests
+
+- Composition: `internal/app/runner_composition.go`,
+  `internal/app/runner_dependencies.go`, `internal/app/object_store.go`
+- Shared subprocess mechanics: `internal/adapters/subprocess`
+- Focused adapters: `internal/adapters/{whisperx,seriatim,audita,scriptorium,notarius,storage,notify}`
+- `internal/adapters/whisperx/http_test.go`
+- `internal/adapters/seriatim/subprocess_test.go`
+- `internal/adapters/audita/subprocess_test.go`
+- `internal/adapters/scriptorium/subprocess_test.go`
+- `internal/adapters/notarius/subprocess_test.go`
+- `internal/adapters/storage/*_test.go`
+- `internal/app/runner_test.go`
+
+See the [WhisperX](../integrations/whisperx.md),
+[Seriatim](../integrations/seriatim.md), [Audita](../integrations/audita.md),
+[Scriptorium](../integrations/scriptorium.md), and
+[Notarius](../integrations/notarius.md) contracts before changing an
+externally visible boundary. Operator-selected values belong in
+[Configuration](../config.md).

docs/internal/artifacts.md

@@ -0,0 +1,302 @@
+# Internal: Artifacts
+
+## Purpose
+
+Explain the artifact registry, runtime catalog, resolver, previous-input
+requirements, and shared remote current-state mechanics implemented by
+`internal/artifacts`. Configuration fields that accept source IDs belong in
+[Configuration](../config.md); physical placement belongs in
+[Operations](../operations.md).
+
+## Built-in Source IDs
+
+The internal registry recognizes these stable built-in source IDs:
+
+- `narratio.transcript.base`
+- `narratio.transcript.polished`
+- `narratio.transcript.final`
+- `narratio.transcript.final_trimmed`
+- `narratio.transcript.final_markdown`
+- `narratio.transcript.final_trimmed_markdown`
+- `narratio.bounds.session`
+
+Registry entries bind each ID to its producer, output kind, and content
+validator. The focused stage documents own their input/output flow;
+[Configuration](../config.md) owns where operators may select these IDs.
+
+## Configured, Extraction, And Previous-Session Sources
+
+- configured source ID format: `narratio.artifact.<artifact_key>`
+- extraction source ID format: `narratio.extraction.<output_key>`
+- previous-session source ID format: `narratio.previous_session.artifact.<artifact_key>`
+
+All formats are validated by strict source-policy rules. Configured artifact and
+extraction keys use `^[a-z][a-z0-9_]*$`; source parsers never normalize an
+unrecognized token into a valid source. Extraction sources are registered only
+from `pipeline.notarius.outputs`; the Notarius index has no selectable source
+ID.
+
+Prepared stable source IDs are `narratio.input.players`,
+`narratio.input.party`, `narratio.input.glossary`, and
+`narratio.input.spell_catalog`. Artifact policy owns their canonical manifest
+kind and prepared filename vocabulary. Canonical party mode preserves the
+party source bytes in the party record and supplies the players record from the
+deterministic `derived_from_party` projection; both remain ordinary prepared
+source IDs for consumers.
+
+## Runtime Catalog
+
+`ArtifactCatalog` tracks:
+
+- `planned`: source registered for run context;
+- `executable`: included in the effective analyze artifact set;
+- `available`: the source's canonical evidence owner validates its current
+  manifest record and durable bytes;
+- `provenance`: availability source.
+
+Configured definitions are always registered. Without an explicit selection,
+the effective analyze set contains enabled definitions. With `--artifacts`, the
+exact named configured definitions become the effective set for that invocation,
+regardless of their `enabled` value. The effective-set resolver itself does not
+expand dependencies. Configuration has already materialized the sorted union of
+explicit ordering edges and configured-artifact input edges, and the analyze
+work planner closes targets over that one concrete prerequisite graph.
+Availability is separate from executability. Configuration may normalize a
+family selection into its concrete generated members before this resolver runs.
+The effective set retains optional family
+and character origin metadata, but its keys, catalog sources, and runtime
+lookups remain concrete configured-artifact identities.
+Family publish policies are likewise expanded into ordinary configured-source
+publish rules during configuration resolution.
+Configured outputs, including non-executable prerequisites, become available
+only when the versioned analyze state identifies a current result whose source,
+contract, canonical configured path, size, and checksum match a confined
+no-follow regular file. An incidental canonical file is unavailable, and a
+manifest with an unsupported aggregate-only representation is rejected before
+catalog construction.
+Extraction entries are registered from configuration and become available only
+after compatible extraction evidence is hydrated.
+
+During an analyze invocation, a newly validated and atomically materialized
+configured output is marked available with its producer run ID, contract,
+checksum, and size. Later scheduled dependents therefore observe the same
+semantic identity whether their prerequisite was reused from current manifest
+evidence or produced earlier in the invocation.
+
+Current provenance values:
+
+- `generated.current_analyze_run`
+- `manifest.current_analyze_artifact`
+- `manifest.inputs.previous_cache`
+- `current_session.previous_cache`
+
+## Resolution Rules
+
+Built-ins:
+
+- resolve only from the current manifest's succeeded registered producer;
+- require exactly one output with the registered source ID and output kind, a
+  safe session-rooted path, producer run ID, canonical checksum, and valid
+  content; and
+- verify the registered checksum against the durable bytes. Incidental files,
+  canonical filenames, duplicate outputs, failed/stale producer records, and
+  malformed evidence are unavailable.
+
+Configured sources (`narratio.artifact.*`):
+
+- resolve only through runtime catalog availability;
+- use the shared typed analyze-evidence inspection in
+  `analyze_evidence.go` for prior current-session results;
+- require the supported analyze-state and fingerprint versions, a `current`
+  record for the exact configured key and source ID, a complete contract, the
+  configured canonical relative path, positive stored size, and stored
+  checksum matching bytes read from a confined no-follow regular file; and
+- treat non-current statuses, malformed records, removed keys,
+  unsafe or missing files, and size/checksum mismatches as unavailable without
+  rewriting manifest state. Catalog construction iterates current
+  configuration, so removed or renamed records are not advertised.
+
+`narratio.member_artifact.*` is not a runtime source family. Configuration
+resolution accepts it only in an artifact-family declaration and rewrites it
+to the corresponding configured source before this catalog is built.
+
+Prepared stable sources (`narratio.input.*`):
+
+- resolve only from the current manifest's exact prepared-input record;
+- require the policy-owned canonical path below the session root, a confined
+  non-symlink regular file, a non-empty payload, and a matching SHA-256
+  checksum; and
+- return an immutable source/path/checksum/size identity shared by extract and
+  analyze rather than falling back to campaign/session source paths.
+
+Extraction sources (`narratio.extraction.*`):
+
+- use the shared typed bundle evidence inspection in `extraction_evidence.go`;
+- require a current successful extract record with the exact configured source,
+  compatible contract and Notarius provenance, a confined regular durable
+  payload, matching checksum, and the current resolved trimmed-transcript
+  identity;
+- remain unavailable unless catalog hydration receives valid evidence. Resume
+  treats absent or obsolete evidence as a rerun decision and unsafe evidence as
+  an error; and
+- are never inferred by scanning the Notarius bundle directory.
+
+Previous-session sources (`narratio.previous_session.artifact.*`):
+
+- resolve only from one matching `previous_artifact` input record in the
+  current manifest, including its exact source ID and checksum;
+- are hydrated only from the unique `published_output` declaration for their
+  canonical source in the selected remote commit; and
+- never infer a path from a previous manifest, current configuration, basename,
+  or existing cache file.
+
+Source absence is evaluated by the consuming artifact input. An optional input
+is omitted from that invocation; a required input fails resolution. This is
+separate from a stage's lifecycle outcome.
+
+Validation by content type:
+
+- transcript JSON built-ins: JSON with top-level `segments` array;
+- transcript Markdown built-ins: non-empty text file;
+- bounds built-in: valid JSON;
+- configured/previous-session artifact files: non-empty text file.
+
+## Previous Requirement Collection
+
+`CollectPreviousArtifactRequirements`:
+
+- scans the effective configured artifact set;
+- extracts only canonical previous-session sources;
+- deduplicates by artifact key;
+- merges required and optional references (required wins);
+- returns deterministic ordering and source locations.
+
+## Current-State Helpers
+
+Artifacts package owns shared remote current-state loading mechanics used by
+restore, status and validation checks, and previous-cache planning.
+
+The selected session manifest is accepted only with
+`schema_version: narratio.session-manifest.v1` and complete stage identity.
+An unsupported local or downloaded manifest must be replaced by regeneration,
+verified restore, or republication; artifact resolution never fills in missing
+authority markers.
+
+For a new-protocol current state, the pointer-selected immutable commit is the
+complete restore authority. Callers receive its declared object identities and
+must not supplement them by listing mutable session prefixes.
+
+A `published_output` declaration must use a canonical publish-compatible source
+ID and an exact `runs/{run_id}/published/{destination}` key in the same session.
+Only this declaration type may refer to an earlier run; other declared objects
+remain confined to the selected commit's run.
+
+The reader opens each small control object directly and enforces owner-specific
+limits before decoding: 64 KiB for the mutable commit pointer, 4 MiB for the
+immutable commit manifest, and 8 MiB for the selected session manifest. These
+are exposed as `MaxCurrentCommitPointerBytes`,
+`MaxRemoteCommitManifestBytes`, and `MaxRemoteSessionManifestBytes`.
+
+Each read uses the generation and size metadata returned with its opened body.
+Actual bytes remain subject to a limit-plus-one read even if size metadata is
+absent or inaccurate. Immutable selections then retain their declared-size,
+checksum, generation, and identity checks. No current-state control object is
+downloaded through a temporary file. Artifact and previous-session downloads
+validate their complete sibling temporary file against available committed
+identity before replacing a canonical local path.
+
+Core helpers:
+
+- `LoadCurrentState`
+- `ValidateCurrentStateIdentity`
+- `RemoteCommitManifest` and `CurrentCommitPointer`
+
+Typed missing-state error:
+
+- `CurrentCommitPointerMissingError` (`ErrCurrentCommitPointerMissing`)
+
+Identity validation supports caller-provided expectations:
+
+- expected campaign;
+- expected session ID;
+- expected run ID, or pointer/manifest run-ID consistency check.
+
+Caller policy is intentionally outside artifacts helpers:
+
+- some callers fail on missing current state;
+- some callers downgrade missing state to status/findings;
+- some callers skip optional behavior when state is missing.
+
+## Remote Lock Objects
+
+The artifacts package defines the independent remote-lock payload and identity
+contract. Each canonical publish source maps to one object below the validated
+session `locks/` prefix; its filename is the lowercase SHA-256 digest of the
+exact source followed by `.lock`.
+
+Lock payloads are strict versioned JSON containing the canonical source, an
+optional reason, and a UTC creation time. Reads are bounded to
+`MaxRemoteLockObjectBytes` (64 KiB), reject unknown fields and trailing values,
+and verify that the payload source matches the key digest. Exact lookup reports
+presence explicitly, prefix listing validates every object and sorts by source,
+creation uses an absence precondition, forced replacement uses one observed
+generation, and removal is idempotent.
+
+## Key Path Helpers
+
+`internal/artifacts/paths.go` and S3-key helpers define canonical helpers for:
+
+- session/work/run paths;
+- previous-cache paths;
+- spool/cache paths;
+- S3 session/run/current-state key layout.
+
+New publication creates run-scoped immutable objects, including
+`runs/{run_id}/commit.json` and `runs/{run_id}/session-manifest.json`. The sole
+mutable selector is `current/commit-pointer.json`; readers verify its selected
+commit and declared object generations/checksums.
+
+A `published_output` declaration may identify its commit's run or an earlier
+immutable run in the same campaign/session namespace. The key parser validates
+that namespace, rejects mutable and malformed keys, and derives the originating
+run from the exact declaration. All other commit artifacts remain confined to
+the commit's own run prefix.
+
+Campaign, session, and Narratio run IDs are validated as portable opaque
+segments at configuration and artifact boundaries before they can be used in a
+workspace or S3 namespace. Previous-artifact destinations remain typed,
+multi-segment relative paths and are confined beneath `previous/artifacts`; they
+are not treated as opaque identifiers.
+
+See [Workspace Internals](workspace.md) for how callers consume local helpers
+and [Operations](../operations.md#local-state-layout) for the authoritative
+physical layout.
+
+## Invariants
+
+- source ID formats are stable contracts;
+- artifact resolution is deterministic and manifest-aware;
+- extraction sources are available only from a compatible successful manifest
+  record;
+- previous-session source resolution in `analyze` is local-only;
+- remote current-state key construction remains centralized in artifacts helpers.
+
+## Implementation And Tests
+
+- Registry and resolution: `internal/artifacts/artifact_resolver.go`,
+  `internal/artifacts/catalog.go`, `internal/artifacts/transcripts.go`,
+  `internal/artifacts/extraction_catalog.go`,
+  `internal/artifacts/extraction_evidence.go`,
+  `internal/artifacts/extraction_input.go`,
+  `internal/artifacts/prepared_input.go`
+- Current state: `internal/artifacts/current_state.go`,
+  `internal/artifacts/current_state_commit.go`,
+- Paths and keys: `internal/artifacts/paths.go`,
+  `internal/artifacts/s3_keys.go`
+- Previous requirements: `internal/artifacts/previous_requirements.go`
+- Tests: `internal/artifacts/artifact_resolver_test.go`,
+  `internal/artifacts/catalog_test.go`,
+  `internal/artifacts/extraction_catalog_test.go`,
+  `internal/artifacts/current_state_test.go`,
+  `internal/artifacts/paths_model_test.go`,
+  `internal/artifacts/previous_requirements_test.go`

docs/internal/command-restore.md

@@ -0,0 +1,127 @@
+# Internal: Command Restore
+
+## Purpose
+
+Explain the implemented restore discovery, planning, installation, and
+reporting flow in `internal/app`. User invocation belongs in
+[CLI](../cli.md#session-restore), and the operator recovery procedure and
+physical restore scope belong in
+[Operations](../operations.md#restore-workflow).
+
+Restore separates remote authority, local conflict policy, and filesystem
+mutation so each remains testable independently.
+
+The public command supplies production defaults to one private,
+invocation-scoped dependency value. Discovery, planning, and execution use that
+same value, so focused sequencing tests can replace one boundary without
+mutable package hooks.
+
+## Discovery Contract
+
+Discovery delegates current-state pointer and manifest loading to
+`internal/artifacts`, then validates the result against the resolved request:
+
+- campaign must match;
+- session ID must match.
+- run ID must match the pointer-selected committed run.
+
+Restore treats any missing or invalid remote current state as a command error.
+
+## Planning Contract
+
+Restore planner action kinds:
+
+- `download`;
+- `skip_same`;
+- `conflict`.
+
+Planner behavior:
+
+- a new-protocol restore uses only the selected commit's declared artifact set;
+  each action carries that artifact's immutable key, checksum, size, and
+  generation;
+- remote-to-local mapping is traversal-safe;
+- actions are sorted by local relative path and then remote key;
+- planning opens the selected session root and inspects every target by its
+  normalized session-relative path. Missing ancestors produce a missing target;
+  an ancestor or leaf symlink is a non-regular conflict and is never opened for
+  hashing;
+- force converts differing eligible regular files from conflicts to downloads;
+  directories, symlinks, and other non-regular targets remain conflicts.
+
+For a non-dry-run restore, planning/classification happens only after acquiring
+the session lock. Runner manifest/reuse checks acquire that same lock first.
+
+Previous-cache readiness is resolved through `previouscache.Resolve` for restore,
+prepare, status, and validation. Each requirement selects the unique
+`published_output` for its canonical source in the pointer-selected commit,
+then strips only that run's immutable `published/` prefix for its safe local
+cache path. It never probes candidate objects or derives paths from manifests,
+configuration, or basenames.
+
+## Execution Contract
+
+Execution order and safety:
+
+- non-manifest downloads happen before manifest install;
+- `manifest.json` installs last;
+- downloads use sibling temp files, validate the complete object, then use an
+  atomic rename;
+- manifest replacement is validated before rename;
+- each committed object is verified against its declared checksum, size, and
+  generation before installation;
+- installation reopens the selected session root, creates only validated
+  missing directory components with workspace modes, and rejects symlinked or
+  non-regular ancestors and leaves before replacement;
+- a committed manifest already verified during discovery is retained for the
+  matching restore action and revalidated before installation, avoiding a
+  second body transfer;
+- failed installs do not roll back files already written in the same execution.
+- a durable `.restore-incomplete.json` marker is written before installation.
+  It blocks runners until a restore retry completes all verified installs and
+  the local manifest replacement, at which point it is removed.
+- restored manifest local references are rebased beneath the selected local
+  session root. Unsafe relative references and producer-machine absolute paths
+  outside the manifest's producer session root are rejected; producer-local
+  spool paths and run-local diagnostic references are not restored as
+  authority. Cleanup remains ordinary restored stage state and is not reset or
+  synthesized by restore.
+
+Audio restore path:
+
+- uses `audio.MaterializeS3Audio`;
+- integrates spool and S3 audio cache paths;
+- reuses cached audio only when its opened regular file, content digest, and
+  identity sidecar all match the selected remote object version; otherwise it
+  refreshes through the validated download path.
+
+## Reporting Contract
+
+- dry-run mode prints a summary, performs no durable session writes, and may
+  read remote current-state or object-identity data to produce that summary;
+- execution mode persists the canonical restore report described in
+  [Operations](../operations.md#restore-workflow);
+- report includes plan counts, per-action status, and execution failures.
+
+## Invariants
+
+- restore uses committed remote current state as authority;
+- one restore or status inspection observes the single pointer-selected commit
+  loaded at discovery; later pointer changes cannot add objects or substitute a
+  different run into its plan;
+- a verified `current/commit-pointer.json` and its selected immutable commit
+  establish remote commitment; when absent, republish the session;
+- restore does not execute pipeline stages.
+
+## Implementation And Tests
+
+- Discovery: `internal/app/restore_discovery.go`
+- Planning: `internal/app/restore_plan.go`, `internal/previouscache`
+- Execution: `internal/app/restore_execute.go`
+- Reporting and command coordination: `internal/app/restore_report.go`,
+  `internal/app/restore.go`
+- Tests: `internal/app/restore_discovery_test.go`,
+  `internal/app/restore_plan_test.go`,
+  `internal/app/restore_execution_test.go`,
+  `internal/app/restore_recovery_test.go`,
+  `internal/app/restore_workflow_test.go`

docs/internal/configuration.md

@@ -0,0 +1,211 @@
+# Configuration Internals
+
+User-visible fields, defaults, and selection behavior belong in the
+[Configuration Reference](../config.md). This document describes the internal
+pipeline-loading boundary implemented by `internal/config`.
+
+## Pipeline Loading
+
+`LoadPipeline` assembles and validates a pipeline in this order:
+
+1. Give ConfigKit a fresh typed `PipelineConfig` default value, the root file,
+   enabled imports, and the caller's profile selection. Omitted, empty, and
+   exact `default` selections use the base; named profiles use the fixed
+   `profiles/<name>.yml` convention.
+2. ConfigKit parses the root and declared imports, additively merges base
+   sources, applies only the selected overlay, and strictly decodes the result.
+3. The definition resolver applies recorded entry-level and inherited defaults,
+   runs deterministic domain normalization, records derived publish destinations,
+   and resolves ordinary paths relative to the root pipeline file.
+4. The definition validator performs read-only checks that require only the
+   pipeline file. Checks requiring canonical-party family expansion remain at
+   the existing explicit pipeline/campaign resolution boundary.
+5. Narratio inspects public ConfigKit metadata to reject artifact and family keys
+   declared by multiple base files, adapts source ownership records, attaches
+   runtime-only state, and computes the normalized effective digest.
+
+This ordering preserves monolithic configuration behavior. Moving a field to
+an imported fragment changes its source ownership, not its path base, default,
+or schema semantics.
+
+## Loaded Context Resolution
+
+`LoadedPipelineCampaign` carries one already composed pipeline and its selected
+campaign into session resolution. `LoadSessionWithPipelineCampaignOptions`
+loads a local session against that context, while
+`ResolveLoadedPipelineCampaign` also accepts an already loaded remote session
+or no session while a caller retrieves one. Compatibility loaders route through
+these functions after their initial pipeline and campaign reads.
+
+Application commands own pipeline and campaign discovery, campaign-file versus
+registry selection, and the corresponding mutual-exclusion rules. Once they
+have a `LoadedPipelineCampaign`, local session discovery and remote-session
+download retain that exact pipeline object and its private provenance. Removing
+a temporary downloaded session file therefore cannot invalidate the resolved
+pipeline or campaign context.
+
+Campaign and session files use independent ConfigKit definitions with imports
+and profiles disabled. Campaign files normally load directly from their source;
+session files are first read through the 8 MiB bounded-read boundary because the
+application must reject unresolved template placeholders before passing an exact
+byte snapshot to ConfigKit. Local and remotely downloaded session bytes then use
+the same typed loader and retain the existing post-load identity expectations,
+stable-input precedence, and source-relative path semantics.
+
+The application also has a separate read-only inspection resolver for `config
+validate`, `config show`, `config sources`, and `config artifacts`. It uses the same production root/profile and
+campaign selection functions, but never routes through session discovery,
+remote-session download, adapter composition, workspace
+initialization, manifest access, or cleanup. A pipeline with retained artifact
+family declarations must resolve its selected campaign before ordinary pipeline
+validation, which expands its canonical-party members and generated publish
+rules. A pipeline without those declarations may be validated by itself.
+Secret-directory bootstrap remains invocation-scoped and happens before the
+ordinary ConfigKit load because explicit typed references must resolve even for
+inspection.
+
+`MarshalEffectivePipeline` is the configuration-owned projection for `config
+show`. It serializes one explicit typed safe projection with the existing YAML
+encoder. The projection has no slots for resolution-only artifact family
+declarations or private provenance. The result has one trailing newline;
+commands do not marshal runtime objects directly.
+
+`EffectivePipelineSources` and `EffectiveCampaignSources` provide the separate
+safe provenance projection for `config sources`. Pipeline ownership begins with
+the complete logical field paths projected from the schema and classifies
+each contributor as root, import, profile, or centralized default. The
+projection replaces generated concrete member paths with paired family and
+canonical-party records, and does the same for generated publish rules.
+Campaign records identify campaign-owned fields and party inputs; canonical
+derived players point to the party source. The application command only joins these sorted
+records with selection metadata and never reparses configuration files.
+
+`EffectiveArtifactCatalog` projects the same finalized concrete artifact map for
+`config artifacts`. It derives source identity from each map key, carries sorted
+effective dependencies, aggregates ConfigKit definition contributors, and adds
+family/character/party provenance for generated members. It does not inspect
+files named after artifacts or infer another runtime graph.
+
+`config diff` uses a thin paired loader that performs two independent ConfigKit
+loads. When campaign
+resolution is needed, the command loads one selected campaign and party and
+expands both effective pipelines from that same party value. The configuration
+owner projects each normalized effective mapping directly from that same typed
+projection into sorted logical paths using JSON number preservation;
+mapping leaves are compared individually while sequence values remain atomic.
+Values are compact deterministic JSON representations for command output, not
+raw YAML fragments, ownership records, or secret material. A differing digest
+with no projected difference is treated as an internal consistency error.
+
+Campaign context construction reads the campaign-owned party source through
+`ParseParty`. The required canonical party retains its raw bytes and normalized
+roster in runtime-only `ResolvedParty` provenance. Canonical resolution creates
+a deterministic `derived_from_party` players projection. The party is the only
+configured roster authority; player and session-party fields are unsupported.
+
+## Canonical Party Domain
+
+`ParseParty` is the package-owned boundary for parsing a party source. Its
+single-source ConfigKit definition disables imports and profiles, requires
+`schema_version` through `VersionRule`, and strictly decodes the
+character-keyed typed schema before applying Narratio's party-domain checks.
+Successful metadata coordinates recover authored character-map order without a
+second YAML parse; class and alias slice order comes directly from typed decode.
+The canonical value retains a separate exact byte copy of its source so
+consumers can materialize the authored party document without reserializing it.
+Its `PlayersYAML` method deterministically derives the ID-sorted, versioned
+players-only projection.
+
+An unversioned source is rejected before resolution, family expansion, or
+prepare-stage materialization.
+
+## Validation Ownership
+
+Configuration validation remains within `internal/config`, with the public
+entry points and pipeline-domain sequencing separated from the private owners
+for core pipeline/session rules, external adapters, publish/storage, Notarius,
+Scriptorium, and shared scalar/path checks. These are package-local ownership
+boundaries rather than extension points or a general validation framework.
+
+Validation is read-only, fails fast, and does not define which of several
+independent invalid fields is reported first. Map-backed declarations are
+nevertheless traversed deterministically. Reusable domain normalization is an
+explicit preceding operation for both loaded and programmatically constructed
+pipelines. Notarius normalization commits canonical scalar, reference, and output
+values only after the complete normalization succeeds; its validator then checks
+that candidate without changing it. Scriptorium validates artifact declarations
+and outputs, dependencies and inputs, variables, referenced outputs, and the final
+dependency graph in deterministic artifact order.
+
+Diagnostics identify the relevant semantic field. Merge conflicts also
+retain every claiming source needed to correct an additive ownership conflict.
+Complete prose, punctuation, formatting, and private validation choreography are
+not compatibility contracts.
+
+## Diagnostics And Runtime Metadata
+
+Syntax, duplicate-key, merge, profile, and schema failures include the
+relevant source name and full field path. Additive conflicts report every
+claiming source so operators can repair the split without repeatedly
+rediscovering additional conflicts.
+
+The loaded pipeline retains private runtime metadata for the absolute root
+path, ordered imports, selected profile name and selection source, selected
+overlay, original ConfigKit metadata and source snapshots, effective digest,
+and leaf ownership. Base
+leaves retain their root/import owners, replaced leaves belong to the selected
+overlay, and centrally supplied values use the synthetic `default` owner. This
+metadata does not participate in YAML decoding or alter the public
+configuration model.
+
+The effective digest is SHA-256 over a versioned deterministic JSON encoding of
+the same typed safe projection used by show and diff. Runtime Notarius paths remain absolute for
+execution, but the digest substitutes their normalized logical values captured
+before root-relative resolution, so relocating an equivalent configuration
+bundle does not change provenance. Because ConfigKit and resolution metadata
+are private, the digest excludes source layout, profile name, and ownership.
+Configuration stores credentials as ConfigKit typed references. A successful
+load resolves their values inside the opaque typed boundary, while this safe
+projection serializes only `EnvName()` identities. Credential-value changes are
+therefore never hashed; reference-name changes remain ordinary provenance.
+`recomputePipelineEffectiveDigest` is the single package-owned refresh point
+for later runtime expansion.
+
+## Test Surfaces
+
+`pipeline_configkit_test.go` exercises real imports and fixed-path profiles,
+base selection, unselected-overlay isolation, source attribution, and Narratio's
+unique artifact/family base-declaration rule. Existing focused configuration
+tests protect defaults, explicit-value semantics, strict decoding, validation,
+metadata, and root-relative path behavior. Application configuration-loader tests protect the
+single-read boundary by changing the pipeline file after its initial load and
+confirming local session resolution retains the original pipeline. Other
+configuration tests continue to protect defaults and validation after assembly.
+`party_test.go` protects the versioned typed party schema, domain invariants,
+authored character order, retained bytes, and deterministic players projection
+without involving campaign or runtime wiring.
+`party_resolution_test.go` protects campaign-owned party loading, canonical
+input restrictions, source provenance, and virtual players input selection.
+
+## Artifact Family Resolution
+
+Pipeline loading retains `scriptorium.artifact_families` as a resolution-only
+declaration. Once campaign party resolution establishes a canonical roster,
+configuration expands families in sorted family-key and character-ID order
+into ordinary `ScriptoriumArtifactConfig` values. The expansion owns the narrow
+`{character_id}` output substitution, closed member-variable selectors, key and
+output collision checks, and the runtime-only family-origin catalog. It then
+removes family declarations from `ScriptoriumConfig`, runs ordinary Scriptorium
+validation, and refreshes the effective pipeline digest. Stages and adapters
+therefore receive only concrete artifact maps.
+
+The catalog retains sorted family member keys plus family/character/source
+origins and the typed dependency/publish declarations for their later owners.
+`member_dependencies` add corresponding ordinary concrete dependencies, while
+the family-only `narratio.member_artifact.<family>` input form is rewritten to
+the matching ordinary configured-artifact source. The catalog records those
+resolved dependency and input identities with their declaring family and party
+member. No member-artifact source is registered as a runtime policy source.
+An enabled family publish declaration expands to ordinary configured-artifact
+publish rules before the existing publish and lock validators run. Runtime
+publication consequently receives no family wildcard or special matcher.

docs/internal/fileops.md

@@ -0,0 +1,55 @@
+# Internal: File Operations
+
+`internal/fileops` owns the narrow mechanics for atomically visible replacement of one
+byte file. Callers keep ownership of serialization, validation, cancellation,
+and destination-directory policy.
+
+## Destination Paths
+
+Ordinary workspace replacement uses standard operating-system path semantics.
+Callers with a root-scoped safety contract own that validation at their boundary.
+
+Remote object acquisition is owned by its domain caller. The caller writes a
+sibling temporary file, validates the complete result, and passes that file to
+`InstallSiblingTempFile` only when it is ready to become canonical. Callers own
+remote object selection, validation, conflict handling, and temporary-file
+cleanup.
+
+Directory bundles are not promoted by `internal/fileops`; extraction owns its
+validated Notarius bundle rename.
+
+## Cleanup Contract
+
+`RemoveAllUnderRoot` accepts an explicit root and a proper descendant, opens
+that root with `os.OpenRoot`, and removes the normalized root-relative target
+with `Root.RemoveAll`. It rejects root, traversal, and out-of-root targets; a
+missing target is successful. `RemoveFileUnderRoot` additionally requires an
+existing leaf to be regular. Command and post-publish policy remains owned by
+`internal/app`.
+
+## Bounded Reads
+
+`ReadFileBounded` opens a normal path and reads at most its caller-supplied
+limit. Credential mode policy and environment precedence remain owned by
+`internal/app`.
+
+## Replacement Contract
+
+`ReplaceFileAtomic` requires an existing destination directory. It creates a
+sibling temporary file, writes the complete byte sequence, applies the
+caller-supplied mode, closes the file, runs an optional pre-rename check, and
+replaces the destination with a rename.
+
+The pre-rename check is the last point at which a caller can cancel without
+installing a new destination. A failure before the rename leaves the old
+destination unchanged and removes the temporary file. Successful replacement
+provides atomic visibility; it makes no crash- or power-loss durability claim.
+
+Replacement follows the operating system's same-filesystem rename semantics.
+If a platform cannot replace an existing destination, the operation returns an
+error and never removes the old file as an emulation step.
+
+`WriteFileAtomic` and the copy helpers create the destination parent with the
+repository's workspace permissions. `InstallSiblingTempFile` requires the
+caller to provide an existing temporary file in the existing destination
+directory.

docs/internal/manifest.md

@@ -0,0 +1,315 @@
+# Internal: Manifest
+
+## Purpose
+
+Explain the session-progress model implemented by `internal/manifest`. Physical
+manifest placement belongs in
+[Operations](../operations.md#local-state-layout).
+
+## Session Manifest
+
+`manifest.Manifest` records:
+
+- required `schema_version: narratio.session-manifest.v1`
+- identity (`session_id`, `campaign`, `run_id`)
+- local path metadata (`local_workdir`, `local_spool_dir`)
+- remote identity metadata (`s3_bucket`, `s3_session_prefix`, `s3_run_prefix`)
+- `inputs` records
+- durable `artifacts` records
+- per-stage `stages` map, including the ordinary cleanup stage record
+
+Session, campaign, and run identities in local and downloaded manifests must be
+portable opaque segments. Unsafe legacy identities are rejected with migration
+guidance rather than being normalized into a different workspace or remote
+namespace.
+
+Decoding is strict at every typed manifest field. Unknown fields are rejected;
+the schema version remains `narratio.session-manifest.v1` and there is no
+compatibility reader that silently drops retired state. Cleanup progress uses
+only the cleanup entry in `stages`.
+
+Every persisted session manifest has a non-nil `stages` map. Each entry is
+non-nil and its embedded `name` exactly matches its map key; loading never
+invents either marker. A missing or unsupported schema version, or incomplete
+stage identity, is unsupported local state: move it aside and regenerate it,
+or restore a verified current committed snapshot when that is authoritative.
+
+Prepare records independent `party` and `players` input checksums. In canonical
+party mode, the party record retains its campaign source identity while the
+players record uses `derived_from_party`; raw roster content is never embedded
+in manifest metadata. Both records remain the durable authority for consumers
+of their prepared input source IDs.
+
+The model admits these stage states:
+
+- `pending`
+- `running`
+- `succeeded`
+- `failed`
+- `skipped`
+- `stale`
+- `interrupted`
+
+### Analyze-owned artifact state
+
+Every `analyze` stage record carries `analyze_state_version: 1` and an
+`analyze_artifacts` map keyed by normalized configured artifact key. The
+version is the authority marker: version 1 with no entries is a valid evaluated
+empty set. An absent or unsupported version is unsupported local state and is
+rejected rather than being treated as aggregate-only authority.
+
+Each analyze artifact record has one disposition:
+
+- `current`: the configured artifact is available and carries a versioned
+  fingerprint plus a complete output record and separate output size;
+- `stale`: the recorded semantic identity is no longer current;
+- `missing`: no validated current result exists;
+- `failed`: the attempted work failed and carries a bounded diagnostic; or
+- `unselected`: the artifact was intentionally outside the evaluated set.
+
+Records bind their normalized key and dependencies, fingerprint contract when
+evaluated, canonical session-relative output identity when current, producing
+Narratio run, update time, and bounded non-secret Scriptorium provenance and
+diagnostic paths. A current output includes its configured source ID, contract,
+checksum, and positive byte size. Non-current records cannot carry an output,
+so an older file is not advertised through stale, missing, failed, or
+unselected state.
+
+Family-produced records additionally retain optional `family` and
+`character_id` provenance supplied by configuration resolution. These fields
+do not replace the concrete configured key or infer family membership from a
+name, so older records without them remain valid.
+
+The session-stage collection is the reconciled authority across invocations.
+These records remain analyze-owned data inside the fixed stage; they are not
+dynamic stages or generic subtasks.
+
+Analyze checkpoints one complete validated candidate collection after each
+successful artifact and before exposing that artifact to later work. It saves a
+detached manifest candidate and installs its version, collection, and updated
+timestamp into live state only after the store reports success. Aggregate stage
+outputs are ordinary result outputs rebuilt in configured-key order from current
+session records only.
+
+Successful incremental execution replaces only evaluated artifact records and
+preserves valid unrelated current records. Rebuilt outputs are compared by
+bytes and contract: an unchanged identity permits an unselected dependent with
+the same recomputed fingerprint to remain current, while a changed identity
+removes output authority from every unselected transitive dependent by marking
+it stale. A partial analyze invocation can therefore succeed while unrelated
+configured records remain stale. Existing canonical files never create current
+records without validated execution and a successful checkpoint.
+
+Aggregate analyze status is deliberately coarser than this collection. Resume
+validation may skip a succeeded aggregate record when the selected artifact
+closure is current even if unrelated records are stale. Conversely, a stale
+aggregate record may cross the ordinary runner boundary and perform zero
+Scriptorium calls when reconciliation proves every selected artifact current;
+ordinary successful stage completion then restores the aggregate status.
+
+On an artifact error, analyze installs a complete retry collection on the live
+manifest before returning. The active record is failed without output, current
+transitive dependents are stale, unrelated current records survive, and only
+earlier checkpointed completions remain current. The runner's ordinary failure
+path marks aggregate analyze failed, clears aggregate outputs, invalidates
+delivery dependents, and saves that collection. A terminal session-save failure
+is joined with the original adapter, validation, filesystem, cancellation, or
+checkpoint cause.
+
+If an artifact checkpoint fails, analyze rolls its working collection back to
+the last committed collection before constructing the active failed record.
+Newly materialized canonical bytes therefore remain incidental and cannot
+become current authority. A later ordinary run reconciles the durable records,
+reuses earlier current artifacts, and resumes with the failed or stale work.
+The same recovery applies after process interruption: the aggregate record may
+be non-successful, but every completed artifact checkpoint remains reusable.
+
+## Remote Commit Manifest
+
+`artifacts.RemoteCommitManifest` is a separate, versioned remote snapshot
+contract. It is not a serialized session manifest and contains no local
+post-publication assertion such as `current_pointer_written`. A remote commit
+identifies one campaign, session, and run and declares its immutable artifact
+set. Each artifact has a typed source, immutable destination key, SHA-256
+checksum, size, and storage generation.
+
+`current/commit-pointer.json` is the sole mutable selector for the new
+contract. It identifies exactly one run-scoped `runs/{run_id}/commit.json` and
+binds that object by checksum, size, and generation. Readers strictly reject
+unknown fields, version mismatches, pointer/commit identity mismatches, and
+objects that do not match their declaration.
+
+An absent commit pointer means no remote state is published. Republish to
+establish a current immutable snapshot.
+
+## Persistence Semantics
+
+`manifest.LocalStore`:
+
+- accepts only the exact supported session-manifest schema version;
+- rejects unknown fields and multiple JSON values;
+- validates stage-map identity without synthesizing missing maps, records, or
+  names;
+- retains only representation-preserving collection normalization;
+- writes through a sibling temporary file and atomic replacement visibility;
+- updates `updated_at` on save.
+
+Manifest replacement provides atomic visibility, not a crash- or power-loss
+durability guarantee. Callers reload after a returned save error before retrying.
+An analyze checkpoint error leaves the prior manifest as per-artifact authority.
+The runner then attempts to persist the bounded failed/stale retry collection;
+if that terminal save also fails, its error is joined with the originating
+failure and the next invocation must reconcile from the last readable manifest.
+
+## Execution Semantics
+
+Planning and execution share one pure pre-execution decision over the recorded
+stage status, force selection, and normalized resume-validation result. That
+decision identifies run versus reuse and any required stale and dependent
+invalidation transition. Planning applies it only to a cloned manifest.
+Execution remains the sole owner of observed stage results and persistence.
+
+For each selected stage, the application runner computes semantic evidence,
+performs any eligible resume validation, applies the pre-execution decision,
+and persists required invalidation before either reusing the existing success
+or beginning an attempt. The same one-stage execution owner marks the attempt
+running before invoking the stage and then records and persists its failed,
+skipped, or successful terminal state before a later stage may start. On
+success it records outputs, logs, generated configuration references, metadata,
+and—when the stage implements the optional contract—a versioned
+semantic-configuration fingerprint. Artifact
+records may include optional contract and external provenance objects; old
+manifests remain compatible when those fields are absent. A successful forced
+rerun marks only succeeded transitive dependent session-stage records stale.
+The application owns a fixed dependency relation distinct from execution order;
+dependents are returned in canonical order. Render and extract therefore never
+stale one another, while either can stale analyze, publish, notify, and cleanup.
+
+Starting an execution clears the current session-stage record's prior outputs,
+logs, generated configuration references, metadata, and semantic fingerprint.
+Failed and skipped
+transitions enforce the same clearing rule directly, while success repopulates
+only fields returned by the new result. Marking a record stale does not clear
+those details because resume validation and diagnosis may still require them
+before execution begins.
+
+Aggregate lifecycle clearing deliberately preserves the analyze-owned
+per-artifact collection. This lets later reconciliation replace only evaluated
+entries without erasing unrelated current results. Other stages retain their
+existing aggregate-only lifecycle behavior and are forbidden from carrying the
+analyze-specific fields.
+
+A stage may explicitly return a skipped disposition and stable reason. The
+runner persists that outcome in the session manifest, clears older outputs,
+logs, generated configuration references, and metadata, then applies any
+bounded details from the current skip and continues. This self-skip is distinct from deciding not to execute an
+already-succeeded stage and is reconsidered on later runs. Skipped results
+cannot contain outputs. An intentional self-skip records the current semantic
+fingerprint because it is a completed, reusable stage result; failed or
+interrupted work never promotes one.
+
+When an already-succeeded stage is skipped, the session manifest deliberately
+retains its existing succeeded record because it remains the cross-invocation
+progress authority. If a stage supplies semantic configuration evidence, reuse first
+requires the persisted positive schema version and lowercase SHA-256 digest to
+match the current resolved stage semantics. Missing semantic evidence, malformed
+evidence, or a mismatch makes the stage and its fixed transitive dependents
+stale. The existing stage-specific resume
+validator runs only after this semantic check succeeds; both checks are
+required. Extraction, analyze, and cleanup have resume validators and may
+reject an otherwise eligible skip when their selected durable evidence is
+obsolete; the runner marks the aggregate record stale and executes it.
+Analyze's validator can still accept a partial selection when only unrelated
+artifact records are stale.
+
+Implemented reuse coverage is deliberately split between aggregate semantic
+evidence and focused durable validators:
+
+| Work | Reuse authority | Focused owners |
+| --- | --- | --- |
+| prepare | aggregate semantic fingerprint | [prepare](stage-prepare.md) |
+| transcribe | aggregate semantic fingerprint | [transcribe](stage-transcribe.md), [WhisperX](../integrations/whisperx.md) |
+| merge | aggregate semantic fingerprint | [merge](stage-merge.md), [Seriatim](../integrations/seriatim.md) |
+| polish | aggregate semantic fingerprint | [polish](stage-polish.md), [Audita](../integrations/audita.md) |
+| normalize | aggregate semantic fingerprint | [normalize](stage-normalize.md), [Seriatim](../integrations/seriatim.md) |
+| trim | aggregate semantic fingerprint | [trim](stage-trim.md), [Scriptorium](../integrations/scriptorium.md), [Seriatim](../integrations/seriatim.md) |
+| render | aggregate semantic fingerprint | [render](stage-render.md), [Seriatim](../integrations/seriatim.md) |
+| extract | aggregate semantic fingerprint plus reference/output validator | [extract](stage-extract.md), [Notarius](../integrations/notarius.md) |
+| analyze artifacts | per-artifact fingerprint, reconciliation, and output validator | [analyze](stage-analyze.md), [Scriptorium](../integrations/scriptorium.md) |
+| publish | aggregate semantic fingerprint plus immediate lock/commit checks | [publish](stage-publish.md), [storage adapter](adapters.md) |
+| notify | aggregate delivery-mode fingerprint | [pipeline overview](overview.md), [configuration](../config.md#notifications) |
+| cleanup | aggregate semantic fingerprint plus publish-evidence and target-absence validator | [cleanup](stage-cleanup.md), [workspace](workspace.md) |
+
+These contracts record resolved choices Narratio can observe, not operational
+runner tuning. External model, module, prompt, profile, and configuration-file
+contents that a tool privately loads remain outside the contract when their
+configured identifier is unchanged; operators must force the affected work
+after such a private content change.
+
+The session manifest is the authoritative stage-progress ledger across
+invocations. It retains the most recently resolved invocation's bounded
+configuration provenance. It identifies the selected profile name and source
+(`default` or `cli`) plus the effective configuration digest, but never a raw
+secret or profile content. This provenance is informational: it does not
+participate in stage resume or cache decisions. A profile change therefore
+invalidates only stages whose semantic configuration changed. When a private
+external-tool model, module, prompt, or profile changes behind an unchanged
+configured identifier, use `--force` for the affected work.
+
+`session plan` computes the same current fingerprint and applies the shared
+decision and invalidation rules to a cloned manifest. It predicts the runner
+decision without persisting session state. The semantic fingerprint owner
+hashes deterministic JSON from stage-owned typed structs; stage providers must
+exclude secrets, complete effective-configuration dumps, and operational
+values that cannot affect canonical results. Concrete coverage is owned by the
+focused stage and integration documents linked above.
+
+Before an explicitly bounded execution starts after `prepare`, the application
+reads the session manifest and accepts only `succeeded` or `skipped` for every
+excluded canonical prefix stage. The first other status or absent record fails
+the request before layout mutation, adapter initialization, session-manifest
+writes, or run-directory creation. Excluded prefix records are not passed to
+resume validators. Records after the selected end are not prerequisites and
+may be made stale by selected work without being scheduled.
+
+After a publish commits remotely, any configured local cleanup is first recorded
+as a session-manifest obligation before deletion begins. Each target becomes
+complete only after its confined deletion (or safe absence check) and a
+successful manifest save. An incomplete obligation is retried when publish
+executes again and retains the committed run and remote identity that authorized
+it; an invocation that does not execute publish does not perform cleanup.
+
+Each invocation derives campaign, session, run, local-path, and remote-prefix
+metadata from the validated resolved configuration as one projection. A persisted
+session manifest must agree on campaign and session identity before execution;
+the current projection is refreshed for every invocation while stage progress,
+inputs, and durable artifacts remain session history.
+
+For handled failures after the session identity transition, the runner records
+the failure on the session ledger and atomically attempts its save while
+preserving the originating error. Abrupt process death remains an accepted case
+where the last session transition can require operator interpretation.
+
+## Invariants
+
+- stage resume/skip decisions are session-manifest driven.
+- semantic fingerprint comparison precedes stage-specific resume validation.
+- only successful and intentional-skipped results promote current semantic
+  evidence.
+- running, failed, and self-skipped stages do not retain result payloads from
+  an earlier success.
+- stale stages retain prior details until replacement execution starts.
+- force reruns stale succeeded stages in the fixed dependency relation.
+- the session manifest is the only durable local progress ledger.
+- remote commitment is established by a verified current pointer and remote
+  commit relationship, never by a mutable session-manifest boolean.
+
+## Implementation And Tests
+
+- Models and transitions: `internal/manifest/manifest.go`
+- Remote commit model and readers: `internal/artifacts/remote_commit.go`,
+  `internal/artifacts/current_state_commit.go`,
+- Persistence and validation: `internal/manifest/store.go`
+- Package tests: `internal/manifest/*_test.go`
+- Assembled execution behavior: `internal/app/runner_test.go`,
+  `internal/app/run_stage_test.go`

docs/internal/overview.md

@@ -0,0 +1,139 @@
+# Internal Overview
+
+This document is the implemented component map for Narratio. Normative system
+boundaries and dependency direction belong in
+[Architecture](../policy/architecture.md). User and operator contracts belong
+in the [CLI](../cli.md), [Configuration](../config.md),
+[Operations](../operations.md), and [Troubleshooting](../troubleshooting.md).
+Externally observable tool and format contracts belong under
+[Integrations](../integrations/).
+
+## Execution Path
+
+```text
+cmd/narratio -> internal/app -> configuration and production composition
+             -> internal/stage -> adapters and external systems
+             -> manifests and artifact resolution -> durable local/remote output
+```
+
+The executable delegates process behavior to the application boundary. The
+application resolves configuration, composes concrete collaborators, acquires
+session safety controls, and runs commands. Pipeline commands execute the
+canonical stage sequence through adapter interfaces, while manifests record
+progress and artifact services resolve durable inputs and outputs.
+
+The pipeline runner follows one explicit invocation lifecycle: inspect bounded
+prerequisites without mutation, prepare the workspace layout, acquire one
+session lock, load or create the authoritative session manifest, compose only
+the collaborators needed by the selected range, execute each selected stage in
+order, and then return the summary. Cleanup, when selected, is the ordinary
+final stage rather than runner finalization work.
+The lock remains held from manifest initialization through finalization, and a
+failed stage or required manifest save prevents later stages from starting.
+
+## Components
+
+| Area | Implemented owners | Responsibility |
+| --- | --- | --- |
+| Executable | `cmd/narratio` | Process entry, standard stream wiring, argument handoff, and exit status. |
+| Application orchestration | `internal/app` | Command dispatch, configuration selection, secret-file environment loading, production composition, session locking, planning, execution, restore, and user-facing reporting. |
+| Configuration | [`internal/config`](configuration.md) | ConfigKit-backed pipeline imports and profiles, canonical party and family expansion, strict YAML loading, defaults, normalization, session templating, and validation. |
+| Pipeline stages | `internal/stage` | Canonical stage registry, shared stage contract, execution dependencies, and implemented stage behavior. |
+| External boundaries | `internal/adapters`, `internal/audio` | WhisperX HTTP, downstream subprocesses, notification, object storage, and S3 audio materialization behind Narratio contracts. |
+| Manifests | `internal/manifest` | Durable session progress, stage transitions, validation, and atomic persistence. |
+| Artifacts and paths | `internal/artifacts`, `internal/pathsafe` | Artifact identities and resolution, local and remote path/key models, current-state discovery, and confined relative destinations. |
+| Previous-session cache | `internal/previouscache` | Deterministic planning and materialization requirements for configured previous-session inputs. |
+| Artifact policy | `internal/artifactpolicy` | Source and destination policy, configured artifact identity validation, and publish destination safety. |
+| Shared models and file operations | `internal/artifactmodel`, `internal/contracts`, [`internal/fileops`](fileops.md) | Transcript and artifact data contracts plus atomic single-file replacement, bounded reads, workspace modes, and root-scoped cleanup helpers. |
+| Logging | `internal/logging` | Application logger construction and shared structured logging behavior. |
+
+The application boundary composes concrete implementations. Stages depend on
+Narratio-level contracts; external transport and SDK details remain in
+adapters. The normative rules for these relationships remain in
+[Architecture](../policy/architecture.md).
+
+Command execution normalizes one private dependency value per invocation.
+Configuration reads, secret-directory reads, remote downloads, object-store
+construction, and restore phase functions use that value throughout the
+command. Public command entry points supply production defaults, while focused
+tests replace only the external or sequencing boundary they exercise without
+mutating package state.
+
+Pipeline execution and `session plan` share the same inclusive contiguous-range
+model and the same pure pre-execution decision over recorded status, force, and
+normalized resume validation. Planning applies that decision and modeled
+selected-stage transitions to cloned session state; it does not persist, create
+invocation state, initialize stage-execution adapters, or claim to observe an
+execution result. Command configuration loading can still retrieve a
+missing session file through configured remote storage. It retains the initially
+composed pipeline and selected campaign while resolving either a local or
+downloaded remote session, so one invocation cannot mix pipeline revisions.
+Analyze planning additionally exposes the artifact closure's targets,
+prerequisite rebuilds, execution order, and current reuse.
+
+## Pipeline Stage Set
+
+The implemented canonical order is:
+
+1. [`prepare`](stage-prepare.md)
+2. [`transcribe`](stage-transcribe.md)
+3. [`merge`](stage-merge.md)
+4. [`polish`](stage-polish.md)
+5. [`normalize`](stage-normalize.md)
+6. [`trim`](stage-trim.md)
+7. [`render`](stage-render.md)
+8. [`extract`](stage-extract.md)
+9. [`analyze`](stage-analyze.md)
+10. [`publish`](stage-publish.md)
+11. `notify` (no-op)
+12. [`cleanup`](stage-cleanup.md)
+
+`notify` currently has no persisted pipeline outputs and uses the explicit
+`noop` notification mode. Its versioned semantic evidence records that delivery
+mode and excludes adapter credentials and response data. The focused stage
+documents own implementation mechanics. The
+[CLI](../cli.md) and [Operations](../operations.md) own user-visible invocation
+and execution semantics.
+
+Execution order and invalidation are separate application contracts. The stage
+registry owns the flat execution sequence. The application orchestration owner
+uses a fixed, validated dependency relation to find transitive dependents in
+canonical order. In particular, `render` and `extract` are sibling consumers of
+trimmed transcript state: neither invalidates the other, while either can stale
+`analyze`, `publish`, `notify`, and `cleanup`. Cleanup is a dependent of notify,
+so any earlier replacement can stale a prior cleanup result.
+
+## Focused Documentation
+
+- [Configuration Internals](configuration.md): pipeline imports and profiles,
+  field ownership, decoding, and root-relative path semantics.
+- [Adapter Internals](adapters.md): external adapter boundaries, composition,
+  failure behavior, and test surfaces.
+- [Artifact Internals](artifacts.md): source identities, runtime catalog,
+  resolution, previous requirements, and current-state helpers.
+- [Manifest Internals](manifest.md): session records, persistence, and execution
+  transitions.
+- [Storage Internals](storage.md): object-store interface and S3 behavior.
+- [Workspace Internals](workspace.md): local layout, locking, and cleanup
+  guardrails.
+- [Restore Internals](command-restore.md): discovery, planning, execution, and
+  reporting.
+- [`prepare`](stage-prepare.md)
+- [`transcribe`](stage-transcribe.md)
+- [`merge`](stage-merge.md)
+- [`polish`](stage-polish.md)
+- [`normalize`](stage-normalize.md)
+- [`trim`](stage-trim.md)
+- [`render`](stage-render.md)
+- [`extract`](stage-extract.md)
+- [`analyze`](stage-analyze.md)
+- [`publish`](stage-publish.md)
+- [`cleanup`](stage-cleanup.md)
+
+Use this map to find an owner, then read its focused documentation and tests
+before changing behavior.
+
+The stage registry is implemented in `internal/stage/placeholders.go` and its
+ordering is protected by `internal/app/planner_test.go`. Cross-invocation skip,
+force, failure, and invalidation behavior is exercised in
+`internal/app/runner_test.go` and `internal/app/run_stage_test.go`.

docs/internal/stage-analyze.md

@@ -0,0 +1,210 @@
+# Stage: analyze
+
+## Purpose
+
+Reconcile configured Scriptorium artifacts, execute only required work in
+dependency order, and safely materialize validated outputs.
+
+## Inputs
+
+- ordinary configured artifacts from `pipeline.scriptorium.artifacts`; canonical
+  party artifact families have already expanded into this map during
+  configuration resolution, including corresponding member dependencies and
+  rewritten member-artifact input sources. Each artifact's dependency list is
+  already the normalized union of explicit ordering edges and configured-artifact
+  input edges.
+- optional selected artifact keys supplied through the stage environment
+- built-in, configured, extraction, and previous-session source references in
+  artifact inputs
+
+Supported source families:
+- built-ins: `narratio.transcript.*`, `narratio.bounds.session`
+- prepared stable inputs: `narratio.input.players`, `narratio.input.party`,
+  `narratio.input.glossary`, `narratio.input.spell_catalog`
+- configured artifacts: `narratio.artifact.<key>`
+- extraction lanes: `narratio.extraction.<key>`
+- previous-session cache: `narratio.previous_session.artifact.<key>`
+
+## Outputs
+
+- one current per-artifact manifest record per validated materialized output
+- stage metadata describing selected/generated/reused artifacts
+
+## Key Behavior
+
+- when `pipeline.scriptorium` is absent or no configured artifact is
+  executable, completes successfully and records explanatory metadata. Before
+  returning, it reconciles and checkpoints the complete session collection:
+  records removed from configuration are dropped, while valid records for
+  configured but unselected artifacts remain authoritative. This is not an
+  explicit self-skip: the session manifest records success, satisfies publish's
+  prerequisite, and an ordinary later run reuses the result while the effective
+  set remains empty. Enabling or selecting an artifact later makes missing
+  versioned evidence non-resumable and schedules it without requiring force.
+- builds a runtime artifact catalog containing built-ins, configured artifacts,
+  and configured extraction lanes. Extraction availability is hydrated only
+  from compatible successful extraction evidence.
+- uses enabled configured artifacts by default. An explicit `--artifacts`
+  selection is a one-invocation override that makes exactly the named
+  configured artifacts explicit targets even when disabled. The work planner
+  adds required configured prerequisites, reuses current ones, and schedules
+  stale, missing, or otherwise non-current prerequisites before dependents.
+- makes a non-executable configured artifact reusable only when its current
+  manifest record and durable output pass the configured-artifact evidence
+  contract; an incidental or stale canonical file is unavailable.
+- validates selected artifact dependency order (cycle-safe topo ordering).
+- consumes only the concrete dependency graph finalized by configuration;
+  planning, prerequisite closure, execution, reuse, invalidation, and semantic
+  evidence do not infer a second graph from inputs.
+- resolves required/optional inputs per artifact source definition into an
+  ordered semantic identity. Each identity records the configured input name,
+  canonical source ID, required policy, explicit presence, source contract,
+  checksum, size, and a source-based logical identity. Workspace paths and
+  producer run IDs are excluded.
+- orders input identities by configured input name independently of Go map
+  iteration. Runtime adapter paths remain a separate execution-only map.
+- omits an unavailable optional input from the adapter request while retaining
+  explicit absence in its semantic identity; an unavailable required input
+  fails.
+- resolves prepared stable input sources through the shared manifest-authoritative
+  identity resolver; it does not accept incidental files or fall back to
+  campaign/session source paths.
+- reuses checksums and sizes from validated prepared, extraction, and current
+  configured-artifact evidence. Other resolved inputs are hashed as confined
+  regular files with streaming reads and the central resolved-artifact size
+  limit.
+- owns a versioned SHA-256 fingerprint contract with one fixed-field canonical
+  JSON payload and no map serialization. Configured artifacts are fingerprinted
+  in deterministic dependency order.
+- fingerprints the normalized artifact key, prompt and profile identifiers,
+  effective render-debug behavior, session-relative output identity, sorted
+  dependency keys, ordered input declarations and semantic identities,
+  validated current dependency-output identities, and sorted effective
+  Scriptorium variables (including Narratio's sticky session variable).
+- provides read-only reconciliation that classifies each configured record as
+  current, stale, missing, failed, or otherwise non-resumable, and
+  separately identifies manifest records removed from current configuration.
+  A record is current only when its fingerprint version and value match and its
+  configured output still passes manifest-authoritative evidence validation.
+- owns a read-only typed work planner. Its explicit targets are enabled
+  artifacts by default or the exact normalized `--artifacts` selection when
+  supplied. It closes targets over configured prerequisites, orders the closure
+  topologically, reuses current members, and schedules every non-current member
+  before its dependents.
+- force applies only to explicit targets. A current prerequisite is reused
+  unless it is itself an explicit forced target; disabled prerequisites may be
+  rebuilt when required, while unrelated disabled artifacts are excluded.
+- the work plan carries explicit targets, prerequisite-only work, deterministic
+  execution and reuse lists, invalidated and removed records, and a cloned
+  reconciled starting collection. Valid unrelated configured records survive
+  reconciliation, removed records are omitted, and incidental files never
+  become current without manifest-backed regeneration.
+- implements aggregate resume validation by running the same read-only catalog,
+  fingerprint reconciliation, and work planner used by execution. A succeeded
+  aggregate record is reusable exactly when the selected closure schedules no
+  artifact work and no removed configuration records require a checkpoint;
+  stale unrelated records do not block a partial selection.
+- exposes the typed artifact decision to `session plan`. Planning applies it to
+  a cloned manifest after modeling earlier selected stage transitions, so
+  aggregate run/skip and artifact execute/reuse decisions match the ordinary
+  runner without creating durable state or invoking Scriptorium.
+- executes only the work plan's scheduled entries. Manifest-validated current
+  prerequisites remain available through the runtime catalog without invoking
+  Scriptorium; newly produced prerequisites enter that catalog with the same
+  contract, checksum, and size identity used for persisted current evidence.
+- before the first adapter call, checkpoints the reconciled starting collection
+  when it removed configuration records or cleared obsolete authority. After
+  each successful artifact, checkpoints the complete collection before making
+  that artifact available to the next scheduled dependent.
+- keeps adapter output in the invocation's run-local analyze directory until
+  it is a safe, non-empty, bounded regular file with a calculated checksum and
+  complete output contract. Canonical replacement uses the shared atomic file
+  operation boundary and verifies that the installed checksum matches the
+  validated run-local bytes.
+- records each successful artifact's freshly computed fingerprint, canonical
+  relative output path, contract, checksum, size, producer run ID, bounded
+  Scriptorium provenance, logs, and generated configuration references in the
+  analyze-owned manifest collection.
+- preserves valid unrelated current records during partial execution. If a
+  rebuilt output's bytes and contract are unchanged, unselected dependents may
+  remain current. If that semantic identity changes, unselected transitive
+  dependents become stale without being executed; dependents included in the
+  invocation are evaluated in dependency order instead.
+- returns aggregate outputs from every current session record in normalized-key
+  order after the complete reconciled collection has been checkpointed;
+  unrelated stale records do not make an otherwise successful partial
+  invocation fail.
+- resolves previous-session sources from local `previous/` cache only.
+- runs optional render-debug, then artifact execution.
+- validates non-empty output files and materializes canonical outputs.
+
+## Failure Semantics
+
+- required missing configured/previous-session inputs fail.
+- missing required prepared stable input source includes prepare rerun guidance.
+- missing required previous-session source includes prepare rerun guidance.
+- missing required `narratio.transcript.final_markdown` or
+  `narratio.transcript.final_trimmed_markdown` inputs includes render rerun
+  guidance.
+- dependency cycles or unavailable required dependencies fail.
+- adapter validation failures fail stage.
+- a scheduled artifact failure installs a retry collection on the live manifest
+  with the active artifact marked `failed`, a bounded error, and no output
+  authority. Current transitive dependents become stale without execution.
+- earlier artifacts remain current only after their run-local output passed
+  validation, canonical materialization, and a successful complete-collection
+  manifest checkpoint; unattempted later artifacts retain their reconciled
+  session state.
+- unrelated current records survive a partial failure. Old canonical bytes for
+  the failed artifact and newly materialized bytes whose checkpoint fails are
+  incidental, not current evidence.
+- a checkpoint failure stops the queue immediately. The in-memory retry
+  collection starts from the last successful checkpoint, marks the active
+  artifact failed without output authority, and stales its current transitive
+  dependents.
+- the runner's ordinary error path marks aggregate analyze failed, clears its
+  aggregate outputs, invalidates publish, notify, and cleanup through the
+  application dependency relation, and persists the live retry collection.
+  Terminal-save errors are joined with the original artifact or checkpoint
+  failure.
+
+## Invariants
+
+- `analyze` performs no remote storage calls for previous-session source resolution.
+- input-identity resolution is read-only: it does not invoke adapters,
+  materialize outputs, update status, or create run-local files.
+- fingerprints exclude timeouts, retries, timestamps, producer and Narratio run
+  IDs, executable and config paths, workspace roots, diagnostic locations, and
+  executable or private transitive configuration contents. A change that is
+  visible only inside Scriptorium—such as a file privately loaded by its config
+  path—requires an explicit forced regeneration.
+- output provenance and metadata are deterministic per execution.
+- a canonical file without current per-artifact manifest evidence is never
+  promoted to current state.
+- an ordinary retry after interruption or partial failure reuses every valid
+  earlier checkpoint and schedules only failed, stale, missing, or otherwise
+  non-current work in dependency order.
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#scriptorium-artifact-entries) owns artifact
+  fields and source-selection rules, including
+  [artifact families](../config.md#scriptorium-artifact-families).
+- [CLI](../cli.md) owns user-visible artifact selection.
+- [Scriptorium](../integrations/scriptorium.md) owns the subprocess contract.
+- Implementation and tests: `internal/stage/analyze.go`,
+  `internal/stage/analyze_artifact_execution.go`,
+  `internal/stage/analyze_execution_state.go`,
+  `internal/stage/analyze_input_identity.go`,
+  `internal/stage/analyze_fingerprint.go`,
+  `internal/stage/analyze_reconciliation.go`, `internal/stage/analyze_plan.go`,
+  `internal/stage/analyze_resume.go`, `internal/stage/analyze_test.go`,
+  `internal/stage/analyze_checkpoint_test.go`,
+  `internal/stage/analyze_input_identity_test.go`,
+  `internal/stage/analyze_fingerprint_test.go`,
+  `internal/stage/analyze_reconciliation_test.go`,
+  `internal/stage/analyze_plan_test.go`,
+  `internal/stage/analyze_incremental_execution_test.go`,
+  `internal/stage/analyze_failure_test.go`,
+  `internal/stage/analyze_resume_test.go`, and
+  `internal/app/analyze_checkpoint_recovery_test.go`.

docs/internal/stage-cleanup.md

@@ -0,0 +1,54 @@
+# Stage: cleanup
+
+## Purpose
+
+Remove explicitly configured local data for the run selected by a successful
+publish commit. Cleanup is the twelfth and final canonical stage, after
+`notify`.
+
+## Authorization And Targets
+
+Cleanup is disabled unless at least one of these policies is true:
+
+- `pipeline.spool.delete_audio_after_publish`
+- `pipeline.workspace.cleanup_after_publish`
+
+An enabled cleanup requires the manifest's publish stage to be succeeded with
+`uploaded=true`, a canonical `published_run_id`, and the exact remote commit and
+current commit-pointer keys for the manifest campaign/session prefix. The
+configured campaign and session must match the manifest. Cleanup derives its
+only possible targets through the canonical artifact path helpers:
+
+- the published run's session spool-audio directory; and
+- the published run's workspace run directory.
+
+It does not scan parent directories, infer the newest run, or read object
+storage. Every deletion uses the root-confined file operation, which rejects a
+root itself, traversal, symlink escape, and out-of-root targets.
+
+## Execution And Resume
+
+Disabled cleanup records an explicit skipped result. Enabled cleanup removes
+targets in deterministic policy order and treats an already-missing target as
+success. A successful record contains the published run ID and completed policy
+names.
+
+A previous cleanup success is reusable only while its semantic configuration
+matches and every selected target remains absent. A recreated target makes the
+stage non-resumable. A failure is recorded against cleanup; publish remains
+succeeded and remote current state is unchanged. On a later normal full run,
+valid upstream stages are skipped and cleanup is retried without republishing.
+
+`narratio publish` selects publish only and never invokes cleanup. A bounded run
+through `notify` or any earlier endpoint also excludes it. Operators may run
+`narratio run-stage cleanup <session_id>` directly; missing, failed, malformed,
+or identity-mismatched publication evidence fails before deletion.
+
+## Implementation And Tests
+
+- Planning and evidence validation: `internal/stage/cleanup_plan.go`
+- Execution and resume: `internal/stage/cleanup.go`
+- Root-confined deletion: `internal/fileops/cleanup.go`
+- Focused tests: `internal/stage/cleanup_plan_test.go`,
+  `internal/stage/cleanup_test.go`, and
+  `internal/app/cleanup_orchestration_test.go`

docs/internal/stage-extract.md

@@ -0,0 +1,118 @@
+# Internal: Extract Stage
+
+## Responsibility
+
+`extract` runs after `render` and before `analyze`. It converts the canonical
+`narratio.transcript.final_trimmed` JSON into configured Notarius lane artifacts.
+An omitted or disabled Notarius section makes the stage explicitly self-skip
+with reason `notarius_disabled`, no outputs, and no Notarius runner.
+
+The external protocol is documented in the
+[Notarius integration contract](../integrations/notarius.md). Configuration
+fields belong in [Configuration](../config.md), and physical paths and force
+procedures belong in [Operations](../operations.md).
+
+## Lifecycle
+
+`internal/stage/extract.go`:
+
+1. resolves the final trimmed transcript from the shared artifact catalog;
+2. resolves every configured prepared reference through the shared
+   manifest-authoritative identity resolver before creating run-local output;
+3. streams each verified reference into an invocation-local snapshot and
+   rejects any source change observed while copying;
+4. fingerprints the byte- and provenance-bearing Notarius invocation evidence,
+   including sorted reference identities;
+5. creates a run-local staging directory and invokes the injected
+   `notarius.Runner`;
+6. revalidates the reference snapshots, then validates the v2 successful
+   receipt, confined index, management documents, configured required lane
+   descriptors, validation summaries, and regular payload files;
+7. atomically promotes the complete bundle to its immutable durable location;
+8. records one non-selectable `notarius_index` output and one selectable
+   `notarius_lane` output per configured lane; and
+9. registers each lane as `narratio.extraction.<output_key>` for downstream
+   Scriptorium and publish resolution.
+
+Lane records retain checksum, contract, producer run ID, and Notarius system,
+run, pipeline, and lane provenance. Stage metadata retains the durable bundle
+root, receipt, diagnostic paths, rejection/warning summaries, producing
+Narratio run ID, the resolved trimmed-input identity, and invocation
+fingerprint. The input identity binds the exact transcript bytes, canonical
+source ID, producer stage/output/run identity, and resolution provenance.
+Reference metadata contains only selector, source ID, canonical session-relative
+path, checksum, and size; adapter requests receive selector and absolute
+invocation-local snapshot path, never payload contents. Snapshot bytes must
+match the prepared identity both before and after Notarius runs, so a concurrent
+prepared-file replacement cannot make recorded provenance describe different
+bytes from those supplied to Notarius.
+Validation completes before the single durable rename, so a rejected result
+cannot expose a partial durable bundle.
+
+Any executed extraction outcome that replaces a different effective outcome
+marks succeeded analysis and delivery dependents stale. Render is an independent
+sibling and remains current. Repeating the same disabled self-skip with no
+outputs is stable and does not repeatedly invalidate dependent stages.
+
+## Resume Validation
+
+Before the focused validator runs, the application compares extract's versioned
+semantic fingerprint. It covers enablement, Notarius pipeline identity, sorted
+reference selector/source mappings, sorted declared output contracts, and each
+canonical `narratio.extraction.<key>` output identity. It excludes executable,
+timeout, working directory, config path, and private Notarius config contents.
+
+`internal/stage/extract_resume.go` then permits a skip only when the existing
+stage record still matches the current byte- and provenance-bearing invocation
+evidence. That evidence covers the current direct trimmed-transcript identity,
+sorted prepared-reference identities, pipeline identity, and configured output
+contracts. The same reference helper and transcript identity are resolved again
+for artifact evidence, so changing current transcript bytes, reference bytes,
+or producer identity makes the prior extraction obsolete. Operational runner
+settings do not invalidate otherwise current durable evidence.
+
+A valid prepared-reference change makes extraction non-resumable. Missing,
+unsafe, or checksum-inconsistent prepared evidence is a hard validation error
+with prepare-force guidance because an immediate extract rerun cannot succeed.
+
+The validator then checks the producing run identity, canonical immutable
+bundle root, path confinement and absence of symlink components, receipt
+identity, exactly one canonical index, the exact configured source set,
+contracts and provenance, regular-file status, and stored checksums. Missing or
+obsolete results are non-resumable and run again; unsafe filesystem conditions
+return an error rather than silently accepting or replacing data.
+
+Neither contract can observe files imported by Notarius configuration, profile
+contents, prompt/module definitions, or other transitive inputs. Operators must
+force extraction after changing any such private input behind a stable
+identifier.
+
+## Failure Behavior
+
+Adapter startup, timeout, nonzero exit, receipt decoding, path confinement,
+index compatibility, inconsistent warning or diagnostic envelopes,
+required-lane rejection or incomplete validation, payload inspection,
+checksum, or installation errors fail the stage through ordinary manifest
+transition handling.
+Stdout receipt and stderr diagnostics remain separate. Downstream stages are
+not given selectable extraction sources unless the complete configured result
+has passed validation and installation.
+
+When a replacement attempt begins, the current session-stage record no longer
+advertises payload from the previous success. A failed replacement therefore
+has no current outputs, logs, generated configuration references, or metadata,
+while the earlier immutable promoted bundle remains available for audit and
+recovery.
+
+## Implementation And Focused Tests
+
+- Stage execution, selection, and resume validation: `internal/stage/extract.go`,
+  `internal/stage/extract_resume.go`,
+  `internal/stage/extract_test.go`,
+  `internal/stage/semantic_contracts_delivery.go`
+- Subprocess boundary: `internal/adapters/notarius/subprocess.go`,
+  `internal/adapters/notarius/subprocess_test.go`
+- Catalog hydration: `internal/artifacts/extraction_catalog.go`,
+  `internal/artifacts/extraction_catalog_test.go`
+- Composition and downstream behavior: `internal/app/runner_test.go`,
+  `internal/stage/analyze_test.go`, `internal/stage/publish_test.go`

docs/internal/stage-merge.md

@@ -0,0 +1,51 @@
+# Stage: merge
+
+## Purpose
+
+Normalize raw transcript inputs and merge into base transcript via Seriatim.
+
+## Inputs
+
+- `transcripts/raw/*.json`
+- `inputs/speakers.yml`
+- `inputs/autocorrect.yml`
+
+## Outputs
+
+- `transcripts/base.json`
+- optional `artifacts/seriatim.report.json`
+
+## Key Behavior
+
+- discovers and validates raw transcript inputs.
+- normalizes each raw transcript (`seriatim.Normalize`) into run-local scratch output.
+- merges normalized inputs (`seriatim.Run`) into base transcript.
+- validates merged transcript and optional report JSON.
+- materializes canonical outputs and records stage logs/generated configs.
+
+## Invariants
+
+- merge always consumes normalized forms of raw inputs.
+- base transcript must validate before stage success.
+- report output is config-gated.
+
+## Resume Evidence
+
+Merge records a versioned semantic-configuration fingerprint for the Seriatim
+merge operation, output schema, coalesce gap, and every configured advanced
+merge transformation. A change reruns merge and stales only its fixed
+descendants; prepare and transcribe remain reusable. Binary path, timeout,
+report emission, logs, and diagnostic retention are operational exclusions.
+
+Configuration or resources loaded privately inside Seriatim are outside
+Narratio's observable contract and require `--force` when changed. An existing
+successful merge record without evidence reruns once when selected.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns subprocess and output semantics.
+- [Configuration](../config.md#pipeline) owns operator-selected Seriatim values.
+- Implementation and tests: `internal/stage/merge.go`,
+  `internal/stage/merge_test.go`,
+  `internal/stage/semantic_contracts_initial.go`, and
+  `internal/stage/semantic_contracts_initial_test.go`

docs/internal/stage-normalize.md

@@ -0,0 +1,42 @@
+# Stage: normalize
+
+## Purpose
+
+Normalize polished transcript into final transcript using Seriatim.
+
+## Inputs
+
+- `transcripts/polished.json`
+
+## Outputs
+
+- `transcripts/final.json` (or configured normalize output path)
+- optional `artifacts/seriatim.normalize.report.json`
+
+## Key Behavior
+
+- resolves polished transcript only from verified manifest output evidence.
+- applies `pipeline.normalize` config or default normalize config.
+- runs Seriatim normalize with configured timeout/binary.
+- validates normalized transcript and optional report.
+- materializes canonical outputs and records logs/generated configs.
+
+## Invariants
+
+- final transcript must validate as processed transcript JSON (`segments` array).
+- normalize defaults are applied when `pipeline.normalize` is unset.
+
+## Resume Semantics
+
+The versioned semantic fingerprint covers the Seriatim normalize operation,
+output schema and canonical output identity, plus the configured transcript
+transformations. Seriatim's executable and timeout and optional report
+generation are operational and do not invalidate the normalized transcript.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns subprocess and output semantics.
+- [Configuration](../config.md#pipeline) owns normalize fields and defaults.
+- Implementation and tests: `internal/stage/normalize.go`,
+  `internal/stage/normalize_test.go`,
+  `internal/stage/semantic_contracts_refinement.go`

docs/internal/stage-polish.md

@@ -0,0 +1,54 @@
+# Stage: polish
+
+## Purpose
+
+Run Audita polishing on base transcript and produce polished transcript.
+
+## Inputs
+
+- `transcripts/base.json`
+- `inputs/glossary.yml`
+
+## Outputs
+
+- `transcripts/polished.json`
+- optional `artifacts/audita.report.json`
+
+## Key Behavior
+
+- resolves base transcript only from verified merge manifest output evidence.
+- invokes an Audita runner configured with static model/runtime options; the
+  invocation supplies paths and modules.
+- validates processed transcript structure (`segments` array required).
+- validates optional report JSON.
+- materializes canonical outputs; records logs/generated config and adapter metadata.
+
+## Invariants
+
+- polished transcript schema validation is mandatory.
+- report output is config-gated.
+
+## Resume Semantics
+
+The versioned semantic fingerprint covers the Audita service endpoint, model,
+validation model, module set, transcript description, output schema, selected
+external configuration path, and canonical polished-transcript identity. Module
+ordering is normalized because the configured modules form a set. Audita's
+executable, timeouts, concurrency, report and debug behavior, work retention,
+and resolved credential value are operational and do not invalidate a
+successful result. The public credential reference name may participate in
+configuration provenance but never exposes its resolved value.
+
+Narratio can fingerprint a selected model, module, or configuration identifier,
+but it cannot inspect content that Audita privately resolves behind that stable
+identifier. Force `polish` after changing such private content without changing
+its identifier.
+
+## Related Contracts And Tests
+
+- [Audita](../integrations/audita.md) owns subprocess, validation, and failure
+  semantics.
+- [Configuration](../config.md#pipeline) owns operator-selected Audita values.
+- Implementation and tests: `internal/stage/polish.go`,
+  `internal/stage/polish_test.go`,
+  `internal/stage/semantic_contracts_refinement.go`

docs/internal/stage-prepare.md

@@ -0,0 +1,99 @@
+# Stage: prepare
+
+## Purpose
+
+Materialize canonical current-session inputs before processing stages.
+
+## Inputs
+
+- resolved campaign, session, and pipeline configuration
+- stable input files (`speakers`, `autocorrect`, `glossary`, `players`, `party`)
+- optional spell-catalog overlay
+- one resolved local or S3 audio source
+- enabled configured artifact input requirements for previous-session sources
+
+## Outputs
+
+- `inputs/campaign.yml`
+- `inputs/session.yml`
+- `inputs/pipeline.resolved.yml`
+- `inputs/speakers.yml`
+- `inputs/autocorrect.yml`
+- `inputs/glossary.yml`
+- `inputs/players.yml`
+- `inputs/party.yml`
+- optional `inputs/spell_catalog.json`
+- `audio/*.flac`
+- optional `previous/manifest.json`
+- optional `previous/artifacts/**`
+- deterministic `manifest.inputs` entries (checksums + provenance)
+
+## Key Behavior
+
+- validates required config/store state.
+- enforces local audio vs S3 audio mutual exclusivity.
+- rejects duplicate explicit local audio sources after resolution.
+- gives distinct local source paths with the same basename deterministic unique
+  prepared filenames so neither source is overwritten.
+- materializes S3 audio through spool/cache-aware logic.
+- materializes a configured spell catalog with checksum and provenance, or
+  safely removes an obsolete canonical spell catalog and its manifest record
+  when the effective input is omitted.
+- copies the validated canonical party bytes unchanged and deterministically
+  generates the prepared players projection.
+- scans enabled configured artifact inputs for `narratio.previous_session.artifact.*` requirements.
+- clears managed `previous/` state on every invocation, then, when requirements exist:
+  - resolves the pointer-selected previous source through the shared resolver;
+  - downloads previous manifest/artifacts;
+  - records previous inputs in `manifest.inputs`.
+
+Required previous-session inputs fail when unavailable; optional missing inputs
+are typed skipped results. Each requirement resolves only from its unique
+`published_output` declaration in the pointer-selected remote commit. Prepare
+strips that selected run's immutable `published/` prefix to derive the local
+cache path, then validates the committed checksum, size, and generation.
+
+## Invariants
+
+- only `prepare` hydrates canonical `previous/` cache state.
+- managed previous artifacts are stored under `previous/artifacts/**` without
+  duplicate `artifacts/artifacts/` nesting.
+- managed `previous/` state represents only the current requirement set.
+- `manifest.inputs` ordering is deterministic (`kind`, `path`).
+
+## Resume Evidence
+
+Prepare records a versioned semantic-configuration fingerprint for the
+resolved campaign/session selection, local-versus-S3 audio mode and canonical
+audio names, stable-input ownership/presence, previous-session identity, and
+the party mode plus canonical players projection version, and the effective
+previous-artifact requirement set. A change reruns prepare and
+stales its fixed descendants. Existing successful records without this
+evidence rerun once when selected.
+
+Workspace, spool, and cache placement and absolute source relocation are not
+semantic when logical selection, canonical names, and bytes are equivalent.
+The fingerprint deliberately does not read or rehash large audio. Prepared
+input checksums remain the content provenance. Before reusing success, prepare
+validates every durable prepared copy and compares current stable-input bytes,
+canonical party and derived-player bytes, local audio membership/checksums, or
+S3 key/size/entity-tag identity with that provenance. Source relocation with
+equivalent names and bytes remains reusable; changed or unavailable evidence
+causes a normal prepare rerun.
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md) owns audio selection, stable input fields, and
+  previous-session settings.
+- [Operations](../operations.md) owns physical input, audio, spool, cache, and
+  previous-state layout.
+- [Storage Internals](storage.md) and [Artifact Internals](artifacts.md) explain
+  the internal collaborators.
+- Implementation and tests: `internal/stage/prepare.go`,
+  `internal/stage/prepare_test.go`,
+  `internal/stage/prepare_resume.go`,
+  `internal/stage/prepare_resume_test.go`,
+  `internal/stage/semantic_contracts_initial.go`,
+  `internal/stage/semantic_contracts_initial_test.go`,
+  `internal/audio/s3_audio_test.go`,
+  `internal/previouscache/*_test.go`

docs/internal/stage-publish.md

@@ -0,0 +1,147 @@
+# Stage: publish
+
+## Purpose
+
+Upload run/session outputs to object storage and atomically advance remote current state.
+
+## Inputs
+
+- successful preceding stages from the [canonical stage set](overview.md#pipeline-stage-set)
+- declared run-local diagnostic files
+- resolved publish output rules
+- independent per-source remote lock objects read at planning start
+- the validated prior current commit when an initially locked source may carry
+- durable previous-session cache files when present
+
+## Outputs
+
+- uploaded session manifest, declared run-local diagnostics, and selected
+  publish outputs;
+- uploaded durable previous-session cache files when present;
+- immutable run-scoped commit manifest; and
+- current commit pointer, written last.
+
+Exact remote placement and the operator workflow belong in
+[Operations](../operations.md#publish-workflow).
+
+## Key Behavior
+
+`publish.go` retains the transaction-owning `Run` path. Source discovery,
+manifest-derived archive metadata, confined opening, and checksum validation
+live in `publish_sources.go`; artifact-catalog matching, selection, required
+and optional rules, destination resolution, and collision inputs live in
+`publish_outputs.go`. Copying, hashing, immutable upload, reuse checks, and
+remote verification live in `publish_transfer.go`; committed-manifest,
+immutable-commit, and current-pointer encoding and writes live in
+`publish_commit.go`. `publish_output_plan.go` owns the deterministic, I/O-free
+five-outcome model and its secret-free metadata projection.
+
+- when publishing or run upload is disabled, completes successfully with no
+  outputs and records explanatory metadata. This is not an explicit self-skip:
+  the session manifest records success. Enablement and upload policy are
+  fingerprinted, so changing either automatically makes the prior result
+  non-resumable.
+- validates prerequisite stage success and object-store availability.
+- derives a deterministic run-archive allowlist from the session manifest's
+  declared current-run outputs, logs, and generated configurations. References
+  outside the current run root, including canonical session files and prior-run
+  diagnostics, are not archive candidates. No distinguished diagnostic file is
+  read or uploaded.
+- opens each archive candidate beneath its archive root without following
+  symlinked ancestors or leaf entries, verifies that it is a regular file and
+  checks a declared checksum when present, then streams the opened descriptor.
+- derives the durable previous-cache archive from its validated manifest using
+  the same confinement and regular-file checks.
+- resolves publish output sources through runtime artifact catalog and
+  manifest-aware resolution. Configured Scriptorium outputs are publishable
+  only from validated `current` per-artifact analyze evidence; an incidental
+  canonical file, aggregate output, stale/failed/unselected record, or
+  mismatched path, size, or checksum remains unavailable. Built-in and
+  extraction sources likewise require their own current manifest evidence.
+- publishes extraction lanes only through explicit configured output rules;
+  neither run-local nor durable Notarius bundles are scanned or uploaded wholesale.
+- selected artifact filter applies to configured artifact sources only.
+- plans every configured output into exactly one source-sorted outcome:
+  `published`, `carried_locked`, `locked_unpublished`, `optional_missing`, or
+  `unselected`.
+- an initially locked selected output carries its exact prior
+  `published_output` declaration when one exists. The declaration may point to
+  an earlier run in the same session and retains its source, key, checksum,
+  size, and generation; the bytes are not copied.
+- verifies a carried object's exact key, size, and generation through object
+  metadata without downloading or opening the local output source.
+- an initially locked source with no prior declaration remains absent even
+  when the output rule is required. Optional missing unlocked outputs remain
+  absent; required missing unlocked outputs fail.
+- creates one complete immutable source-to-destination mapping before upload;
+- uploads and verifies every declared immutable object and the commit manifest;
+- writes the immutable commit from the combined new and exact carried
+  declarations.
+- rereads every initially unlocked participating source after the immutable
+  commit. A newly present lock aborts selection, leaving the uploaded immutable
+  attempt unselected. Removing an initial lock may conservatively produce one
+  carry because the initial snapshot controls the plan.
+- updates `current/commit-pointer.json` exactly once, last, and writes no other
+  mutable publication control record.
+
+## Metadata Signals
+
+Includes counts/lists for:
+- run uploads
+- published output uploads
+- previous uploads
+- one `publish_output_outcomes` collection containing all five output states,
+  exact declared keys, and key-derived origin runs where applicable
+- remote commit and current-pointer key paths
+- the run identifier selected by the commit
+
+## Invariants
+
+- `current/commit-pointer.json` is the remote commit marker and is written last.
+- run files, selected outputs, previous-cache files, and the committed session
+  manifest are all declared by an immutable commit under the run prefix.
+- run and previous uploads contain only manifest-declared regular files opened
+  from verified descriptors; symlinks, special files, replacement races, and
+  undeclared entries are rejected or ignored before uploads begin.
+- run-local diagnostics, including Notarius receipt and stderr files, are
+  archived only when recorded by the session manifest. A run with no declared
+  current-run diagnostics legitimately uploads none.
+- independent publish locks are not overridden by `--force`; every initially
+  unlocked relevant source is reread immediately before current-state
+  selection.
+- successful uploaded publication records the exact run ID, remote commit key,
+  and current-pointer key consumed by the later cleanup stage.
+
+## Resume Semantics
+
+The versioned semantic fingerprint covers enabled behavior, run-upload policy,
+normalized source/destination/required output rules, and the remote backend,
+bucket, region, endpoint, and root-prefix identity. Rule ordering is
+canonicalized. Credential
+environment names, path-addressing transport mode, local workspace placement,
+and run identifiers are excluded. Independent remote locks remain mutable state
+and are read directly by publish at planning and commit boundaries; semantic
+evidence does not replace those safety checks.
+
+The commit boundary and cleanup authorization are normative architecture invariants; see
+[Architecture](../policy/architecture.md#publish-commit-boundary).
+
+Publish never deletes local state. The standalone `publish` command selects
+only this stage; a full run selects notify and cleanup afterward.
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#publish-configuration-summary) owns output
+  rules.
+- [Operations](../operations.md#publish-locks) owns remote lock lifecycle and
+  physical remote state.
+- [Artifact Internals](artifacts.md) explains source resolution and current-state
+  helpers.
+- Implementation and tests: `internal/stage/publish.go`,
+  `internal/stage/publish_sources.go`, `internal/stage/publish_outputs.go`,
+  `internal/stage/publish_output_plan.go`,
+  `internal/stage/publish_transfer.go`, `internal/stage/publish_commit.go`,
+  `internal/stage/publish_test.go`,
+  `internal/stage/semantic_contracts_delivery.go`,
+  `internal/app/operator_helpers_test.go`, and
+  `internal/app/cleanup_orchestration_test.go`

docs/internal/stage-render.md

@@ -0,0 +1,60 @@
+# Stage: render
+
+## Purpose
+
+Render Markdown transcript artifacts from normalized JSON transcripts via Seriatim.
+It runs after `trim` and before `extract` in the canonical sequence. Render and
+extract are independent sibling consumers: replacing render output does not
+invalidate extraction, but it does invalidate succeeded analysis and delivery
+records that may consume rendered transcripts.
+
+## Inputs
+
+- `narratio.transcript.final` (`transcripts/final.json`)
+- `narratio.transcript.final_trimmed` (`transcripts/final.trimmed.json`)
+
+## Outputs
+
+- `narratio.transcript.final_markdown` -> `transcripts/final.md`
+- `narratio.transcript.final_trimmed_markdown` -> `transcripts/final.trimmed.md`
+
+## Key Behavior
+
+- uses `pipeline.render` settings (enabled/format/title/booleans).
+- resolves inputs only from verified manifest output evidence.
+- writes run-local outputs first, then materializes canonical session outputs.
+- records input provenance, output paths, adapter metadata, logs, and generated config refs.
+- when `pipeline.render.enabled=false`, completes successfully with no outputs
+  and records explanatory metadata. This is not an explicit self-skip: both
+  manifests record success. Because enablement is fingerprinted, enabling
+  render later automatically makes the prior result non-resumable.
+
+## Failure Semantics
+
+- missing normalized input fails with normalize rerun guidance.
+- missing trimmed input fails with trim rerun guidance.
+- adapter/subprocess failure fails stage.
+- empty render output files fail validation.
+
+## Invariants
+
+- only `format: markdown` is supported.
+- render stage owns production of built-in Markdown transcript sources.
+
+## Resume Semantics
+
+The versioned semantic fingerprint covers enablement, final format, resolved
+title (including the session-title fallback), timestamp, segment-ID and
+metadata inclusion, both canonical input identities, and both Markdown output
+identities. Seriatim's executable, timeout, and report behavior are operational
+and do not invalidate rendered transcripts. A render-only change leaves the
+independent `extract` sibling reusable while invalidating their shared
+downstream consumers.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns render subprocess behavior.
+- [Configuration](../config.md#pipeline) owns render fields and defaults.
+- Implementation and tests: `internal/stage/render.go`,
+  `internal/stage/render_test.go`,
+  `internal/stage/semantic_contracts_refinement.go`

docs/internal/stage-transcribe.md

@@ -0,0 +1,56 @@
+# Stage: transcribe
+
+## Purpose
+
+Generate raw per-speaker transcripts from prepared audio using WhisperX.
+
+## Inputs
+
+- `audio/*.flac` from `prepare`
+
+## Outputs
+
+- `transcripts/raw/<speaker>.json`
+
+## Key Behavior
+
+- discovers prepared audio from manifest inputs or canonical audio directory.
+- derives the transcript identity from the prepared `.flac` filename.
+- dispatches one WhisperX request per prepared file through a bounded errgroup.
+- validates each output as JSON.
+- writes run-local outputs then materializes canonical transcript outputs only
+  after every planned request succeeds.
+
+## Invariants
+
+- prepared audio identities must be unique; prepare disambiguates distinct
+  source paths that share a basename.
+- output path returned by adapter must match requested output path.
+- an empty adapter result path means the requested path; adapters cannot select
+  an alternate destination.
+- each successful output is validated before stage success, and cancellation or
+  incomplete dispatch cannot be reported as a successful result.
+
+## Resume Evidence
+
+Transcribe records a versioned semantic-configuration fingerprint containing
+the Narratio-visible WhisperX service URL and recognition language. Changes to
+either rerun transcription and stale its fixed descendants while leaving
+prepare reusable. Retry count/delay, concurrency, timeout, credentials, and
+diagnostic locations are operational and do not change this evidence.
+
+WhisperX models or private service configuration not exposed by Narratio's
+adapter contract cannot be fingerprinted; use `--force` after changing them.
+An existing successful transcribe record without evidence reruns once when
+selected.
+
+## Related Contracts And Tests
+
+- [WhisperX](../integrations/whisperx.md) owns HTTP, retry, timeout, and
+  cancellation semantics.
+- [Configuration](../config.md#pipeline) owns concurrency and other
+  operator-selected values.
+- Implementation and tests: `internal/stage/transcribe.go`,
+  `internal/stage/transcribe_test.go`,
+  `internal/stage/semantic_contracts_initial.go`, and
+  `internal/stage/semantic_contracts_initial_test.go`

docs/internal/stage-trim.md

@@ -0,0 +1,57 @@
+# Stage: trim
+
+## Purpose
+
+Produce a final-trimmed transcript. By default, the stage generates bounds and
+applies a bounds-driven trim.
+
+## Inputs
+
+- `transcripts/final.json`
+
+## Outputs
+
+- `transcripts/final.trimmed.json` (or configured trim output path)
+- when trim enabled: `artifacts/session_bounds.json`
+
+## Key Behavior
+
+When `trim.enabled=true`:
+- runs Scriptorium bounds artifact generation;
+- optionally runs render-debug output generation;
+- validates bounds payload against transcript;
+- derives keep selector;
+- either copies unchanged transcript or runs Seriatim trim;
+- validates trimmed transcript and materializes bounds output.
+
+When `trim.enabled=false`:
+- copies normalized transcript to trimmed output.
+
+## Invariants
+
+- normalized transcript is required input.
+- bounds output exists only in enabled trim path.
+- render-debug output is diagnostic and not a declared stage output.
+
+## Resume Semantics
+
+The versioned semantic fingerprint covers enablement, the bounds prompt and
+profile identifiers, the Scriptorium configuration identity, transcript input
+name, sticky session variable, bounds and trimmed output identities, and the
+Seriatim trim operation. Diagnostic bounds rendering, diagnostic output paths,
+timeouts, executable paths, and optional reports are operational and do not
+invalidate the canonical trimmed transcript.
+
+Narratio cannot inspect prompt, profile, or configuration content that
+Scriptorium or Seriatim privately resolves behind a stable identifier. Force
+`trim` after changing such private content without changing its identifier.
+
+## Related Contracts And Tests
+
+- [Scriptorium](../integrations/scriptorium.md) owns bounds generation and
+  debug-render subprocess behavior.
+- [Seriatim](../integrations/seriatim.md) owns transcript trimming behavior.
+- [Configuration](../config.md#pipeline) owns trim fields and defaults.
+- Implementation and tests: `internal/stage/trim.go`,
+  `internal/stage/trim_test.go`,
+  `internal/stage/semantic_contracts_refinement.go`

docs/internal/storage.md

@@ -0,0 +1,72 @@
+# Internal: Storage
+
+## Purpose
+
+Explain the object-store interface and S3 implementation used by Narratio.
+Remote key layout and lifecycle belong in [Operations](../operations.md), while
+operator-selected storage fields and credential mechanisms belong in
+[Configuration](../config.md).
+
+## Primary Contract
+
+`storage.ObjectStore` interface:
+
+- `List(ctx, prefix)`
+- `Read(ctx, key)` returns an object body and the generation observed with it
+- `Download(ctx, key, localPath)`
+- `Upload(ctx, localPath, key, opts)`
+- `UploadConditional(ctx, source, key, opts, condition)`
+- `Delete(ctx, key)` removes one object and succeeds when it is already absent
+- `Exists(ctx, key)`
+
+Key invariant:
+- callers pass full bucket-relative keys;
+- storage implementations do not infer campaign/session/run prefixes.
+
+`ReadObjectBounded` is the shared mechanism for small control objects. It opens
+one object version, returns the metadata observed with that body, rejects an
+oversized known size before transfer, and still performs a context-aware
+limit-plus-one read. It closes the body on every exit. Callers own the policy
+limit and add the control-object category to errors; this helper is not used for
+large artifact payloads.
+
+## Composition
+
+`NewObjectStoreFromConfig` constructs the S3-backed implementation from
+resolved configuration. The application loads configured filesystem secrets
+before calling it. The storage adapter consumes already-resolved values; it does
+not own discovery, defaults, or configuration validation.
+
+## S3 Backend Behavior
+
+- normalizes object keys.
+- `List` paginates and returns normalized `ObjectInfo`.
+- A truncated S3 listing must supply a new, non-empty continuation token;
+  otherwise listing fails with bucket and prefix context instead of looping.
+- `Download` writes local files with parent directory creation.
+- `Upload` streams local file and returns remote metadata.
+- `Read` binds a returned body to its S3 ETag. `UploadConditional` maps an ETag
+  match or absence precondition directly to the provider request and reports a
+  failed precondition without performing a local check-then-write replacement.
+- `Delete` maps one normalized key to the provider delete-object operation and
+  treats an absent object as success.
+- `Exists` maps not-found responses to `false`.
+
+## Invariants
+
+- storage layer is stateless regarding manifest/stage progression.
+- bounded reads never retain more than the caller's limit plus one byte and do
+  not replace owner-specific size policy.
+- publish ordering semantics are owned by stage/app code, not storage adapters.
+- A caller that installs remote content into a canonical workspace path first
+  writes a sibling temporary file, validates the complete object against its
+  owner-specific contract, and only then replaces the destination. Temporary
+  files used solely for remote comparisons are removed by their callers and do
+  not represent installed workspace state.
+
+## Implementation And Tests
+
+- Contract and S3 adapter: `internal/adapters/storage`
+- Composition: `internal/app/object_store.go`
+- Tests: `internal/adapters/storage/*_test.go`,
+  `internal/app/object_store_test.go`

docs/internal/workspace.md

@@ -0,0 +1,97 @@
+# Internal: Workspace
+
+## Purpose
+
+Explain the helpers that construct local session and run paths, coordinate
+single-writer access, and confine cleanup. The authoritative physical layout and
+retention workflow belong in [Operations](../operations.md#local-state-layout).
+
+## Path Ownership
+
+`internal/artifacts` owns canonical session, run, spool, cache, and
+previous-cache path construction. `SessionPathsFor` provides the session-scoped
+path model, and layout creation goes through `EnsureLayoutFor`. Callers should
+consume those helpers instead of rebuilding relative paths.
+
+`internal/pathsafe` validates relative destinations. `internal/fileops` opens
+explicit cleanup roots and removes only normalized proper descendants through
+the resulting root handle.
+
+`internal/fileops` owns the ordinary workspace mode contract. On POSIX,
+`WorkspaceDirectoryMode` is setgid `02775` and `WorkspaceFileMode` is `0664`.
+`EnsureWorkspaceDirectory` reapplies the directory mode after creation so a
+restrictive umask cannot remove group access, while retaining existing ownership
+and group. Credential paths are outside this contract; the platform-specific
+operational requirements are in [Operations](../operations.md#workspace-permissions).
+
+## Run-Local Stage Layout
+
+`internal/stage/run_local.go` maps stage outputs and diagnostics into an
+invocation-scoped layout. Successful outputs are validated and atomically
+materialized into canonical session paths before stage success. Managed
+previous-session cache paths remain session-durable and are never redirected
+into run-local output space.
+
+Extraction uses run-local receipt, stderr, and output-root helpers, then
+validates and renames the external bundle to the unique immutable Notarius
+bundle path supplied by `internal/artifacts`. Exact physical paths belong in
+[Operations](../operations.md#extraction-workflow).
+
+## Locking
+
+`artifacts.LocalStore` enforces the single-writer session lock via an
+operating-system lock held on `.lock` (`ErrLockConflict` on contention). The
+file retains owner metadata after release or process death; its existence is
+not evidence that a lock is active. Command and restore flows wait for this
+lock only while their context remains active, and report a release failure.
+
+The runner assigns every invocation a run-workspace identity but creates the
+physical directory only before an executing stage that requires it. Fully
+reused invocations and cleanup-only invocations therefore do not accumulate
+empty run directories.
+
+## Cleanup Semantics
+
+The canonical cleanup stage:
+
+- runs after notify only when selected by the invocation range;
+- requires a succeeded publish record with `uploaded=true`, the published run
+  ID, remote commit key, and current commit-pointer key;
+- consumes the resolved cleanup policy described in
+  [Configuration](../config.md);
+- refuses unsafe deletes (root delete, traversal, and out-of-root targets);
+- treats missing targets as successful, idempotent deletion; and
+- uses its ordinary failed/succeeded manifest stage record for retry. A later
+  full run can skip reusable upstream stages and retry cleanup without
+  republishing.
+
+Manual cleanup uses the same root-confined deletion mechanism. Invocation
+syntax and exact deletion scope belong in [CLI](../cli.md#clean) and
+[Operations](../operations.md#cleanup).
+
+## Invariants
+
+- campaign-aware session root is mandatory.
+- manifest-driven stage state is durable across runs.
+- cleanup guardrails prevent destructive root/out-of-scope deletion.
+- restore installation is confined to validated relative paths beneath the
+  selected session root.
+- ordinary workspace paths retain group-writable directory and file modes across
+  nested creation, replacement, and Notarius promotion.
+
+## Implementation And Tests
+
+- Path model and local store: `internal/artifacts/paths.go`,
+  `internal/artifacts/local.go`
+- Run-local materialization: `internal/stage/run_local.go`
+- Immutable bundle installation: `internal/stage/extract.go`
+- Workspace modes: `internal/fileops/modes.go`
+- Cleanup confinement: `internal/fileops/cleanup.go`,
+  `internal/stage/cleanup.go`, `internal/stage/cleanup_plan.go`, and
+  `internal/app/cleanup_targets.go`
+- Tests: `internal/artifacts/paths_model_test.go`,
+  `internal/artifacts/local_test.go`, `internal/stage/run_local_test.go`,
+  `internal/fileops/modes_posix_test.go`,
+  `internal/fileops/cleanup_test.go`, `internal/app/cleanup_targets_test.go`,
+  `internal/stage/cleanup_test.go`, and
+  `internal/app/cleanup_orchestration_test.go`

docs/operations.md

@@ -0,0 +1,641 @@
+# Operations Guide
+
+Operator workflow for running, recovering, and publishing Narratio sessions.
+
+For command syntax, see [docs/cli.md](./cli.md). For field-level config, see [docs/config.md](./config.md).
+
+## Campaign and Session Selection
+
+Campaign selection priority:
+
+- `--campaign-file`
+- `--campaign`
+- `pipeline.campaigns.default_campaign_id`
+
+Session source priority:
+
+- `--session`
+- local default search paths
+- remote session object (S3) when local session file is not found and storage is configured
+
+## Session Initialization
+
+Use `session init` to generate a concrete session file for local or remote use.
+
+Local file:
+
+```bash
+narratio session init 2026-04-04 --output ./session.yml --date 2026-04-04 --title "Session 12"
+```
+
+Remote session object:
+
+```bash
+narratio session init 2026-04-04 --remote --force
+```
+
+If `campaign.yml` sets `session_template_file`, `session init` renders it. Template variables must resolve to concrete values.
+
+Campaigns must provide stable input files for speakers, autocorrect, glossary,
+players, and party, and may provide an optional spell-catalog overlay. Session
+files may override those paths for one session. The `prepare` stage
+materializes them under `inputs/`; configured consumers use the prepared files,
+never the original campaign or session source paths. Field definitions and
+source IDs are in [Configuration](./config.md#notarius-reference-bindings).
+
+## Standard Session Workflow
+
+1. Select pipeline/campaign/session config.
+2. Validate session readiness:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+3. (Optional) inspect stage decisions:
+
+```bash
+narratio session plan 2026-04-04
+```
+
+4. Run the pipeline:
+
+```bash
+narratio run 2026-04-04
+```
+
+5. Check state:
+
+```bash
+narratio session status 2026-04-04
+```
+
+Run, plan, and status output identify the resolved pipeline profile (or `none`)
+and effective configuration digest. Status distinguishes the current resolved
+value from the last value persisted in the session manifest, which helps
+diagnose profile switches without changing resume authority.
+
+Before switching an operational profile, compare its effective meaning with the
+production base through `narratio config diff default <profile>`.
+The command is read-only and succeeds whether it finds differences or not. Its
+sorted records describe defaulted, expanded concrete configuration—not source
+file layout—so it can be used to review model, artifact, and publish changes
+without creating a session or run. Use `narratio config artifacts` to review the
+concrete artifact identities, definitions, outputs, and dependencies. Select the
+same campaign explicitly when profiles could resolve different campaign paths;
+see the [CLI reference](cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
+for syntax and record format.
+
+## Stage Execution and Continuation Behavior
+
+Canonical stage order:
+
+1. `prepare`
+2. `transcribe`
+3. `merge`
+4. `polish`
+5. `normalize`
+6. `trim`
+7. `render`
+8. `extract`
+9. `analyze`
+10. `publish`
+11. `notify`
+12. `cleanup`
+
+Execution rules:
+
+- succeeded stages are skipped unless `--force` is set; stages with semantic
+  configuration contracts additionally require matching versioned evidence;
+- `run` continues interrupted or partially completed sessions by running non-succeeded stages;
+- forcing a stage marks succeeded transitive dependents as `stale` before the
+  replacement runs; render and extract are independent siblings; and
+- an executed failure, changed self-skip, or success that replaces a different
+  effective outcome uses the same fixed dependency relation. A
+  repeated self-skip with the same reason and no outputs is stable and does not
+  perpetually rerun dependent work.
+
+Every aggregate stage except analyze currently provides semantic-configuration
+evidence; analyze retains its more precise per-artifact fingerprints and
+validator.
+Prepare additionally validates current stable/local/S3 source identity and the
+checksums of its durable prepared copies before reuse. Changed bytes, audio
+membership, S3 object identity, or missing/tampered copies rerun prepare and
+its fixed descendants without requiring `--force`. Changing prepare selection
+semantics likewise reruns all fixed descendants; changing
+WhisperX language/service identity reuses prepare; changing a Seriatim merge
+transformation reuses prepare and transcribe; and changing an Audita model
+reuses prepare, transcribe, and merge while rebuilding transcript refinement.
+A trim change invalidates both render and extract through the fixed dependency
+relation, while a render-only change preserves the extract sibling.
+
+Operational timeouts, retry/concurrency tuning, executable paths,
+workspace/cache/spool placement, reports, diagnostics, and secret values are
+excluded. Configuration, models, prompts, modules, or resources loaded
+privately inside external tools remain unobservable to Narratio. If their
+contents change behind the same configured identifier, explicitly force the
+affected stage.
+
+An explicit self-skip is a durable `skipped` stage outcome that later runs
+reconsider. It differs from successful no-output execution: disabled `render`
+and `publish`, and absent or no-executable `analyze`, record `succeeded` with
+metadata and no outputs. Ordinary later runs reuse those successful results;
+force the affected stage after enabling or configuring it. Optional artifact
+inputs are omitted only from the consuming artifact invocation and do not make
+the stage self-skip.
+
+Single-stage execution:
+
+```bash
+narratio run-stage normalize 2026-04-04 --force
+```
+
+Contiguous bounded execution uses inclusive canonical endpoints:
+
+```bash
+narratio session plan 2026-04-04 --from extract --through analyze --force
+narratio run 2026-04-04 --from extract --through analyze --force
+```
+
+Omitting `--from` selects from `prepare`; omitting `--through` selects through
+`cleanup`. Force applies only within the selected range. Repeating `--from`,
+`--through`, or `--force` is rejected instead of resolving by argument order.
+The plan command uses the same selection contract and prints only the selected
+range. Planning is read-only: it clones the loaded manifest, models selected
+stage transitions and invalidation in memory, and invokes resume validation
+without writing the manifest, creating run directories, materializing files,
+or invoking pipeline adapters. Analyze detail separates explicit targets,
+prerequisite rebuilds, scheduled execution, and current reuse. This lets a
+coarsely stale aggregate analyze stage show zero artifact executions when its
+selected artifact evidence is still semantically current.
+
+Before a bounded run or plan whose range starts after `prepare`, every excluded
+prefix stage must already have a session-manifest status of `succeeded` or
+`skipped`. Narratio reports the first absent, pending, running, failed, stale,
+or interrupted prerequisite without creating a run-local directory or changing
+session state. Widen `--from` to include that stage, or recover it explicitly
+before retrying. Excluded prefix stages are not resume-validated or repaired as
+part of the bounded invocation; selected stages still reject missing, unsafe,
+or manifest-inconsistent inputs at their owning boundary.
+
+Stages after `--through` are not prerequisites and are never scheduled by the
+bounded invocation. A selected forced stage can mark one of those succeeded
+dependents stale through the fixed invalidation relation, but the dependent
+does not execute until a later invocation selects it. Production composition
+likewise initializes only collaborators needed by the selected range and
+shared session lifecycle. In particular, render does not require Notarius or
+Scriptorium, extract does not require Scriptorium, and analyze does not require
+the transcription, Seriatim, Audita, or Notarius adapters.
+
+For the common post-transcript development loop, use:
+
+```bash
+narratio regenerate-artifacts 2026-04-04
+narratio regenerate-artifacts 2026-04-04 --artifacts session_recap,player_handout
+```
+
+This command is a transparent expansion to a forced bounded `run` from
+`extract` through `analyze`. Extraction always rebuilds its complete configured
+bundle. Analysis rebuilds the selected targets and their required analysis
+prerequisites, or uses the normal default selection when no artifact names are
+given. The command does not run publish, notify, or cleanup; delivery and
+retention remain separate operator actions.
+
+Inspect current artifact evidence, then publish explicitly when the regenerated
+set is ready:
+
+```bash
+narratio session artifacts 2026-04-04
+narratio publish 2026-04-04
+```
+
+If planning or execution reports stale, missing, failed, or tampered
+analysis evidence, regenerate the affected target instead of copying an older
+canonical file into place or editing the manifest. See
+[Troubleshooting: Analysis artifact evidence is not current](./troubleshooting.md#analysis-artifact-evidence-is-not-current).
+
+Analyze checkpoints each completed configured artifact before starting later
+artifact work. After a partial failure or interruption, retry normally: valid
+earlier checkpoints are reused and execution resumes with failed, stale, or
+missing work in dependency order. `--force` is unnecessary unless the operator
+intends to replace otherwise-current artifacts.
+
+## Artifact Selection
+
+For a configured artifact family, selecting its family key expands to every
+concrete character artifact. Select a concrete generated key to operate on one
+member only. Manifests and plan output retain the concrete key as the durable
+identity and include the family and character ID as optional provenance.
+
+`--artifacts` can be used on `run`, `session plan`, `run-stage`, `analyze`, and
+`publish`. For a bounded run or plan, the selected range must contain `analyze`
+or `publish`.
+
+Selection behavior:
+
+- validates names against `pipeline.scriptorium.artifacts`;
+- selects explicit analyze targets and permits their required configured
+  prerequisites to be reused or rebuilt first;
+- filters publish rules for `narratio.artifact.<name>` sources only;
+- does not suppress built-in transcript, bounds, or explicitly configured
+  `narratio.extraction.<name>` publish sources; and
+- never partially selects Notarius lanes.
+
+## Extraction Workflow
+
+When Notarius is omitted or disabled, `extract` records an explicit skipped
+outcome with reason `notarius_disabled` and no outputs. A later invocation
+reconsiders the skipped stage, so enabling Notarius does not require force.
+
+When Notarius extraction is enabled, the stage consumes the final trimmed JSON
+and preserves the complete validated Notarius bundle at:
+
+- `artifacts/notarius/{narratio_run_id}/`
+
+The directory is immutable once promoted. Configured lanes become
+`narratio.extraction.<name>` sources for Scriptorium and explicit publish rules;
+the bundle and `index.json` are retained for audit and resume validation but
+are not selectable or published implicitly.
+
+Configured Notarius references resolve only from the current manifest-backed
+prepared inputs. Their canonical locations are `inputs/party.yml`,
+`inputs/players.yml`, `inputs/glossary.yml`, and, when configured,
+`inputs/spell_catalog.json`. Extraction supplies Notarius with verified copies
+under `runs/<run_id>/extract/references/` so a concurrent refresh of canonical
+prepared files cannot change the bytes consumed by an in-flight invocation.
+For a canonical party, preparation retains the validated authored party bytes
+at `inputs/party.yml` and generates `inputs/players.yml` from that roster.
+The manifest records their checksums separately, with the players input marked
+as derived from the party; refresh preparation after changing the roster rather
+than editing either prepared file.
+Inspect the effective stable-input inventory and
+prepared-file readiness with:
+
+```bash
+narratio session status 2026-04-04
+narratio session validate 2026-04-04
+```
+
+Reference metadata records selector, source ID, session-relative path,
+checksum, and byte size, but never payload contents. Changing a prepared
+reference changes extraction identity: ordinary continuation rejects the old
+result, reruns Notarius, and marks successful downstream stages stale. If the
+prepared file is missing or inconsistent with its manifest checksum, repair
+the source configuration and refresh prepared state first:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Starting a replacement clears the previous extraction payload from the current
+session-stage record. If that replacement fails or self-skips, the current
+record does not fall back to the earlier outputs. The earlier immutable bundle
+remains available for inspection, but downstream resolution requires a new
+current successful extraction record.
+
+Notarius extraction validates its run-local bundle and moves it once to the
+unique durable bundle path. A pre-existing durable destination is a collision;
+no duplicate promotion tree is created.
+
+## External Command Lifecycle
+
+When an external command is cancelled or times out, Narratio terminates its
+owned descendants as well as the command itself. Cancellation first requests
+termination where the platform supports it, then force terminates after a
+bounded wait. A command is not considered finished until its leader has been
+reaped, and descendants that keep standard output or error open cannot keep
+the invocation blocked. Other operating systems fail closed rather than launch
+a command without tree ownership.
+
+Subprocess stdout and stderr diagnostics are separately redacted and capped at
+8 MiB per invocation. Narratio does not retain configured credential values in
+these logs or their error tails; reaching a capture limit terminates the command
+tree and reports which stream exceeded the limit.
+
+Run-local diagnostics are:
+
+- `runs/{run_id}/extract/notarius.receipt.json`
+- `runs/{run_id}/extract/notarius.stderr.log`
+- `runs/{run_id}/extract/notarius-output/` before durable promotion
+
+The run upload is an allowlist derived from the session manifest's declared
+current-run outputs, logs, and generated configurations, not a workspace scan.
+References to canonical session files and prior-run paths are excluded. Each
+declared source is opened without following symlinked ancestors or the leaf,
+verified as a regular file, and streamed from that verified descriptor.
+Unlisted files and unsafe entries are never uploaded; a run with no declared
+current-run diagnostics uploads no run-local files. No distinguished diagnostic
+file is uploaded. The durable bundle is never scanned for implicit publication;
+only lanes named by explicit `pipeline.publish.outputs` rules are uploaded.
+
+To intentionally replace the current extraction result, run:
+
+```bash
+narratio run-stage extract 2026-04-04 --force
+```
+
+Narratio automatically reruns extraction when its recorded invocation contract,
+prepared Narratio reference identities, or durable output validation changes.
+The semantic portion covers Notarius enablement, pipeline identity, declared
+reference mapping, and output contracts. Executable, timeout, working directory,
+and private config-file paths are operational and do not invalidate a current
+result.
+It cannot fingerprint configuration files, profiles, prompts, modules, or
+other references loaded transitively by Notarius itself. Force extraction after
+changing any of those inputs, even when the top-level Narratio and Notarius
+config paths remain the same. A forced extract
+marks successful downstream stages stale. Ordinary extraction failures or
+outcome changes also stale affected downstream stages, while an identical
+repeated `notarius_disabled` self-skip does not repeatedly invalidate them.
+
+Publish reuse additionally tracks enabled/run-upload behavior, normalized
+output rules, and remote backend/bucket/region/endpoint/root identity.
+Credential environment names, local workspace placement, and run IDs are
+excluded. Regardless of semantic reuse evidence, executing publish reads
+independent remote locks directly and rechecks initially unlocked sources
+before commit selection.
+
+## Publish Workflow
+
+Run publish only:
+
+```bash
+narratio publish 2026-04-04
+```
+
+Equivalent:
+
+```bash
+narratio run-stage publish 2026-04-04 --force
+```
+
+This standalone command stops after publish and does not invoke cleanup. A
+default full `run` continues through notify and cleanup.
+
+Publish commit model:
+
+- uploads eligible run files under `{session_prefix}/runs/{run_id}/`, excluding
+  audio and the run-local Notarius staging bundle;
+- uploads configured published outputs and `previous/**` cache files into the
+  same immutable run scope, including only explicitly configured extraction
+  lanes;
+- writes `{session_prefix}/runs/{run_id}/commit.json` after all declared
+  immutable objects are uploaded and verified; and
+- writes `{session_prefix}/current/commit-pointer.json` once, last.
+
+`current/commit-pointer.json` is the remote current-state commit marker. It
+selects exactly one immutable commit, which declares the complete object set.
+
+## Remote Commit Migration
+
+The immutable remote commit contract uses
+`runs/{run_id}/commit.json` to declare a run's complete object set and a small
+`current/commit-pointer.json` to select it. The pointer binds the selected
+commit by version, checksum, size, and storage generation; committed artifacts
+are also checksum- and generation-bound. Readers accept this contract now and
+strictly reject mismatched or unknown data.
+
+A selected commit may carry a `published_output` declaration from an earlier
+run in the same campaign/session. Restore, previous-session caching, status,
+and artifact inspection use that declaration's exact immutable key and stable
+source; they do not reconstruct a current-run key or scan the session prefix.
+Other artifact types remain confined to the selected commit's own run.
+
+When the pointer is absent, remote state is not published; republish the
+session to establish a committed snapshot. Remote commit state does not carry
+local `current_pointer_written` metadata.
+
+## Publish Locks
+
+Each mutable lock is an independent strict JSON object at
+`{session_prefix}/locks/{sha256(source)}.lock`. The lowercase digest is computed
+from the exact canonical source. Listing validates every object in that
+reserved prefix and reports locks sorted by source.
+
+Upgrades do not read, migrate, or delete the retired `{session_prefix}/locks.yml`
+object. It has no effect on publication and may be removed manually after
+verifying lock commands and publication behavior with the upgraded Narratio
+version. No migration command is required.
+
+Operational rules:
+
+- add uses conditional creation and reports an existing lock;
+- add with `--force` observes the current generation and attempts one
+  generation-matched replacement. A concurrent change is reported for an
+  explicit operator retry;
+- remove is idempotent and reports whether a lock was present;
+- the initial publish lock snapshot controls output planning. A locked source
+  carries its exact declaration from the prior current commit when available;
+  a never-published locked source remains absent even when required;
+- carried bytes are not copied. Publish verifies the prior object's exact key,
+  size, and generation without downloading it; and
+- after immutable uploads and the immutable commit, publish re-reads every
+  initially unlocked participating source. A newly added lock prevents the
+  current pointer from advancing, although the immutable attempt may remain.
+  A removed initial lock may conservatively cause one carry.
+
+Examples:
+
+```bash
+narratio session locks 2026-04-04
+narratio session locks add 2026-04-04 narratio.artifact.session_recap --reason "manual edits" --force
+narratio session locks remove 2026-04-04 narratio.artifact.session_recap
+```
+
+## Restore Workflow
+
+Use restore when local durable session state is missing or stale and remote committed current state is authoritative.
+
+Dry run:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Apply:
+
+```bash
+narratio session restore 2026-04-04
+```
+
+`--dry-run` does not write durable session files. It still reads the selected
+remote current state and may read object identity/content needed to classify the
+plan, so it is not a network-free operation.
+
+Default restore scope:
+
+- the committed session manifest and the committed transcript/artifact objects
+  declared by the selected remote commit
+- `previous/**` when needed by configured previous-session artifact inputs;
+  each is selected from its exact committed `published_output` declaration
+
+Optional:
+
+- `--include-audio` to include `audio/**`
+- `--force` to overwrite eligible conflicting regular files; it never replaces
+  directories, symlinks, or other non-regular local targets
+
+Restore treats a local leaf symlink or any symlinked ancestor beneath the
+selected session root as a conflict, even when the linked content matches the
+selected remote object. `--force` does not override that conflict.
+
+Before each install, restore validates the destination again beneath the
+selected session root. A target changed into a symlink, directory, or other
+non-regular entry is not overwritten.
+
+Restore writes an execution report at `reports/restore-latest.json`.
+
+If restore fails after beginning installation, it leaves a durable
+`.restore-incomplete.json` marker in the session root. Pipeline runs will stop
+until you rerun the same restore command and it completes. Restore intentionally
+does not try to roll back files already installed; retrying the selected remote
+snapshot is the recovery procedure.
+
+## Local State Layout
+
+Session root:
+
+- `{workspace.root}/work/{campaign}/{session_id}`
+
+Durable session paths:
+
+- `manifest.json`
+- `inputs/**`
+- `audio/**`
+- `transcripts/**`
+- `artifacts/**`
+- `previous/**`
+- `reports/**`
+- `logs/**`
+- `config/**`
+- `runs/**`
+
+`manifest.json` is the versioned session-progress document. Narratio accepts
+only `narratio.session-manifest.v1`. Do not edit unsupported session state to
+add missing markers. Move it aside and regenerate it, restore a verified
+current committed snapshot, or republish healthy regenerated state as the
+relevant workflow requires.
+
+Validated Notarius bundles live below `artifacts/notarius/{run_id}/`; receipt,
+stderr, and pre-promotion output remain in the producing run's `extract`
+directory as described in [Extraction Workflow](#extraction-workflow).
+
+Run-local layout:
+
+- `runs/{run_id}/{stage}/outputs`
+- `runs/{run_id}/{stage}/logs`
+- `runs/{run_id}/{stage}/reports`
+- `runs/{run_id}/{stage}/config`
+- `runs/{run_id}/{stage}/scratch`
+
+Spool layout (runtime/transient):
+
+- `{spool.root}/{campaign}/{session_id}/{run_id}/...`
+- restore audio spool under `{spool.root}/{campaign}/{session_id}/restore/audio`
+
+Cache layout (durable S3 audio cache):
+
+- `{cache.root}/s3/{bucket}/...`
+
+Each cached audio file has an adjacent managed identity record. It binds the
+file to its remote object version and verified digest; deleting or altering the
+record simply causes Narratio to download and verify the object again.
+Remote audio and restored files are fully transferred into sibling temporary
+files and validated before they replace their canonical workspace paths.
+
+### Workspace Permissions
+
+Ordinary Narratio workspace content is intentionally shareable with the
+workspace group. On POSIX systems, Narratio-created workspace, spool, and cache
+directories converge on setgid `02775`; ordinary files, including manifests,
+transcripts, generated configuration, logs, reports, and Notarius artifacts,
+converge on `0664`. Narratio explicitly applies these modes so a restrictive
+caller umask does not remove group write or setgid. It does not change file or
+directory ownership: the configured workspace's existing group is inherited.
+
+API keys are credentials, not ordinary workspace data. Store them outside the
+shared workspace or in a separately restricted credential location; ordinary
+workspace group access must never be treated as authorization to read keys.
+Provision a credential directory as `0700` and credential files as `0600`;
+these permission modes are an operator requirement and are not enforced by
+ConfigKit.
+
+Every pipeline-loading command bootstraps credentials once, before the first
+configuration load and before any adapter is constructed. `--secrets-dir`
+selects an explicit directory (relative paths use caller CWD). Otherwise a
+present `secrets/` beside the resolved root pipeline is used, while an absent
+conventional directory is skipped. Explicit missing/empty selections and
+invalid existing directories fail the command. ConfigKit follows a symlink used
+as the selected directory, but immediate entries must be regular non-symlink
+files. Existing environment variables—including empty values—always win.
+
+S3 `access_key_id` and `secret_access_key` are an optional pair of typed
+environment references. When both are absent, the AWS SDK retains its default
+credential provider chain. When both are configured, Narratio supplies their
+load-time resolved values through an explicit static provider. Audita's optional
+`llm_api_key` is likewise captured at pipeline load and supplied only through
+the child process's `AUDITA_LLM_API_KEY` override.
+
+External adapter results are individually bounded before Narratio validates or
+materializes them. These per-file limits do not reserve disk space: prevent hard
+disk exhaustion with filesystem, service, container, or volume quotas sized for
+the session workload.
+
+## Cleanup
+
+Session-scoped cleanup:
+
+```bash
+narratio clean 2026-04-04
+```
+
+Global cleanup:
+
+```bash
+narratio clean --all
+```
+
+Dry-run and cache variants:
+
+```bash
+narratio clean 2026-04-04 --dry-run --clear-cache
+narratio clean --all --dry-run --clear-cache
+```
+
+Rules:
+
+- `clean` deletes work/spool session state;
+- cache is preserved unless `--clear-cache` is set;
+- each deletion is confined beneath its configured workspace, spool, or cache
+  root; root, traversal, and out-of-root targets are refused;
+- the final cleanup stage is gated by successful publish evidence plus:
+  - `pipeline.spool.delete_audio_after_publish=true`
+  - `pipeline.workspace.cleanup_after_publish=true`
+  - the manifest publish record has `uploaded=true`, the published run ID, the
+    exact remote commit key, and the current commit-pointer key.
+
+Cleanup derives only the configured run-scoped spool and workspace targets,
+uses root-confined deletion, and does not read or mutate object storage. A
+failure leaves publish succeeded and the remote committed snapshot current.
+Retry with a normal full run: valid upstream work, including publish and notify,
+is skipped while the failed cleanup stage runs again. `run-stage cleanup` is
+also available when the manifest has valid publish evidence. A range ending at
+`notify` or earlier and the standalone `publish` command never delete local
+state.
+
+Older local manifests that contain the retired cleanup-obligation field are
+intentionally incompatible with strict manifest decoding. If such an
+obligation is still outstanding, complete it with the pre-upgrade binary.
+Otherwise verify that the remote publication is committed, move aside or
+regenerate the incompatible local manifest (or restore the verified remote
+snapshot), and then use the cleanup stage. Manual JSON editing is not the
+normal migration path.
+
+## Operational Caveats
+
+- Local and S3 audio modes are mutually exclusive.
+- Publish requires prerequisite stages through `render` and `analyze` to be succeeded.
+- Markdown publish defaults require render outputs (`transcripts/final.md` and `transcripts/final.trimmed.md`).
+- Restore requires configured object storage and committed remote current state.
+- Pipeline-loading commands bootstrap filesystem secrets once before configuration
+  resolution and object-store initialization.

docs/policy/architecture.md

@@ -0,0 +1,305 @@
+# Architecture
+
+This document defines Narratio's intended high-level architecture and the
+invariants that changes must preserve. Implemented component details belong in
+the [Internal Overview](../internal/overview.md) and its linked documents.
+Significant architectural decision history belongs under `docs/adr/` when such
+records exist.
+
+## System Shape
+
+Narratio is a small Go application that turns D&D session audio into polished
+transcripts and generated session artifacts. It is an explicit, stage-driven
+orchestrator, not a general workflow engine.
+
+Narratio coordinates specialized external systems rather than reimplementing
+their domains:
+
+- WhisperX performs transcription;
+- Seriatim performs deterministic transcript processing and rendering;
+- Audita performs transcript correction and polishing;
+- Notarius extracts validated structured artifact bundles; and
+- Scriptorium executes prompts and produces configured artifacts.
+
+Narratio owns orchestration, configuration resolution, session and run state,
+artifact and path modeling, manifest persistence, stage sequencing, resume,
+restore, cleanup gates, and publish semantics. External contracts are defined
+in the [integration documentation](../integrations/).
+
+The pipeline has one canonical ordered stage set. Configuration may enable,
+disable, or parameterize supported behavior, but it must not turn that sequence
+into an arbitrary DAG or hide orchestration in generic workflow abstractions.
+An invocation selects either the full sequence or one inclusive contiguous
+range of it. Execution remains flat and canonical even though invalidation is
+dependency-aware: the application owns a separate fixed relation used only to
+stale transitive dependents, including dependents outside a selected range.
+The implemented stage inventory belongs in the
+[Internal Overview](../internal/overview.md).
+
+Narratio is contract-first without being abstraction-heavy. Interfaces and
+extension points should protect demonstrated boundaries. New abstraction is not
+itself an architectural goal.
+
+## Ownership And Dependency Direction
+
+The application boundary owns command dispatch, configuration selection,
+production composition, session locking, and top-level lifecycle. It may depend
+on concrete implementations to assemble a run.
+
+Stage orchestration expresses intent in Narratio-level data and interfaces.
+Stages may depend on configuration, manifest, artifact, path, and adapter
+contracts, but they must not depend on transport-specific request types,
+subprocess argument construction, cloud SDK types, or downstream tool internals.
+
+Adapters translate between Narratio contracts and external systems. They own
+HTTP, subprocess, notification, and object-storage mechanics, including command
+construction, transport behavior, provider response handling, and external
+error adaptation. External dependency types must remain inside the adapter that
+owns them unless that dependency is the adapter's explicit public contract.
+WhisperX HTTP behavior, Seriatim, Audita, Notarius, and Scriptorium command
+construction, notification transport, and object-storage SDK details remain
+behind these boundaries.
+
+State and path services must not infer stage policy. Storage implementations
+receive explicit bucket-relative keys and do not infer campaign, session, run,
+or root-prefix semantics. Manifest persistence records transitions but does not
+choose orchestration policy. Artifact resolution identifies and validates
+artifacts but does not execute producers.
+
+Dependencies should remain narrow and point toward Narratio-owned contracts.
+Prefer the Go standard library. Add an external dependency only when it provides
+a clear correctness, security, interoperability, or complexity benefit, and
+confine it to the boundary that needs it.
+
+## Stage Boundaries
+
+Each stage has one explicit responsibility and declares:
+
+- required input state;
+- produced output state;
+- configuration it consumes;
+- external adapters it uses;
+- manifest references and metadata it reads or writes;
+- skip, force, invalidation, and resume behavior; and
+- failure behavior.
+
+Stages write and validate run-local results before materializing canonical
+outputs where that distinction applies. A stage is complete only after its
+required outputs have been written, validated, and recorded in durable manifest
+state. Later stages depend on recorded success and artifact resolution, not
+merely on incidental files existing on disk.
+
+A failed or interrupted stage must not be presented as successful. Failure
+should preserve enough local state and diagnostics for inspection, recovery,
+and resume. Forcing a stage invalidates succeeded transitive dependents
+according to a fixed application-owned relation that is separate from canonical
+execution order. The relation is validated against the stage inventory and is
+not configurable.
+
+A stage may explicitly self-skip with a stable reason and no outputs. That
+outcome is persisted, clears older outputs owned by the stage, and is
+reconsidered on a later invocation. A stage may also validate whether an
+otherwise successful recorded result is still resumable; an obsolete result
+is staled and rerun, while an unsafe condition that prevents a sound decision
+stops execution.
+
+Shared behavior should live behind a narrow service or helper with one clear
+owner. Stages must not reach across boundaries or reproduce adapter, manifest,
+artifact, or path policy ad hoc.
+
+## Manifest, Resume, And Restore
+
+The session manifest is the durable ledger for progress across invocations. It
+records session and run identity, stage state, input and output references,
+diagnostic references, checksums or provenance where useful, and non-secret
+adapter and publish metadata.
+
+Resume and skip decisions are manifest-driven. Filesystem state may be
+inspected and validated, but file presence alone does not replace recorded
+stage state. Each invocation retains an in-memory command summary and a
+run-local workspace identity, but neither is a durable progress ledger. The
+physical run directory is created on demand only when an executing stage
+requires it; a fully reused invocation or a cleanup-only invocation does not
+leave an empty run directory.
+
+Restore treats committed remote current state as its authority. It must plan
+deterministically, confine remote-to-local paths, protect local conflicts, and
+install the validated session manifest after other restored durable files. The
+physical workflow and recovery procedures belong in
+[Operations](../operations.md).
+
+Restore and runner transitions for one session use the same local lock. A
+durable incomplete-restore marker blocks runner reuse after a partial restore;
+safe retry, rather than rollback of arbitrary local effects, is the recovery
+mechanism. Restored manifest-local references must be confined to the selected
+local session root, never trusted as producer-machine absolute paths.
+
+For the immutable remote-commit protocol, a restore or status operation binds
+to one pointer-selected commit and only its declared object identities. A force
+flag may replace an eligible regular managed file, but never turns a directory
+or other non-regular conflict into a successful restore.
+
+## Configuration
+
+Configuration is strict, explicit, centralized, and operator-oriented.
+
+- YAML decoding rejects unknown fields.
+- Defaults are centralized and testable.
+- Empty configured values do not silently replace meaningful defaults.
+- Validation rejects invalid configuration before stage execution where
+  practical.
+- Root-owned imports and one selected profile resolve deterministically through
+  the configuration owner; commands do not implement their own merge rules.
+- Canonical party rosters are campaign-owned. Their derived players projection
+  and concrete character-family artifacts are resolved before runtime stages
+  or adapters receive configuration.
+- Resume uses stage- or artifact-owned semantic evidence for observable
+  result-affecting configuration; profile identity and an effective digest are
+  provenance, never blanket cache keys.
+- Session templating remains narrow and deterministic rather than becoming a
+  general configuration language.
+- Secret values are supplied indirectly and are not persisted in ordinary
+  configuration.
+
+Narratio must not become a second configuration system for downstream tools.
+External systems own their runtime defaults wherever practical; Narratio passes
+the paths required by its stage contracts and explicit operator overrides. The
+field-level contract and credential-supply mechanisms belong in
+[Configuration](../config.md).
+
+## Artifacts, Paths, And Storage
+
+Artifact identities and local and remote paths are application contracts.
+Canonical helpers own workspace, spool, cache, session, run, input, transcript,
+artifact, log, report, configuration, and publish-current paths. Callers must
+not reconstruct canonical paths through scattered string concatenation.
+
+Reusable audio cache entries require a typed record that binds a confined,
+no-follow regular file and its digest to the selected remote object identity.
+Size alone and unqualified multipart ETags are not content-integrity evidence.
+
+Artifact resolution is deterministic and manifest-aware. Producers materialize
+canonical outputs before reporting success, and consumers resolve declared
+artifact identities rather than infer files from unrelated directory contents.
+External artifact bundles become current only through validated immutable
+promotion and manifest records; directory presence alone never establishes
+availability.
+
+Writes, moves, replacements, and deletions must use narrow, explicit,
+root-confined destinations. Symlinks, traversal, broad roots, and ambiguous
+relative destinations must not expand the scope of an operation. Cleanup is
+permitted only through explicit operator action or the configured final cleanup
+stage, and it must preserve durable cache unless cache removal is explicitly
+requested.
+
+Physical layout, retention, and operational lifecycle belong in
+[Operations](../operations.md). Logical external formats and durable integration
+contracts belong under [Integrations](../integrations/).
+
+## Publish Commit Boundary
+
+Publish has one explicit remote commit boundary. A remote run becomes current
+only after Narratio has successfully uploaded its immutable run-scoped objects,
+the immutable commit manifest, and finally the current commit pointer.
+
+`current/commit-pointer.json` is the sole mutable selector and must be written
+exactly once, last. Failed, incomplete, skipped, or uncommitted publish attempts
+must not be presented as current remote state. Publish locks remain authoritative
+and are not bypassed by a forced run. Each mutable remote lock is an independent
+per-source object; lock mutations use provider-enforced generation preconditions.
+Publish revalidates every initially unlocked participating source after writing
+the immutable commit and before pointer selection. A lock observed then leaves
+the prior committed snapshot current. A lock created after that final check may
+race the pointer write; the commit protocol does not claim a stronger distributed
+transaction boundary.
+
+Automatic local cleanup is permitted only after a successful publish commit,
+only when explicitly configured, and only through the path-safety guardrails.
+It is the final ordinary pipeline stage after `notify`. The stage derives its
+exact run-scoped targets from validated publish metadata and the configured
+campaign/session identity; it never infers a target from directory contents or
+rereads remote storage. A cleanup failure makes the invocation incomplete while
+leaving publish succeeded and the committed remote snapshot authoritative. A
+normal later full run reuses valid upstream stages and retries cleanup.
+
+The standalone `publish` command selects only publish and therefore never
+invokes cleanup. Operators may select cleanup directly only when the manifest
+contains valid successful publication evidence.
+
+## Security, Privacy, And Diagnostics
+
+Narratio distinguishes ordinary workspace data from credentials. Campaign and
+session material—including manifests, transcripts, prompts, generated
+configuration, logs, reports, diagnostics, and Notarius artifacts—is
+intentionally shareable with the configured workspace group. API-key material
+is sensitive and is not covered by the ordinary workspace-sharing policy.
+
+Narratio supports Linux on amd64 and arm64. macOS arm64 is best effort: it has
+a cross-build gate and release asset, but no native runtime release gate.
+Windows and macOS amd64 are unsupported and have no compatibility promise or
+release asset.
+
+On supported systems, Narratio-created ordinary workspace directories converge on
+setgid `02775` and ordinary workspace files on `0664`, even when the caller's
+umask is restrictive. This preserves the existing workspace group for nested
+creation and atomic replacements without changing ownership. API-key storage
+uses a separate restrictive contract.
+
+Raw secrets must not be stored in pipeline, campaign, or session YAML or written
+to manifests, logs, generated configuration, reports, publish metadata,
+documentation, or examples. Pipeline configuration names environment variables;
+command-owned startup may populate previously unset variables from ConfigKit's
+explicit or conventional secrets directory. This bootstrap occurs once per
+invocation before any typed configuration load or adapter construction and is
+not hidden inside library loaders. Safe bootstrap diagnostics may identify
+files and statuses but never values. Diagnostics should avoid transcript and
+prompt content unless a deliberate, bounded inspection mechanism requires it.
+
+Logs, reports, generated invocation files, generated configuration, and render
+debug files are diagnostics, not canonical pipeline products. They should be
+durable and discoverable where configured, and manifest references must preserve
+the distinction between diagnostics and artifacts.
+
+Documentation security rules belong in the
+[Documentation Policy](documentation.md). Credential supply belongs in
+[Configuration](../config.md), while permissions, sensitive runtime-artifact
+handling, and recovery belong in [Operations](../operations.md).
+
+## Determinism And Testability
+
+Narratio prefers deterministic behavior where practical, including stable local
+and remote layouts, sorted operation order, predictable generated
+configuration, repeatable command construction, deterministic artifact
+resolution, and reproducible planning.
+
+Run IDs and timestamps may be intentionally variable, but surrounding behavior
+must remain controllable in tests. Core behavior should be testable without live
+external services; expensive, nondeterministic, destructive, or external
+boundaries should be replaceable with focused test doubles. General testing
+philosophy and sufficiency rules belong in the [Testing Policy](testing.md).
+
+## Documentation And Decision Records
+
+Documentation follows the [Documentation Policy](documentation.md). Current
+behavior belongs in its canonical user, operator, integration, architecture, or
+internal owner. Proposed behavior and implementation status belong under
+`docs/roadmap/`.
+
+Significant architectural decisions may be recorded under `docs/adr/` using the
+format and lifecycle defined by the documentation policy. ADR acceptance does
+not establish that a decision has been implemented.
+
+## Architectural Non-Goals
+
+Narratio does not aim to provide:
+
+- a generic DAG or workflow engine;
+- a replacement configuration layer for WhisperX, Seriatim, Audita,
+  Scriptorium, or other downstream tools;
+- a storage abstraction broader than the needs of this pipeline;
+- stage logic coupled directly to cloud SDKs, transports, subprocess details,
+  or downstream implementation internals;
+- raw-secret persistence;
+- implicit cross-stage behavior that bypasses manifest and artifact contracts;
+  or
+- a prompt-authoring system.

docs/policy/documentation.md

@@ -0,0 +1,150 @@
+# Documentation Policy
+
+## Purpose
+
+This policy assigns each documentation topic to one canonical owner. Its goal is
+to keep Narratio documentation accurate, concise, discoverable, and resistant
+to drift for users, operators, developers, integrators, and LLM coding agents.
+
+## Core Rules
+
+### One Canonical Owner
+
+Each authoritative fact belongs in one document. A non-owning document may give
+a short, stable summary for orientation, but it must link to the canonical owner
+instead of repeating volatile details.
+
+Volatile details include commands, flags, configuration fields and defaults,
+stage or integration keys, schemas, file names, paths, status codes, retry
+behavior, and runtime guarantees. If readers could reasonably treat a statement
+as a contract, maintain it only in the owning document.
+
+### Current And Future Behavior
+
+Outside `docs/roadmap/`, documentation describes implemented behavior only.
+Partial features may be described only to their implemented boundary.
+
+ADRs are the narrow exception: an ADR may record an accepted architectural
+decision before implementation, but acceptance must not be presented as proof
+that the behavior exists. The roadmap owns implementation status and sequencing
+until the decision is implemented. Current architecture, user, operator,
+integration, and internal documentation are updated when the behavior lands.
+
+### Audience And Detail
+
+Write for the document's stated audience and include only the detail needed for
+its owned topic. User and operator docs should not expose implementation detail.
+Developer docs should link to user-facing and external contracts rather than
+restate them.
+
+### Examples
+
+Complete copyable files belong in `examples/`. Documentation may use the
+smallest illustrative snippet needed to explain its owned topic, but should link
+to maintained examples instead of embedding a second complete copy.
+
+Examples must be valid, secret-free, and tested where practical. Commands and
+configuration used in documentation should match the application.
+
+### Security And Privacy
+
+Documentation and examples must not contain real credentials, private keys,
+private environment dumps, sensitive source material, or private infrastructure
+details unless intentionally public. Document secret-handling mechanisms, not
+secret values.
+
+## Canonical Ownership
+
+| Topic | Canonical owner | Owned content | Content owned elsewhere |
+| --- | --- | --- | --- |
+| Product orientation and minimal end-to-end quickstart | `README.md` | What Narratio is, why it is useful, one shortest successful invocation, and links onward. | Complete command reference, configuration reference, operational procedures, implementation detail. |
+| Contributor entry point | `docs/development.md` | Task-oriented reading guide, minimal contributor orientation, baseline validation commands, and links to canonical docs. | Package inventory, architecture rules, subsystem behavior, detailed change recipes. |
+| Maintainer release procedure | `docs/release.md` | Version selection, candidate preparation and validation, guarded tag publication, release completion boundary, failure recovery, and optional asynchronous inspection. | Script implementation mechanics, current application contracts, and historical release summaries. |
+| Historical release summary | `docs/releases/<tag>.md` | Immutable summary, compatibility, upgrade, and changes for one released version. | Current maintainer procedure and current application contract details. |
+| Current application architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
+| Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression-test policy, and criteria for adding, rewriting, or deleting tests. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
+| CLI contract | `docs/cli.md` | Commands, arguments, flags, invocation semantics, output conventions, and exit behavior. | End-to-end operating procedures, configuration field definitions, runtime filesystem layout, stage implementation details. |
+| Configuration contract | `docs/config.md` | Discovery and precedence, file schemas, fields, defaults, environment overrides, validation rules, and user-selectable stage or integration settings. | Complete example files, CLI syntax, runtime state lifecycle, implementation details. |
+| Operations | `docs/operations.md` | Runtime workflows, physical filesystem and remote-state layout, output and diagnostic handling, resume, cleanup, permissions, recovery, and operational limits. | CLI flag syntax, configuration field definitions, logical artifact schemas, implementation mechanics. |
+| Troubleshooting | `docs/troubleshooting.md` | Symptom-driven diagnosis, likely causes, safe inspection steps and remedies, and links to relevant contracts. | CLI syntax, configuration definitions, operational procedures, integration contracts, implementation mechanics. |
+| Public HTTP contract, if introduced | `docs/api.md` | Routes, authentication, media types, request and response schemas, status codes, pagination, caching, idempotency, rate limits, and HTTP retry semantics. | Client walkthroughs, upstream or downstream integration internals, implementation detail. |
+| Consumer guidance, if a public package or API is introduced | `docs/consumers/` | Task-oriented use of the public interface, minimal client examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| External and durable integration contracts | `docs/integrations/` | External file formats and protocols, upstream and downstream contracts, logical artifact paths and schemas, media types, and compatibility behavior. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
+| Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
+| Internal component behavior | Other files under `docs/internal/` | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
+| Architectural decision history | `docs/adr/` | Significant decisions, context, alternatives, rationale, consequences, and supersession history. | Current behavior reference, implementation status, task sequencing. |
+| Future work and implementation status | `docs/roadmap/` | Proposed, accepted, deferred, or rejected work; implementation status; sequencing; and task breakdowns. | Implemented behavior reference and architectural decision rationale. |
+| Complete copyable artifacts | `examples/` | Maintained configuration, inputs, and other files intended to be copied or run. | Field-by-field reference, command reference, prose explanation. |
+
+Documents that do not exist are required only when the corresponding interface
+or responsibility exists. Do not create placeholder API, consumer, integration,
+or operations documents for behavior the application does not have.
+
+## Boundary Rules
+
+### Orientation
+
+The README owns product orientation. The developer guide routes contributors.
+Architecture owns normative structure. Internal overview owns the current
+concrete component map. These documents may link to one another but should not
+maintain parallel package or behavior descriptions.
+
+### Commands, Configuration, Operations, And Troubleshooting
+
+CLI documentation answers how to invoke the application. Configuration
+documentation answers what settings mean. Operations answers what happens to
+runtime state and how to operate or recover the application. Troubleshooting
+starts from observable symptoms and links readers to the owning command,
+configuration, operational, or integration contract. When a workflow crosses
+these topics, choose the document that owns the task and link to the other
+contracts.
+
+### Contracts And Implementation
+
+Integration and API documents define externally observable shapes and
+semantics. Internal documents explain how Narratio implements or consumes those
+contracts. Internal docs may name a field, file, or protocol to identify a
+dependency, but must link to its canonical contract for the definition.
+
+### Security Topics
+
+This policy owns what documentation and examples may contain. Architecture owns
+application security invariants. Configuration owns credential-supply
+mechanisms. Operations owns permissions and handling of sensitive runtime
+artifacts. Troubleshooting owns safe diagnostic and remediation guidance.
+Internal docs own implementation mechanisms only.
+
+## Architecture Decision Records
+
+Use sequentially numbered ADR filenames such as
+`0001-record-architecture-decisions.md`. Follow the lightweight Nygard format:
+
+1. title;
+2. status;
+3. date;
+4. context;
+5. decision;
+6. alternatives considered;
+7. consequences.
+
+Treat the decision content of an accepted ADR as immutable. When a decision
+changes, create a new ADR and update the earlier ADR's status to superseded.
+Rejected architectural alternatives belong in the ADR; rejected product ideas
+belong in the roadmap.
+
+## Maintenance
+
+When behavior changes, update its canonical owner in the same change. If
+ownership moves, remove the old definition and replace it with a link where
+navigation remains useful.
+
+Before completing documentation work:
+
+- verify affected behavior and examples;
+- check commands, flags, fields, defaults, schemas, and paths against their
+  implementation;
+- keep unimplemented behavior in the roadmap, subject to the ADR exception;
+- remove stale references and validate links;
+- confirm that non-owning documents summarize and link rather than redefine;
+- confirm that no secrets or sensitive private data were added.

docs/policy/testing.md

@@ -0,0 +1,296 @@
+# Testing Policy
+
+## Purpose
+
+Our tests exist to make **incorrect changes expensive and correct changes cheap**.
+
+We do not optimize for test count, line coverage, exhaustive isolation, or the fewest possible tests. We optimize for sufficient confidence in important behavior while imposing as little unnecessary friction as possible on future development.
+
+## Every test has a cost
+
+Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
+
+A test must be:
+
+- written and reviewed;
+- understood by future maintainers and coding agents;
+- executed in local and CI workflows;
+- diagnosed when it fails;
+- updated when legitimate behavior changes;
+- maintained as fixtures, APIs, and dependencies evolve; and
+- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
+
+Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+
+A test is warranted only when the confidence it provides justifies these costs.
+
+Apply this cost-benefit analysis at two levels:
+
+1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
+2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
+
+The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
+
+Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
+
+The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
+
+## Default testing style
+
+Use a **classical/Detroit-style** approach:
+
+- Test observable behavior, resulting state, contracts, and invariants.
+- Use real internal collaborators when they are fast and deterministic.
+- Use fakes, stubs, or mocks primarily at expensive, nondeterministic, destructive, or external boundaries.
+- Prefer package-level behavioral tests over tests coupled to private helpers or internal call sequences.
+- Treat exact collaborator interactions as testable behavior only when the interaction itself is a requirement.
+
+Examples of appropriate seams include clocks, randomness, subprocesses, remote APIs, object storage, email, and paid LLM calls.
+
+## Test execution requirements
+
+Tests in the default suite must be deterministic, offline, and independent of real credentials. They must not invoke paid APIs or depend on mutable external services. Tests that require live infrastructure must be explicitly opt-in and clearly separated from the default suite.
+
+Control clocks, randomness, environment variables, and other process-global or machine-specific state when they affect behavior. Tests should be safe to run repeatedly and alongside other tests without depending on execution order or state left by an earlier test.
+
+## What deserves tests
+
+Prioritize tests for:
+
+1. Public and package-level contracts.
+2. Domain rules and important invariants.
+3. Boundary conditions and malformed input.
+4. Failure handling, cancellation, retries, recovery, and partial success.
+5. Serialization, schemas, compatibility, and round trips.
+6. Previously observed or plausible regressions.
+7. Representative integration and end-to-end workflows.
+
+A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
+
+For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+
+Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
+
+## Choose the right test boundary
+
+Test through the narrowest stable boundary that expresses the behavior clearly.
+
+This is often the package API, but it may instead be:
+
+- a smaller pure function when dense domain logic is most clearly isolated there;
+- a package-level operation when several internal collaborators jointly produce the behavior; or
+- a larger integration boundary when correctness emerges from interaction with a real dependency.
+
+Do not force all behavior through oversized end-to-end tests. Do not test every private helper merely because it exists. Choose the boundary that gives durable confidence with the least incidental coupling.
+
+## Test behavior, not implementation
+
+A test should protect a decision, contract, or invariant—not memorialize the current implementation.
+
+Before adding or retaining a test, ask:
+
+> What realistic defect would this test catch?
+
+A test is suspect when its main purpose is to detect that someone:
+
+- changed an internal constant;
+- renamed or split a private helper;
+- reordered equivalent internal operations;
+- changed incidental formatting;
+- replaced one correct algorithm with another; or
+- refactored internal object structure without changing behavior.
+
+Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
+
+A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
+
+## Expected effects of different changes
+
+Use the following expectations when evaluating test failures and test maintenance:
+
+| Change | Expected effect on tests |
+|---|---|
+| Internal refactor that preserves behavior | Existing tests should normally remain unchanged and continue to pass. |
+| Change to an internal default with no contractual significance | Behavioral tests should normally remain unchanged; tests should derive expectations from configuration or relationships rather than duplicate the old value. |
+| Intentional change to public behavior, policy, schema, or compatibility guarantees | The relevant tests should be reviewed and changed deliberately. |
+| Accidental violation of a contract or invariant | Tests should fail; fix the production code rather than rewriting the tests to accept the defect. |
+
+A test failing is not the same as a test needing to be edited. Many tests may correctly fail because of one production defect. The maintenance smell is a correct internal change that requires unrelated expectation updates throughout the suite.
+
+## Separate mechanism from policy
+
+Configurable thresholds and defaults must not be duplicated throughout the test suite.
+
+For example, do not encode an internal concurrency limit indirectly:
+
+```go
+// Production policy:
+const maxConcurrency = 4
+
+// Brittle test:
+err := startProcesses(5)
+require.Error(t, err)
+```
+
+Instead, test the mechanism relationally:
+
+```go
+const limit = 2
+runner := NewRunner(limit)
+
+require.NoError(t, runner.Start(limit))
+require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
+```
+
+The test should prove:
+
+- the configured limit is accepted; and
+- one beyond the configured limit is rejected.
+
+The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
+
+Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
+
+For concurrency limits, test both kinds of behavior when relevant:
+
+1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
+
+Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
+
+## Avoid semantic duplication across layers
+
+Each behavior should have a clear test owner.
+
+- Parser tests own parsing cases.
+- Validator tests own validation rules.
+- Domain tests own transformations and invariants.
+- Adapter tests own external integration behavior.
+- Orchestrator tests own coordination and failure propagation.
+- CLI tests own argument and configuration mapping.
+- End-to-end tests prove that representative assembled workflows work.
+
+Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
+
+Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
+
+## Use test doubles deliberately
+
+Choose the least elaborate test double that provides the required control or observation.
+
+As a default:
+
+1. Prefer real collaborators when they are fast and deterministic.
+2. Use small in-memory fakes when realistic stateful behavior is helpful.
+3. Use stubs when a dependency only needs to provide controlled responses.
+4. Use mocks when the interaction itself is contractual.
+
+Mocks are appropriate when the contract includes facts such as:
+
+- a notification is sent exactly once;
+- a transaction is committed only after successful writes;
+- cancellation reaches a subprocess;
+- an expensive API is called no more than once; or
+- a security audit event is emitted.
+
+Do not use mocks merely to isolate every object or reproduce the implementation's call graph.
+
+## Go-specific guidance
+
+Use:
+
+- table-driven tests for meaningful behavioral categories and boundaries;
+- `t.TempDir()` for real filesystem behavior;
+- `httptest.Server` for realistic HTTP interactions;
+- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- golden files only when the complete output is intentionally stable;
+- integration tests where correctness depends on component interaction; and
+- a small number of representative end-to-end tests.
+
+Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
+
+At CLI boundaries, prefer exit classifications, structured output, and the smallest stable semantic fragment needed to identify the error. Do not snapshot complete diagnostic wording unless it is contractual.
+
+Golden-file updates must require an explicit local flag. CI must not update golden files automatically, and reviewers must inspect the semantic diff before accepting an update.
+
+Keep tests readable and direct. Test helpers and fixture frameworks must earn their own maintenance cost; do not build elaborate test infrastructure for small or isolated needs.
+
+## Coverage
+
+Coverage is a diagnostic, not a target.
+
+Use it to find untested critical branches and unexpectedly weak packages. Do not write low-value tests solely to increase a percentage, and do not infer test quality from coverage alone.
+
+Pure domain logic will often warrant higher coverage than CLI wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
+
+Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
+
+## Regression tests
+
+A bug fix should normally include a regression test that fails before the fix and passes afterward.
+
+Retain the test when the defect could realistically recur and its consequences justify the ongoing cost. Prefer the narrowest durable test of the violated contract or invariant; do not preserve accidental implementation details from the original bug.
+
+Not every historical bug requires a permanent test. If the underlying design has made recurrence impossible, the test has become redundant, or a stronger invariant test now subsumes it, remove or consolidate it.
+
+## Deleting or rewriting tests
+
+Tests are maintained code, not permanent historical artifacts.
+
+Delete or rewrite a test when its maintenance cost exceeds the confidence it provides.
+
+Strong candidates include tests that:
+
+- require updates after harmless internal changes;
+- directly assert private constants without protecting a real contract;
+- duplicate the same policy across several layers;
+- verify mock choreography rather than outcomes;
+- snapshot large amounts of incidental output;
+- test trivial private helpers already exercised through stable package behavior;
+- protect risks already covered more effectively elsewhere;
+- are flaky, misleading, obsolete, or disproportionately expensive to diagnose; or
+- no longer correspond to a plausible failure mode.
+
+Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
+
+Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
+
+## Reviewing a proposed test
+
+Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
+
+1. What realistic defect would it catch?
+2. How likely is that defect?
+3. How consequential would it be?
+4. Is the behavior already protected elsewhere?
+5. At which layer should this behavior be owned?
+6. Does the test assert a durable contract or an incidental implementation detail?
+7. Could the implementation be refactored without changing the behavior and without editing this test?
+8. What should cause this test to fail?
+9. What legitimate changes should not cause this test to fail?
+10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
+11. Is there a smaller or more direct test that protects the same risk?
+
+Do not add the test when its expected lifetime cost exceeds its expected protective value.
+
+When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+
+## Definition of sufficient
+
+A test suite is sufficient when:
+
+- important contracts and invariants are protected;
+- meaningful boundaries and failure modes are exercised;
+- realistic and consequential regressions are credibly protected against silent recurrence;
+- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
+- important external boundaries have realistic integration coverage;
+- representative complete workflows are tested;
+- failures provide useful signal rather than redundant noise;
+- legitimate internal changes usually do not require test edits; and
+- additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
+
+Sufficiency is a risk judgment, not a coverage percentage or test count. Reassess it as the application, its users, and the consequences of failure evolve.
+
+The governing rule is:
+
+> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.

docs/release.md

@@ -0,0 +1,104 @@
+# Releasing Narratio
+
+This document is the maintainer procedure for creating a Narratio source and
+binary release. The synchronous release boundary is a successful push of one
+new tag to `origin`; Woodpecker and Gitea publication happen later and do not
+change that result.
+
+## Go prerequisite
+
+Release preparation requires the maintained Go 1.26 baseline. Tag CI currently
+uses the explicitly pinned `golang:1.26.7` image for candidate validation and
+asset construction; that asynchronous work does not change the tag-push
+completion boundary.
+
+## Choose a version and write its note
+
+Narratio is past `v1.0.0`. Use an unused stable tag in the exact form
+`vMAJOR.MINOR.PATCH`:
+
+- increment `MINOR` for backward-compatible features;
+- increment `PATCH` for backward-compatible fixes; and
+- reserve a new `MAJOR` for an intentional breaking documented contract.
+
+Before preparing the candidate, create
+`docs/releases/vMAJOR.MINOR.PATCH.md` with this structure:
+
+```markdown
+# Narratio vMAJOR.MINOR.PATCH
+
+This release ...
+
+## Summary
+
+## Compatibility
+
+## Upgrade
+
+## Changes
+```
+
+The compatibility section identifies relevant CLI, configuration, artifact,
+integration, or operating-contract changes. The upgrade section states the
+required operator action, or explicitly says that no special action is
+required. Release notes are immutable historical summaries; link to the
+current canonical documentation for detailed behavior.
+
+Commit the note and all candidate changes, then use the ordinary development
+workflow to push that commit to `main`. Do not create a release tag before the
+candidate is committed and `origin/main` contains the exact same commit.
+
+## Validate the candidate
+
+Run the shared checker from any directory:
+
+```sh
+scripts/check-release-candidate.sh vMAJOR.MINOR.PATCH
+```
+
+It validates the version and matching note, module hygiene, formatting,
+whitespace, uncached tests, race tests, static checks, documentation, examples,
+and three official cross-build assets. It uses `GOWORK=off`, does not contact
+application services, CI, or Gitea, and does not create tags or modify tracked
+source. Fix any failure on `main`, commit it, push it normally, and rerun the
+checker.
+
+The checker builds Linux assets for `amd64` and `arm64` and a macOS asset for
+`arm64`. The macOS cross-build proves compilation; it is not native macOS
+runtime evidence.
+
+## Publish the tag
+
+From a clean checkout on `main` whose `HEAD` equals `origin/main`, run:
+
+```sh
+scripts/release.sh vMAJOR.MINOR.PATCH
+```
+
+The command fetches and checks `origin/main`, re-runs candidate validation,
+then fetches and checks again before creating an explicitly unsigned lightweight
+tag for the originally recorded commit. It refuses dirty, divergent, changed,
+or already-tagged candidates. It pushes only:
+
+```text
+refs/tags/vMAJOR.MINOR.PATCH:refs/tags/vMAJOR.MINOR.PATCH
+```
+
+It never commits changes, pushes `main`, force-pushes, moves a tag, or pushes
+all tags. A successful push of that exact ref completes the release command;
+the command prints the tag and commit, then returns without waiting for CI,
+querying Gitea, downloading assets, or checking checksums.
+
+If a failure occurs before the tag is created, correct the candidate on `main`
+and repeat validation. If the push fails after local tag creation, the local tag
+is intentionally retained for inspection and the command must not be retried
+blindly. Once the upstream tag has been pushed, it is immutable. Correct any
+defect or failed asynchronous publication with a new patch version, a new
+release note, and the complete procedure again.
+
+## Optional asynchronous inspection
+
+After a successful tag push, a human may later inspect the tag-triggered
+Woodpecker run and the corresponding Gitea release for binaries and checksums.
+This is optional follow-up only. Automated releasers must not wait for, poll,
+or treat CI/Gitea completion as a condition of the successful tag push.

docs/releases/README.md

@@ -0,0 +1,28 @@
+# Release Notes
+
+This directory contains immutable historical release notes for Narratio.
+Future notes are created with the matching stable version and use this minimum
+structure:
+
+```markdown
+# Narratio vMAJOR.MINOR.PATCH
+
+This release ...
+
+## Summary
+
+## Compatibility
+
+## Upgrade
+
+## Changes
+```
+
+See the [release procedure](../release.md) for creating a candidate and tag.
+When asynchronous publication succeeds, the corresponding Gitea release is the
+canonical source for downloadable binaries and checksums.
+
+- [v2.0.0](v2.0.0.md)
+- [v1.7.0](v1.7.0.md)
+- [v1.6.0](v1.6.0.md)
+- [v1.5.0](v1.5.0.md)

docs/releases/v1.5.0.md

@@ -0,0 +1,47 @@
+# Narratio v1.5.0
+
+Narratio v1.5.0 makes repeated post-transcript artifact development faster and
+more explicit while retaining the fixed, stage-driven pipeline model.
+
+## Highlights
+
+- The canonical pipeline now completes deterministic rendering before
+  extraction, cleanly separating transcript-generating stages from
+  artifact-generating stages.
+- `narratio run` and `narratio session plan` accept inclusive `--from` and
+  `--through` bounds. Excluded transcript stages are not executed or
+  invalidated by a bounded artifact-regeneration run.
+- `narratio regenerate-artifacts SESSION` is an exact convenience alias for a
+  forced run from `extract` through `analyze`, including focused
+  `--artifacts` selections.
+- Configured Scriptorium artifacts now have independent,
+  manifest-authoritative freshness. Narratio reuses validated current work,
+  rebuilds stale prerequisites in dependency order, and persists successful,
+  failed, and newly stale artifact state when an analysis invocation only
+  partially succeeds.
+- Publish consumes only configured artifacts backed by current manifest
+  evidence; incidental or tampered files are not promoted as current output.
+
+## Reliability And Administration
+
+- Bounded prerequisites are checked again under the session lock before any
+  run mutation, closing a concurrent-run race.
+- Analysis fingerprints are stable across executable and configuration path
+  changes and continue to cover only Narratio-observable semantic inputs.
+- Runner composition now carries one validated execution plan from command
+  parsing through prerequisite validation, adapter composition, manifest
+  recording, and stage execution.
+- `narratio version` reports the exact tag embedded in official release
+  binaries; ordinary source builds report `dev`.
+
+## Upgrade Notes
+
+- Existing unbounded commands and direct `run-stage`, `analyze`, and `publish`
+  workflows retain their meanings.
+- Manifests written before artifact-level analysis state remain readable.
+  Legacy aggregate analysis success is not sufficient freshness evidence, so
+  the first analysis evaluation after upgrading may regenerate configured
+  artifacts once.
+- Narratio cannot observe executable contents or configuration, prompt,
+  profile, module, and other files loaded privately by Scriptorium. Explicitly
+  force affected artifacts after changing those private inputs.

docs/releases/v1.6.0.md

@@ -0,0 +1,74 @@
+# Narratio v1.6.0
+
+Narratio v1.6.0 makes large pipeline configurations easier to organize,
+inspect, and vary while adding character-oriented artifact generation and a
+guarded, reproducible release procedure.
+
+## Summary
+
+Pipeline configuration can now be assembled from explicit additive imports and
+a selected production or testing profile. Campaigns can own a canonical,
+versioned party roster, and Scriptorium artifact families can expand one
+definition into concrete per-character artifacts, dependencies, variables, and
+publish rules.
+
+New read-only configuration commands expose the fully resolved pipeline,
+source provenance, semantic digest, and profile differences before a session is
+run. Stage reuse now records configuration-sensitive semantic evidence so
+profile or configuration changes cannot silently reuse incompatible work.
+
+## Compatibility
+
+This is a backward-compatible feature release. Existing monolithic pipeline
+files, concrete Scriptorium artifacts, publish rules, and configurations
+without profiles remain supported. When profiles are declared and no explicit
+profile is selected, Narratio uses the configured production default.
+
+An unversioned party file plus a separate players file remains available as an
+isolated legacy compatibility path, but it cannot drive artifact families. New
+campaigns and new party-oriented features should use the `narratio.party.v1`
+schema. Existing manifests remain readable; missing legacy semantic evidence
+is treated as stale rather than trusted.
+
+Narratio continues to consume the documented Notarius D&D pipeline contract.
+The release workflow still cross-compiles Linux, macOS, and Windows binaries
+for `amd64` and `arm64`; cross-compilation is not native runtime evidence for
+macOS or Windows.
+
+## Upgrade
+
+No special action is required for existing monolithic configurations that do
+not adopt profiles or artifact families. On the first run after upgrading,
+stages recorded by older manifests may regenerate once because those records do
+not contain the new semantic configuration evidence.
+
+To adopt the new configuration model, use the maintained
+`examples/production-testing` bundle as a migration reference: split stable
+settings into explicit imports, define a production default and optional
+testing profile, convert campaign party data to `narratio.party.v1`, remove the
+separate players file, and then introduce character artifact families. Review
+the result with `narratio config validate`, `config show`, `config sources`, and
+`config diff` before running a session.
+
+Campaign, session, previous-session, and run identifiers must satisfy the
+documented portable identity grammar. Existing manifests or remote state with
+unsafe legacy identifiers must be migrated before use.
+
+## Changes
+
+- Added root-owned, non-recursive additive pipeline imports with strict,
+  source-aware conflict detection.
+- Added named pipeline profiles with an explicit production default and
+  deliberate command-line selection.
+- Added `config validate`, `config show`, `config sources`, and `config diff`
+  for read-only inspection of resolved configuration and provenance.
+- Added the strict `narratio.party.v1` campaign roster, including stable
+  character IDs, player and character names, optional aliases, and classes.
+- Added deterministic players derivation and canonical party delivery to
+  downstream integrations.
+- Added character-oriented artifact families, corresponding member
+  dependencies, family selection, and generated publish policies.
+- Added semantic configuration fingerprints and stage-specific resume checks
+  across transcript and artifact stages.
+- Added shared release candidate, asset build, and guarded tag-publication
+  scripts, with tag-triggered publication remaining asynchronous.

docs/releases/v1.7.0.md

@@ -0,0 +1,100 @@
+# Narratio v1.7.0
+
+Narratio v1.7.0 completes the maintainability refactor begun after v1.6.0. It
+simplifies filesystem and process handling, makes resumable work more precise,
+replaces legacy local and remote state with explicit authorities, and turns
+cleanup into the final ordinary pipeline stage.
+
+## Summary
+
+The session manifest is now the sole durable local progress ledger. Analyze
+checkpoints successful artifacts individually, allowing an interrupted or
+partially failed analysis to resume at the first incomplete artifact without
+rerunning successful siblings unless forced.
+
+Remote publication now commits immutable run state before advancing one
+verified current pointer. Publish locks are independent per-source objects;
+locked outputs retain their previously committed declarations rather than
+being overwritten or silently omitted. Local cleanup runs after notification
+as a separately recorded, safely retryable stage.
+
+The release also adopts Go 1.26, simplifies ordinary file operations around
+Narratio's trusted group-owned workspace model, improves subprocess lifecycle
+handling, decomposes several large orchestration and validation components,
+and removes mutable global test seams.
+
+## Compatibility
+
+This is a minor release with deliberate removal of compatibility paths that
+were deprecated and documented as subject to removal:
+
+- Campaigns must use one versioned `narratio.party.v1` `party.yml`. Separate
+  players files, session-level party overrides, and unversioned party documents
+  are rejected.
+- Session manifests must declare `schema_version:
+  narratio.session-manifest.v1` and satisfy the current strict schema. Older
+  unversioned state and manifests containing the retired
+  `post_publish_cleanup` obligation are rejected rather than migrated.
+- Remote current state must be selected by
+  `current/commit-pointer.json` and its referenced immutable run commit. Legacy
+  current-state layouts are no longer inferred.
+- The retired aggregate `locks.yml` object is ignored. Publish locks now live
+  as independent strict objects under the session lock prefix.
+- Run manifests are no longer produced or consumed. The session manifest owns
+  durable stage progress; invocation summaries and run workspaces remain
+  operational data rather than a second progress authority.
+
+The CLI and current pipeline/configuration formats otherwise remain compatible.
+Building and maintaining Narratio now requires Go 1.26. Cross-built Linux,
+macOS, and Windows assets remain available for `amd64` and `arm64`.
+
+## Upgrade
+
+Before upgrading a campaign that still uses compatibility data:
+
+1. Replace legacy party/player inputs with the documented
+   [`narratio.party.v1`](../integrations/party.md) campaign roster.
+2. If a local manifest contains an outstanding `post_publish_cleanup`
+   obligation, complete it with the pre-upgrade binary. Otherwise verify the
+   remote commit, move aside or regenerate the incompatible manifest, or
+   restore a verified committed snapshot. Do not migrate state by manually
+   editing JSON.
+3. Republish sessions that do not yet have a valid immutable run commit and
+   `current/commit-pointer.json`; legacy remote layouts are not current state in
+   v1.7.0.
+4. Recreate any required publication locks with `narratio session locks add`.
+   After verifying the new locks and publication behavior, the ignored legacy
+   `locks.yml` object may be removed manually.
+
+Existing compatible session manifests and already committed remote snapshots
+need no special action. Consult [Operations](../operations.md) and
+[Troubleshooting](../troubleshooting.md) for recovery procedures.
+
+## Changes
+
+- Upgraded the maintained development and release baseline to Go 1.26.
+- Reworked subprocess execution and Windows process-tree supervision while
+  preserving contextual wait and cleanup failures.
+- Replaced custom confined ordinary I/O and directory-promotion machinery with
+  simpler standard-library operations while retaining bounded reads, atomic
+  single-file visibility, workspace modes, validated downloads, and
+  root-confined destructive operations.
+- Removed legacy party, manifest, artifact-discovery, remote-current, and lock
+  compatibility paths.
+- Consolidated durable progress in the session manifest and removed the run
+  manifest domain model.
+- Unified runner planning and execution decisions, separated invocation
+  lifecycle responsibilities, and avoided creating empty run workspaces for
+  fully reused or cleanup-only invocations.
+- Added per-artifact analyze execution, checkpointing, reconciliation, and
+  recovery.
+- Decomposed configuration validation by domain and made validation diagnostics
+  deterministic without treating incidental ordering or wording as a contract.
+- Introduced immutable remote publish commits, one final current pointer,
+  explicit output planning, safe cross-run carried outputs, and independent
+  per-artifact remote locks.
+- Added `cleanup` as the final canonical stage after `notify`, with validated
+  publish evidence, safe retry, explicit selection, and no separate cleanup
+  obligation.
+- Replaced mutable command test hooks with invocation-scoped dependencies and
+  improved structural test diagnostics.

docs/releases/v2.0.0.md

@@ -0,0 +1,94 @@
+# Narratio v2.0.0
+
+Narratio v2.0.0 replaces its configuration infrastructure with ConfigKit,
+simplifies the supported release platforms, and makes artifact definitions and
+their effective dependency graph directly inspectable.
+
+## Summary
+
+Pipeline roots now use ConfigKit v0.1.2 for strict YAML loading, explicit
+imports, fixed-path named profiles, provenance, defaults, and typed credential
+references. The recommended production layout keeps production values in the
+base, defines each ordinary artifact and artifact family in its own imported
+file, and keeps testing changes in `profiles/testing.yml`.
+
+Configured-artifact inputs now create their dependency edges automatically.
+`narratio config artifacts` reports the finalized concrete catalog, including
+generated party-family members, without loading a session or initializing
+runtime adapters.
+
+## Compatibility
+
+This is a breaking configuration and platform release. There is no runtime
+fallback for the retired Narratio composition/profile catalog, `pipeline.secrets`
+or `secrets.env_dir`, credential `_env` fields, or the previous YAML
+presence/default behavior. Defaults apply only to absent values; explicit empty,
+zero, false, and null values retain their typed meaning and may fail Narratio
+domain validation.
+
+The maintained binaries are Linux amd64, Linux arm64, and macOS arm64. Windows
+and macOS amd64 builds are no longer produced or supported. Existing v1 release
+notes remain historical descriptions of their releases.
+
+Inspection commands still avoid session, manifest, workspace, adapter, and
+remote-service activity, but they are not credential-free loaders: every
+explicit typed credential reference in the selected effective configuration must
+resolve; references replaced or removed by that profile need not resolve.
+Pipelines containing artifact families also require a campaign so the canonical
+party can be expanded.
+`config show` is a secret-safe review projection, not a promised reloadable
+authoring file.
+
+## Upgrade
+
+Migrate each operational bundle before running v2:
+
+1. Replace old composition syntax with root-owned
+   [`configkit.imports`](../config.md#pipeline-imports-and-profiles). Imports are
+   additive base sources; they do not override one another.
+2. Fold the old production overlay into the base pipeline. Move testing changes
+   to `profiles/testing.yml`; compare the two independent resolutions with
+   `narratio config diff default testing`.
+3. Follow the maintained
+   [`examples/production-testing`](../../examples/production-testing/pipeline.yml)
+   layout: one explicit artifact entry per `artifacts/<key>.yml` and one family
+   entry per `artifact-families/<key>.yml`. The map key, not the filename, is the
+   stable artifact identity.
+4. Move credential files to an untracked `secrets/` directory beside
+   `pipeline.yml`, or pass `--secrets-dir`. Rename S3 fields to `access_key_id`
+   and `secret_access_key`, and rename the Audita field to `llm_api_key`.
+5. For static S3 credentials, configure both S3 references (commonly
+   `OBJECT_STORAGE_KEY_ID` and `OBJECT_STORAGE_KEY`) and make them resolvable.
+   To use instance roles, workload identity, shared SDK files, or another AWS SDK
+   discovery provider, omit both references. A partial pair is invalid.
+6. Remove a `depends_on` entry when it only duplicates a
+   `narratio.artifact.<key>` input. Keep explicit entries that express ordering
+   without supplying the upstream output as an input.
+7. Review ConfigKit's stricter explicit-value/null semantics and correct values
+   that previously acquired defaults after being explicitly empty. Do not use
+   `config show` output as a source migration mechanism.
+8. Run `config validate`, `config show`, `config sources`, and `config artifacts`
+   with the same profile, campaign, and secrets selection intended for runtime.
+   Resolve the campaign-owned party whenever families are present.
+9. Update deployment and download automation to one of the three maintained
+   platform assets before removing the v1 binary.
+
+The [configuration reference](../config.md), [CLI reference](../cli.md), and
+[maintained examples](../../examples/README.md) are the canonical detailed
+guides.
+
+## Changes
+
+- Adopted ConfigKit v0.1.2 across pipeline, campaign, session, and party loading
+  while keeping Narratio domain validation and post-campaign expansion explicit.
+- Added ConfigKit imports and fixed `profiles/<name>.yml` pipeline overlays,
+  unified safe effective show/diff/digest projection, and source provenance.
+- Added invocation-scoped ConfigKit secrets-directory bootstrap and typed S3 and
+  Audita credential references without exposing resolved values.
+- Inferred, normalized, and validated configured-artifact input dependencies as
+  part of the one concrete graph consumed by analyze planning and resume logic.
+- Added `config artifacts` with stable concrete identities, definition sources,
+  effective dependencies, and generated family/party provenance.
+- Added the maintained production-base/testing-overlay artifact authoring layout.
+- Reduced release assets and active platform support to Linux amd64/arm64 and
+  macOS arm64.

docs/roadmap/configkit-adoption.md

@@ -0,0 +1,375 @@
+# ConfigKit Adoption and Platform Simplification
+
+**Status:** Accepted; implementation pending
+
+## Purpose
+
+Narratio currently owns a substantial YAML configuration subsystem for strict
+decoding, explicit imports, additive composition, profile overlays, presence
+tracking, provenance, and effective-value inspection. ConfigKit was extracted
+from this implementation and now provides those generic responsibilities as a
+maintained library.
+
+Narratio 2.0 will adopt ConfigKit as its configuration foundation, remove the
+superseded generic infrastructure, and retain only Narratio-specific configuration
+policy and domain resolution. The same major release will narrow the supported
+platform set to the systems that Narratio is expected to run on in practice.
+
+This roadmap defines the desired contracts and end state. The
+[implementation plan](implementation.md) owns sequencing and completion gates.
+
+## Goals
+
+- Reduce Narratio-owned configuration code and tests by delegating generic YAML
+  mechanics to ConfigKit.
+- Establish one strict and explainable configuration lifecycle rather than
+  retaining parallel or compatibility loaders.
+- Preserve Narratio's application-specific defaults, validation, campaign/session
+  resolution, party handling, artifact-family expansion, digests, and CLI policy.
+- Make individual artifact definitions easy to organize, identify, inspect, and
+  reuse as dependencies of later artifacts.
+- Keep raw credentials out of YAML and adopt a predictable secrets-directory
+  convention that works before configuration loading.
+- Remove unsupported platform code and release obligations that provide no known
+  user value.
+- Deliver the resulting contract changes as Narratio v2.0.0 with a clear migration
+  guide.
+
+## Platform Contract
+
+Narratio's v2 platform matrix is:
+
+| Platform | Status | Required evidence and deliverables |
+| --- | --- | --- |
+| Linux amd64 | Primary, supported | Default tests, race checks, static analysis, build validation, and a release asset |
+| Linux arm64 | Primary, supported | Cross-build validation and a release asset; native validation is used when an appropriate runner is available |
+| macOS arm64 | Best effort | Cross-build validation and a release asset; native CI and platform-equivalent runtime guarantees are not required release gates |
+| macOS amd64 | Unsupported | No build gate, release asset, compatibility promise, or platform-specific maintenance |
+| Windows | Unsupported | No build gate, release asset, compatibility promise, or platform-specific maintenance |
+
+Windows- and Intel-Mac-specific code, tests, CI entries, release packaging, and
+documentation are removed when they no longer serve a supported build. Portable
+code should remain portable where doing so is natural, but accidental compilation
+on an unsupported target does not create a support commitment.
+
+Best-effort macOS arm64 support means Narratio intends to keep the target buildable
+and will consider reported defects when practical. Lack of a native runner or an
+unverified platform-specific behavior does not block a release. A target for which
+Narratio publishes an asset must still pass its configured cross-build gate.
+
+## ConfigKit Dependency and Ownership
+
+Narratio will depend on the tagged ConfigKit module at an exact reviewed version,
+initially `gitea.maximumdirect.net/eric/ConfigKit@v0.1.1`. ConfigKit is pre-1.0 and
+may make compatibility changes in later minor versions, so dependency upgrades are
+intentional changes accompanied by contract review and focused Narratio integration
+tests; Narratio does not follow an unpinned branch or automatically accept a newer
+minor release.
+
+ConfigKit owns:
+
+- bounded file, byte, and reader input;
+- the accepted YAML syntax and strict typed decoding;
+- import declarations, traversal, containment, and resource limits;
+- additive composition and conflict attribution;
+- selected-profile overlay semantics;
+- explicit input presence and source provenance;
+- application-default and recorded-resolution coordination;
+- structured configuration diagnostics;
+- secret-reference parsing and resolution;
+- secrets-directory loading and its safe report; and
+- detached effective configuration inspection supplied by its public result.
+
+Narratio owns:
+
+- pipeline, campaign, session, party, and runtime domain types;
+- configuration-file discovery and CLI selection;
+- concrete defaults, normalization, and semantic validation;
+- campaign/session identity and stable-input resolution;
+- canonical party parsing requirements and authored-byte retention;
+- character artifact-family expansion and generated ownership attribution;
+- root-relative interpretation of Narratio-owned runtime paths;
+- effective pipeline digests and semantic profile diffs;
+- command output and exit behavior; and
+- adapter and runtime initialization after configuration succeeds.
+
+Narratio must use ConfigKit through its public package. It must not import
+ConfigKit's private YAML implementation or reproduce ConfigKit's parser,
+composition, profile, provenance, or secret-bootstrap mechanics locally.
+
+## Configuration Format
+
+Narratio adopts ConfigKit's configuration conventions directly. The old Narratio
+`composition:` envelope, declared profile catalog, arbitrary overlay paths, and
+`composition.default_profile` are removed without a compatibility parser.
+
+The root pipeline uses `configkit.imports` declarations. Import paths, optional
+imports, globs, recursion, ordering, composition conflicts, containment, and limits
+have the meanings defined by the pinned ConfigKit contract. Narratio does not add a
+second import syntax or narrower application-specific traversal implementation.
+
+The unoverlaid root and its imports are the production configuration. A named
+profile is a ConfigKit overlay at `profiles/<name>.yml`; in particular,
+`profiles/testing.yml` contains inexpensive-model overrides and testing-only
+artifacts. Omitting `--profile`, or selecting ConfigKit's `default` profile, uses the
+production base. `--profile testing` applies the testing overlay. Narratio does not
+maintain a separate configured default-profile mechanism or validate unselected
+profile files.
+
+ConfigKit's rules for explicit false, zero, empty, and null values replace
+Narratio's custom presence tracking. Application defaults and normalization use a
+ConfigKit definition and recorded resolution where the final value depends on
+whether input was present. `NormalizeConfig.UnmarshalYAML` and equivalent
+parser-coupled application hooks are removed.
+
+Pipeline, campaign, session, and party inputs use the common ConfigKit loading
+boundary where its schema model can represent their types. Narratio may perform
+narrow pre- or post-load domain work, such as retaining authored party bytes,
+rejecting unresolved session-template placeholders, resolving stable inputs, and
+expanding artifact families. Such work must not recreate a second generic YAML
+loader.
+
+## Artifact Definition Ergonomics
+
+The canonical v2 configuration layout defines each ordinary Scriptorium artifact
+in its own file under `artifacts/`. The root pipeline imports the directory through
+ConfigKit, for example:
+
+```yaml
+configkit:
+  imports:
+    - glob: artifacts/*.yml
+```
+
+Each file contributes one ordinary entry at
+`scriptorium.artifacts.<artifact_key>`. The map key is the authoritative Narratio
+artifact key and therefore makes the artifact's internal identity explicit in the
+definition itself:
+
+```yaml
+# artifacts/session_recap.yml
+scriptorium:
+  artifacts:
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      output_path: artifacts/session_recap.md
+```
+
+The canonical reusable source ID is derived without another configured identifier:
+`session_recap` becomes `narratio.artifact.session_recap`. The recommended filename
+is `<artifact_key>.yml`, but the filename is organizational rather than an identity
+source. Renaming or moving a file without changing its map key does not change the
+artifact contract. ConfigKit additive composition rejects duplicate artifact keys.
+
+The one-artifact-per-file layout is the documented happy path and is used by
+maintained examples. Narratio does not prohibit an inline artifact or another
+ConfigKit-compatible source organization merely because it differs from that
+layout. Artifact identity always comes from the effective map key, never implicitly
+from a filename or import position.
+
+Artifact families use the corresponding `artifact-families/*.yml` convention, with
+one `scriptorium.artifact_families.<family_key>` entry per file. The family key is
+explicit. Concrete member keys and source IDs continue to use Narratio's existing
+deterministic family/member naming rule after canonical party resolution.
+
+Narratio adds a read-only `narratio config artifacts` command using the ordinary
+pipeline, profile, and campaign selectors. It reports every effective ordinary
+artifact's key, canonical `narratio.artifact.<key>` source ID, defining source,
+enabled state, output path, and effective dependencies. Generated family members
+also report their family origin. Output ordering is deterministic by artifact key,
+and the command does not require a session or initialize runtime adapters.
+
+A configured-artifact input creates an implicit dependency on the referenced
+artifact. For example:
+
+```yaml
+inputs:
+  recap:
+    source: narratio.artifact.session_recap
+```
+
+The effective dependency graph is the union of explicit `depends_on` entries and
+dependencies inferred from `narratio.artifact.<key>` inputs. Inferred and explicit
+duplicates collapse to one dependency. `depends_on` remains available for an
+ordering requirement when the upstream output is not supplied as an input.
+Validation applies the same existence, self-reference, and cycle rules to both
+kinds of dependency, and diagnostics identify the declaration or input that
+introduced an invalid edge. Operators no longer need to repeat an artifact input in
+`depends_on` merely to make that input valid.
+
+## Effective Inspection, Digests, and Profile Diff
+
+ConfigKit results are the authority for composed input presence, base/profile
+origins, diagnostics, and the initial typed effective configuration. Narratio's
+configuration commands adapt those results to existing application-facing command
+formats rather than exposing ConfigKit implementation types directly.
+
+Party-dependent artifact families are expanded only after the selected campaign and
+canonical party are known. That derived state is intentionally outside ConfigKit's
+completed load snapshot. Narratio retains a narrow application-owned layer that:
+
+- records generated family/member and publish-rule attribution;
+- projects the final post-expansion pipeline for operator inspection;
+- recomputes the effective pipeline digest after expansion; and
+- ensures stages and adapters receive only the final concrete configuration.
+
+Narratio continues to own semantic `config diff` behavior because ConfigKit does
+not provide configuration diffs or fingerprints. Each side of a profile comparison
+may be loaded independently through ConfigKit. The former guarantee that both sides
+come from one parsed source snapshot is removed; operators must not rely on a
+meaningful comparison while concurrently modifying its source files. Equality is
+still based on normalized effective values rather than YAML formatting or source
+layout.
+
+## Credential and Secrets Convention
+
+Raw API keys and other credential values remain prohibited in pipeline, campaign,
+session, generated configuration, manifests, diagnostics, examples, and logs.
+Credential-bearing configuration fields use ConfigKit secret references: YAML
+contains an environment-variable name, while the resolved value remains available
+only through ConfigKit's trusted typed boundary. Fields should use credential names
+rather than legacy names ending in `_env` when the v2 schema is updated.
+
+The `pipeline.secrets.env_dir` stanza is removed. Narratio selects a secrets
+directory before parsing the pipeline using this precedence:
+
+1. an explicitly supplied `--secrets-dir PATH`; otherwise
+2. the `secrets` subdirectory beside the selected root pipeline file.
+
+The conventional directory is derived from the resolved root pipeline path, never
+from the current working directory, an imported file, or a selected profile. A
+relative explicit CLI path is resolved from the caller's current working directory,
+consistent with an ordinary command-line path argument.
+
+If an explicitly selected directory is missing or invalid, the command fails with a
+clear bootstrap diagnostic. If the conventional directory is absent, Narratio
+performs no directory bootstrap and continues; pre-existing environment variables
+may satisfy all references. If the conventional directory exists but ConfigKit
+rejects or cannot read it, the command fails. Already-set environment variables,
+including empty values, retain ConfigKit's documented precedence.
+
+Narratio explicitly calls `configkit.LoadSecretsDir` at most once for a command,
+before the first ConfigKit load that may resolve secret references and before any
+concurrent work or runtime client initialization. This application-level convention
+does not make ConfigKit's ordinary `Load` mutate process-global state. The bootstrap
+report is rendered or summarized through a safe Narratio command boundary without
+secret values.
+
+Narratio adopts ConfigKit's secret-directory rules and limits without restoring its
+old permission-mode, selected-directory-symlink, invalid-entry, byte-limit, or
+newline-trimming behavior through compatibility wrappers. Configuration inspection
+may require the referenced environment variables or the selected secrets directory
+because an ordinary successful ConfigKit load resolves effective typed secret
+references. Inspection output may show non-secret reference names where appropriate,
+but must never expose resolved values.
+
+## Validation and Diagnostics
+
+Narratio's semantic rules remain recognizable application code registered with the
+appropriate ConfigKit definition. Validation continues to reject invalid pipeline,
+campaign, session, party, adapter, publish, and artifact relationships before
+runtime work where practical. Validation order and exact prose are not compatibility
+contracts.
+
+ConfigKit diagnostic codes, paths, and source locations should be preserved through
+the application error boundary where useful. Narratio adds command context and exit
+classification but does not parse ConfigKit error strings or expose raw parser
+errors. Secret values must remain absent from success and failure diagnostics.
+
+## Compatibility and Migration
+
+This work intentionally changes supported platforms, pipeline composition syntax,
+profile layout and selection, secret configuration, YAML semantics, diagnostics,
+and some inspection prerequisites. It is released as Narratio v2.0.0.
+
+There is no runtime compatibility layer for the old `composition` or
+`pipeline.secrets` syntax. Maintained examples are converted to the ConfigKit
+format, and the v2 release documentation provides a concise migration guide that
+covers:
+
+- converting imports to `configkit.imports` declarations;
+- folding the production overlay into the base configuration;
+- placing testing overrides in `profiles/testing.yml`;
+- moving ordinary artifacts and artifact families into the recommended individual
+  definition files and using their map keys as canonical identities;
+- removing redundant `depends_on` entries when a configured-artifact input now
+  supplies the same dependency;
+- moving secret files to the conventional sibling `secrets/` directory or using
+  `--secrets-dir`;
+- renaming credential-reference fields where applicable;
+- ConfigKit's stricter YAML and explicit-value semantics;
+- the independent-load behavior of `config diff`; and
+- the removal of Windows and macOS amd64 artifacts.
+
+Obsolete application implementation, tests, documentation, examples, CI paths, and
+release machinery are removed rather than left deprecated inside the v2 codebase.
+
+## Required Changes
+
+- Add and pin the ConfigKit module dependency and establish a small application
+  integration boundary for definitions, result adaptation, and diagnostics.
+- Replace Narratio's generic YAML parsing, composition, import, profile, presence,
+  and provenance implementation with ConfigKit public APIs.
+- Express existing defaults, presence-sensitive normalization, and semantic
+  validation through ConfigKit definitions without moving Narratio domain policy
+  into the library.
+- Adapt pipeline/campaign/session/party loading and preserve the existing
+  application resolution steps that occur after their independent typed loads.
+- Rework effective configuration, sources, and diff commands around ConfigKit
+  results plus Narratio's post-party generated-attribution layer.
+- Add the maintained per-artifact and per-family file layouts, the read-only
+  `config artifacts` catalog, and effective dependency inference from configured
+  artifact inputs.
+- Convert credential configuration to typed references, add the common
+  `--secrets-dir` option, implement conventional-directory selection, and replace
+  the custom API-key directory loader with `configkit.LoadSecretsDir`.
+- Update adapters to consume resolved credentials or safe environment reference
+  names from typed configuration without independently rediscovering secret values.
+- Remove Windows and macOS amd64 build targets, release assets, CI entries,
+  platform-specific implementation, and obsolete tests and documentation.
+- Convert maintained examples and add v2 migration and release documentation.
+- Retain focused Narratio tests for domain policy and assembled application
+  behavior while deleting parser/composer/bootstrap mechanism tests now owned by
+  ConfigKit. Tests remain deterministic, offline, credential-free, and aligned with
+  the repository testing policy.
+- Update canonical architecture, configuration, CLI, operations, development,
+  internal, troubleshooting, and release documentation only as the corresponding
+  behavior is implemented.
+
+## Target End State
+
+Narratio v2.0.0 has one configuration infrastructure dependency and no duplicate
+YAML composition subsystem. Operators maintain a production base configuration,
+optional explicit imports, and fixed named overlays such as `testing`. ConfigKit
+provides strict parsing, composition, typed loading, provenance, diagnostics, and
+credential resolution; Narratio supplies domain meaning and its post-campaign
+artifact expansion.
+
+Operators can add an artifact by creating one discoverable YAML file whose map key
+is its stable Narratio identity. `config artifacts` exposes that key, its canonical
+source ID, origin, output, and effective dependency set. A later artifact can reuse
+the output through `narratio.artifact.<key>` without repeating the same edge in
+`depends_on`, while explicit ordering-only dependencies remain supported.
+
+Secrets normally live in a predictable `secrets/` directory beside the selected
+pipeline, with `--secrets-dir` available for deployment-specific placement. Raw
+credential values never enter YAML or inspection output. The bootstrap is explicit
+in application startup and ordinary ConfigKit loading remains free of process-global
+mutation.
+
+Release and CI maintenance covers Linux amd64 and arm64 as the primary platforms
+and macOS arm64 on a best-effort basis. Windows and macOS amd64 carry no remaining
+support or release burden. The repository contains no legacy configuration parser,
+profile compatibility layer, obsolete platform implementation, or redundant tests
+for behavior owned by ConfigKit.
+
+## Deferred Work
+
+- A ConfigKit reference-only inspection or typed prepare/finalize lifecycle may be
+  considered upstream if credential-free inspection becomes a demonstrated shared
+  requirement. Narratio does not implement a private substitute in this migration.
+- ConfigKit hot reload, remote providers, automatic configuration migration, and
+  application command frameworks remain outside Narratio's scope.
+- Native macOS arm64 CI may be added later when a reliable runner is available; it
+  is not required for v2.0.0.

docs/roadmap/implementation.md

@@ -0,0 +1,522 @@
+# ConfigKit Adoption Implementation Plan
+
+Status: Complete
+
+Implement the [ConfigKit adoption roadmap](configkit-adoption.md) in the order
+below. The target is Narratio v2.0.0: ConfigKit v0.1.1, the reduced platform
+matrix, and the artifact-authoring workflow described there. Each stage is a
+bounded implementation assignment for a gpt-5.6-terra coding agent. Mark a stage
+complete only after its changes and acceptance checks succeed; record a short
+completion and verification note beside its status.
+
+## Instructions for every stage
+
+- Read `docs/development.md` and all three policies under `docs/policy/`.
+  Read the relevant roadmap section and the stage-specific sources listed below.
+  Paths identify starting points, not an exhaustive edit allowlist; inspect callers
+  and tests before changing a package boundary.
+- Use the tagged ConfigKit public API. The local `../ConfigKit` checkout is useful
+  reference material, but verify its version before relying on it. The module
+  cache for v0.1.1 is an alternative. Do not require a sibling checkout, a local
+  `replace`, a Go workspace, an upstream patch, or an unpinned revision to build.
+- Keep Narratio domain policy in `internal/config`, startup and selectors in
+  `internal/app`, and credential delivery in the owning adapters. Do not build
+  a general configuration framework around ConfigKit.
+- Keep each stage buildable and run focused behavioral tests for its changed
+  boundary. Temporary migration helpers may exist between stages only where
+  named here; they must not introduce runtime format auto-detection or fallback.
+- Update each canonical current-behavior document when its behavior changes.
+  The roadmap remains the owner of unimplemented work. Preserve historical
+  release notes; their descriptions of old platforms and formats are history.
+- Tests use real ConfigKit and temporary files, with fake external adapters or
+  storage where needed. Use synthetic credentials and isolate environment
+  mutation; such tests must not run in parallel. Retain domain and security
+  coverage, remove superseded mechanism tests, and avoid exact diagnostic prose
+  or private callback choreography assertions.
+- Do not tag or publish v2.0.0 as part of this plan. Release preparation belongs
+  here; release authorization and tag publication are a later task. The existing
+  release completion boundary remains successful upstream tag push, with no CI
+  polling or waiting.
+
+## Decisions that apply across stages
+
+The roadmap is authoritative for product scope. These implementation choices
+make that scope concrete:
+
+- Enable ConfigKit imports and profile selection for pipeline files only.
+  Campaign, session, and party are single-source definitions with imports and
+  profiles disabled. Preserve their domain schemas and source-relative inputs.
+  Omitted or empty profile and the exact name `default` select the base; other
+  names use ConfigKit's fixed `profiles/<name>.yml` convention.
+- Defaults apply to absent fields. Explicit empty/zero/false/null values follow
+  ConfigKit decoding and then domain validation; do not refill invalid explicit
+  values with legacy defaults. Optional null stage sections remain absent;
+  required sections fail domain validation rather than being resurrected.
+- Keep ConfigKit metadata/result snapshots separate from runtime-only Narratio
+  state. ConfigKit discards ignored and unexported fields when detaching values;
+  attach root paths, retained party bytes, family catalogs, and digest metadata
+  after loading. No validator may mutate bound configuration.
+- Use one Narratio-owned safe effective projection for the final expanded
+  pipeline. It contains configuration values and credential reference names,
+  never resolved credential bytes. It feeds show, semantic diff, digest, and
+  catalog data. ConfigKit metadata remains the authority for input provenance;
+  Narratio adds only domain-generated attribution.
+- Rename `storage.s3.access_key_id_env` to `access_key_id`,
+  `storage.s3.secret_access_key_env` to `secret_access_key`, and
+  `audita.llm_api_key_env` to `llm_api_key`. Each is an optional `*configkit.Secret`.
+  An explicit reference always resolves during loading, even if the associated
+  stage will not execute. Require both S3 fields together or neither. With neither,
+  retain the AWS SDK default credential chain; do not synthesize mandatory secret
+  references that would break role-based authentication or local-only inspection.
+  With no Audita reference, preserve its existing downstream credential discovery.
+  Explicit references replace Narratio's old per-adapter environment lookup.
+- Keep absent S3 credentials out of defaults; migration examples explicitly name
+  the existing AWS variables when static credentials are intended. A non-nil
+  missing/empty reference fails instead of silently falling back. This preserves
+  provider-managed credentials while making explicit configuration trustworthy.
+- `config artifacts` uses the existing configuration command selectors and text
+  output convention. No new JSON mode, artifact creation wizard, inferred filename
+  identity, or profile catalog is added.
+- Preserve the roadmap's unique base artifact definition rule explicitly:
+  ConfigKit rejects conflicting leaves but can compose disjoint children of the
+  same mapping. Use its public source metadata to reject an ordinary artifact key
+  declared by multiple base files, even if their fields are disjoint. Apply the
+  same rule to family keys. A selected profile may override or extend a definition.
+  This is a narrow Narratio declaration rule, not a different merge algorithm.
+- Relax only the obsolete configuration/secrets filesystem restrictions identified
+  by the roadmap. Do not remove session restore, cleanup, publication, workspace
+  confinement, or ordinary shared-workspace mode guarantees as collateral work.
+
+## Stage 1 — Reduce the platform and release matrix
+
+Status: Complete — Reduced CI and release output to Linux amd64/arm64 and macOS
+arm64, removed Windows-only runtime implementations and accommodations, and
+verified focused tests, all three cross-builds, exact asset names, and embedded
+version reporting.
+
+Read: roadmap “Platform Contract”; `scripts/build-release-assets.sh`,
+`scripts/check-release-candidate.sh`, `.woodpecker/`, `internal/releasecheck`,
+`docs/release.md`, and platform-specific files in application, file operations,
+and subprocess packages.
+
+1. Make Linux amd64, Linux arm64, and macOS arm64 the only maintained build and
+   release targets. Remove Windows suffix/packaging branches and macOS amd64
+   targets from scripts, workflows, tests, and current documentation.
+2. Delete Windows-only implementations and tests; simplify paired build-tag
+   files where the remaining implementations can be shared. Preserve portable
+   code when retaining it adds no special support machinery. Preserve functioning
+   Unix subprocess cancellation, locking, and file operations on both supported OSes.
+3. Retain the local Linux verification gates and cross-build gates for all three
+   published assets. Do not introduce a native macOS runner requirement or claim
+   that cross-compilation verifies runtime behavior.
+4. Update architecture/operations platform statements and contributor/release
+   guidance. Keep existing release protections and the no-CI-wait boundary intact.
+
+Acceptance: focused platform/subprocess/fileops/releasecheck tests pass; build all
+three targets with `CGO_ENABLED=0`; verify the asset builder produces exactly the
+three expected versioned binaries and the host binary reports the supplied version.
+Existing release tests still prove guarded publication. No current support promise
+or asset list retains Windows or macOS amd64.
+
+## Stage 2 — Separate configuration normalization from validation
+
+Status: Complete — Added explicit deterministic pipeline normalization before
+validation, made Notarius validation read-only on accepted and rejected inputs,
+documented the boundary, and verified config race tests plus app/stage consumers.
+
+Read: `internal/config/load.go`, `defaults.go`, `normalize.go`, `validate*.go`,
+`artifact_families.go`, and ConfigKit's “Recorded application resolution”,
+“Semantic validation”, and “Defaults and typed decoding” contracts.
+
+1. Separate value-producing normalization from read-only validation, including
+   Notarius reference/output normalization currently performed by
+   `validateNotarius`. Audit all validators for map, slice, pointer, and scalar
+   mutations, including changes that would occur only on error paths.
+2. Keep reusable domain normalization ordinary package code. Validation checks the
+   already normalized candidate and must work for both loaded and programmatically
+   constructed values without silently fixing them.
+3. Inventory defaults in `load.go`: static defaults, per-map-entry/list-element
+   defaults, and inherited or presence-sensitive defaults. Retain named constants
+   in `defaults.go`. This inventory should guide code and concise comments, not
+   become a permanent parallel defaults document.
+4. Keep the current loader calling normalization explicitly before validation
+   during this preparatory stage. The ConfigKit resolver replaces that loading
+   choreography in Stage 4. Preserve family expansion as an explicit domain step.
+
+Acceptance: focused configuration tests prove normalization is deterministic and
+validation does not mutate accepted or rejected inputs. Cover representative
+Notarius normalization/collision handling and existing artifact/publish rules.
+Do not add a separate mutation test for every trivial validator.
+
+## Stage 3 — Replace composition-dependent effective projection
+
+Status: Complete — Introduced one typed safe projection for effective YAML,
+semantic values, source paths, and the versioned v3 digest; removed composition
+parsing from those consumers and verified ordering, expansion, relocation,
+nil/empty, exact-integer, CLI diff, and semantic-resume behavior.
+
+Read: `internal/config/effective_pipeline.go`, `effective_digest.go`,
+`pipeline_composition.go`, `internal/app/config_commands.go`, the config-diff
+implementation, and `docs/internal/configuration.md`.
+
+1. Introduce one package-owned, schema-specific safe projection of the effective
+   pipeline. Use explicit typed projections with stable configuration field names;
+   preserve null versus empty containers and exact integers. Ordinary nested
+   configuration types may be reused where they contain no runtime state.
+   Credential fields must be projected explicitly as reference names when Stage 7
+   changes their types. Do not introduce a generic reflection schema engine.
+2. Render this projection as deterministic YAML using the existing YAML encoder;
+   encoding output is not a replacement YAML loader. Produce semantic records
+   directly from the safe projection using standard JSON encoding/decoding with
+   number preservation if a plain value tree is useful. Do not reparse YAML or
+   reconstruct `compositionNode` trees to compare or hash effective values.
+3. Make show, diff values, and digest use that common projection. Mapping order is
+   immaterial; sequences remain atomic for diff; nil and empty remain distinct.
+   Exclude family declarations once expanded, all runtime metadata, profile names,
+   source locations, and raw credentials. Preserve logical Notarius path handling
+   so moving an equivalent bundle does not change its digest.
+4. Version the changed digest encoding explicitly using the package's digest
+   version constant. Digest/profile metadata remains provenance, never a blanket
+   resume cache key. Keep stage/artifact semantic evidence behavior intact.
+5. Decouple enumeration of effective field paths from the legacy composition
+   renderer. Input ownership can still come from the current loader until Stage 4.
+
+Acceptance: show/diff/digest tests cover equivalent monolithic/split values,
+different map insertion order, null/empty distinction, family-expanded outputs,
+and bundle relocation. A meaningful artifact change changes the digest; formatting
+and profile naming alone do not. Existing semantic resume tests pass. No effective
+projection consumer needs the custom composition parser/renderer.
+
+## Stage 4 — Cut pipeline loading over to ConfigKit
+
+Status: Complete — Replaced pipeline composition with ConfigKit v0.1.1 imports,
+fixed-path profile selection, fresh typed defaults, recorded resolution, source
+metadata adaptation, and duplicate base artifact/family enforcement; removed the
+legacy parser and verified all tests plus the configuration race suite.
+
+Read: ConfigKit v0.1.1 `docs/config.md`, `docs/consumers/README.md`, and its
+Narratio-shaped compiled example; Narratio `internal/config/load.go`,
+`composition.go`, `pipeline_composition.go`, and profile/composition tests.
+
+1. Pin `gitea.maximumdirect.net/eric/ConfigKit v0.1.1` in `go.mod`/`go.sum`.
+   Implement a small pipeline definition using `Load`, `File`, enabled imports,
+   `Options.Profile`, defaults, recorded resolution, and read-only validation.
+2. Migrate static defaults to fresh default values. Apply map-entry, list-element,
+   and inherited defaults with `Resolution.Default`/`IsExplicit`; use recorded
+   normalization for transformations from Stage 2. Do not call old post-decode
+   zero-filling defaults. Remove `NormalizeConfig.UnmarshalYAML` and its private
+   presence bit. Include publication list-entry defaults, optional section
+   allocation, and trim's Scriptorium requirement in this conversion.
+3. Perform file-local semantic checks in the definition. Defer checks needing
+   canonical party expansion or campaign/session context to the existing explicit
+   resolution boundary; do not reject references to future concrete family members
+   merely because the campaign has not loaded yet. Run full checks after expansion.
+4. Replace pipeline composition and profile loaders. `configkit.imports` is the
+   only directive; no `composition` or `_configkit` alias is accepted. Pass profile
+   selection directly under the agreed convention; unselected overlays are not read.
+   `LoadPipelineProfilePair` may remain a thin pair of independent loads, with no
+   shared parse cache or paired source-snapshot promise.
+5. Adapt ConfigKit sources and metadata to Narratio's existing reporting records.
+   Translate paths structurally using public path APIs, not error-message parsing.
+   Preserve multiple contributing locations for composite/profile-owned mappings
+   and distinguish defaults/resolution from authored values.
+   Check artifact/family map-entry ownership for duplicate base declarations using
+   those locations; ignore synthetic defaults and allow the selected overlay.
+6. Attach runtime metadata only after load succeeds. Preserve root-relative runtime
+   path interpretation and capture logical paths before making them absolute.
+   Retain family declarations and original ConfigKit metadata for later expansion.
+7. Replace generic mechanism tests with a small integration set proving actual
+   Narratio wiring. Convert fixtures affected by this stage. Delete
+   `composition.go`, obsolete pipeline traversal/overlay/presence machinery, and
+   their redundant tests once the callers are migrated.
+
+Acceptance: a real root with imports and a testing overlay loads; duplicate claims
+fail with useful source attribution, including two base files contributing disjoint
+fields to one artifact or family key; base/default selection works; malformed
+unselected profiles do not fail base loading. Omitted defaults and representative
+explicit false/zero/empty/null inputs obey the new semantics. Invalid domain values
+still fail. Commands use the new pipeline loader with no format fallback. Update
+`docs/config.md` and configuration internals for the implemented boundary.
+
+## Stage 5 — Migrate campaign, session, and canonical party loading
+
+Status: Complete — Migrated campaign, session, and canonical-party documents to
+independent ConfigKit definitions, preserved bounded session inspection, exact
+party bytes and metadata-derived authored order, and retained identity, source,
+family-expansion, and validation behavior; verified all tests and focused races.
+
+Read: `internal/config/load.go`, `party.go`, `party_resolution.go`,
+`artifact_families.go`, session-template tests, `internal/app/config_loader.go`,
+and ConfigKit source/schema/version contracts.
+
+1. Use separate ConfigKit definitions for campaign, session, and party. Replace
+   `decodeStrictYAML*` and the custom party YAML-node parser. Model the party input
+   as the existing string schema version and character-keyed typed map, with the
+   existing `alias` list, ordered class list, and optional positive class level.
+2. Keep party domain checks: valid IDs, required display names and classes,
+   trimmed/control-free names, case-insensitive name/alias ambiguity, and duplicate
+   class checks. Validate the required version through `VersionRule`. Keep an
+   exact independent copy of authored bytes outside ConfigKit's schema object.
+3. Preserve authored character order using successful metadata source coordinates
+   for the single party input when constructing `CanonicalParty.Characters`.
+   Preserve class/alias sequence order naturally. Family expansion and players
+   projection continue using their existing deterministic ID sorting. Do not keep
+   a second YAML parse just to recover mapping order.
+4. Preserve the narrow session placeholder rejection before typed load, expected
+   session/previous-session identity checks, campaign selection, stable-input
+   precedence, and local versus remotely downloaded session provenance.
+   Retain the initially loaded pipeline/campaign during remote session acquisition.
+5. Where bytes must be retained or inspected before `ConfigKit.Bytes`, bound the
+   initial read too, using the existing fileops bounded-read helper and a named
+   package constant of 8 MiB matching the pinned source limit. Otherwise prefer
+   `ConfigKit.File`. Do not read an unbounded file and rely on later rejection.
+6. Complete post-party family and publish-rule expansion, then full cross-config
+   validation and digest refresh. Keep generated origins outside the library
+   snapshot. Remove unused decoder plumbing and obsolete party parsing tests.
+
+Acceptance: retained authored party bytes are identical; domain errors, aliases,
+class order, family origins, and derived players are correct. Local and remote
+session loads preserve identity, input path bases, and the original pipeline.
+Unknown fields and disabled directives fail through ConfigKit. Focused config,
+prepare/extract, and application loading tests pass.
+
+## Stage 6 — Move secrets bootstrap before all configuration loads
+
+Status: Complete — Added invocation-scoped ConfigKit secrets bootstrap before
+all pipeline loads and adapter construction, exposed consistent explicit and
+conventional directory selection across the CLI, removed the retired pipeline
+schema and custom loader, and verified focused, full, and race test suites.
+
+Read: roadmap “Credential and Secrets Convention”; ConfigKit's bootstrap
+contract; `internal/app/secrets_env*.go`, `config_loader.go`, `config_commands.go`,
+`object_store.go`, command dependencies, and command argument owners.
+
+1. Add a shared `--secrets-dir` selector to every CLI path that loads a pipeline,
+   including configuration inspection/diff, execution aliases, planning, publish,
+   and session/storage operations. Route all through one command-owned bootstrap
+   context. Help/version and commands that do not load configuration do not bootstrap.
+2. Discover the root pipeline first. An explicit relative directory is based on
+   caller CWD; otherwise choose `secrets/` beside the resolved root pipeline.
+   An explicitly empty flag is invalid. Missing conventional directories are
+   skipped; missing explicit directories and unreadable/invalid existing directories
+   fail. Use ConfigKit's selected-directory symlink behavior without custom rejection.
+3. Call `LoadSecretsDir` at most once per command before any ConfigKit load or
+   adapter creation, including both loads of config diff and remote session lookup.
+   Use invocation state, not a package-global `sync.Once`. A new command gets its
+   own bootstrap; library config loaders themselves never mutate the environment.
+4. Present safe bootstrap warnings/failures through stderr or the existing safe
+   diagnostic boundary, never into YAML show output. Preserve the partial report
+   on bootstrap failure. Do not add rollback or reinterpret ConfigKit file rules.
+5. Remove late bootstrap calls in object-store and runner composition, the custom
+   directory reader and permission platform files, `pipeline.secrets`, its defaults
+   and validators, and their obsolete tests. Convert affected fixtures immediately.
+
+Acceptance: test absent conventional directory, explicit override from another
+CWD, invalid explicit directory, already-set empty/nonempty environment values,
+and bootstrap failure before adapter construction. A diff invocation and a remote
+session acquisition each bootstrap once. Config inspection requires no session,
+workspace writes, manifests, or runtime clients. Update CLI, config, operations,
+architecture, and troubleshooting to the new bootstrap/permission contract.
+
+## Stage 7 — Resolve typed credentials and protect all output boundaries
+
+Status: Complete — Migrated S3 and Audita credentials to optional ConfigKit
+typed references, passed resolved values directly to runtime adapters, preserved
+the AWS default provider chain when unconfigured, and verified reference-only
+inspection, provenance, generated configuration, manifest, and diagnostic
+boundaries with focused, full, and race test suites.
+
+Read: ConfigKit “Typed secret references” and rendering contracts;
+`internal/config/config.go`, `defaults.go`, `validate_adapters.go`, the safe
+projection from Stage 3, `internal/adapters/storage/s3_backend.go`,
+`internal/adapters/audita/subprocess.go`, and their composition callers.
+
+1. Implement the three renamed optional typed fields and S3 pair policy above.
+   Reject old `_env` schema keys; remove their default-name and lookup helpers.
+   Convert maintained fixtures with synthetic environment setup where references
+   are present. Do not add credentials to unrelated local pipeline fixtures.
+2. Pass resolved S3 values to the SDK adapter's explicit static provider only when
+   the pair is configured. With neither, preserve its default provider chain.
+   Pass Audita the resolved value and safe reference name; deliver the value only
+   through the child environment's existing `AUDITA_LLM_API_KEY` override.
+   Do not re-read the environment for an explicitly resolved reference.
+3. Adapt every serialization, digest, source/diff output, diagnostic, generated
+   invocation config, manifest, and provenance caller to safe reference names.
+   Never YAML-marshal ConfigKit's opaque secret directly or call `Value()` in an
+   inspection path. Secret changes alone must not alter effective digests or
+   invalidate artifact evidence; reference-name changes may change provenance.
+4. Keep ConfigKit errors structured and safe. Domain validation messages must
+   avoid raw credential values; do not unwrap private parser errors or interpolate
+   whole trusted configuration structs into error messages.
+5. Preserve downstream tool ownership for Notarius/Scriptorium credentials that
+   Narratio does not configure. Do not add a parallel credential schema for them.
+
+Acceptance: use distinctive synthetic secrets to prove no leakage in show, sources,
+diff, catalog-ready projections, success/failure diagnostics, manifests, or generated
+Audita configuration. Verify the child receives the credential and an environment
+change after loading does not change explicit delivery. Test S3 paired references,
+partial-pair rejection, missing effective references, and nil SDK fallback offline.
+Update adapter and credential documentation and remove obsolete lookup tests.
+
+## Stage 8 — Infer and normalize artifact dependencies once
+
+Status: Complete — Finalized one normalized artifact graph from explicit ordering
+edges and configured-artifact inputs (including rewritten family members), retained
+edge origins for diagnostics, and verified focused/full tests plus the full race
+suite and an assembled input-only analyze workflow.
+
+Read: `internal/config/validate_scriptorium.go`, `artifact_families.go`,
+`internal/artifactpolicy`, `internal/artifacts`, analyze planning/execution, and
+their artifact dependency and semantic-evidence tests.
+
+1. At the final configuration resolution boundary, union explicit `depends_on`
+   with configured-artifact input references. Use the existing source-ID parser;
+   do not infer edges from filenames, arbitrary strings, or non-artifact inputs.
+2. Infer after family member inputs become concrete `narratio.artifact.<key>`
+   sources. Normalize, deduplicate, and sort the effective dependency list once.
+   Keep explicit ordering-only dependencies valid. Repeated finalization must not
+   accumulate edges or alter ordering. Store source-edge attribution separately
+   from the effective graph for diagnostics and generated provenance.
+3. Apply existing existence, self-reference, enabled-dependency, output, and cycle
+   validation to the union. Remove the error requiring an input to be repeated in
+   `depends_on`. Retain diagnostics pointing to the explicit list entry or the
+   input that contributed an invalid edge, including family origin where relevant.
+4. Ensure execution, targeted analysis, planning, prerequisite closure, reuse,
+   forced regeneration, and semantic evidence consume this same concrete graph.
+   No adapter or command may infer a second graph. Refresh final projection/digest
+   after inference; equivalent explicit and inferred declarations compare equally.
+
+Acceptance: cover input-only dependency ordering, duplicate explicit/inferred edges,
+ordering-only edges, missing/self/cyclic references, and rewritten family inputs.
+Use an assembled fake-artifact workflow to prove upstream output is supplied to a
+dependent artifact and failure/resume retains successful predecessors. Existing
+force/selection and invalidation tests remain valid. Update config and analyze
+internals with the implemented inference rule.
+
+## Stage 9 — Add the effective artifact catalog command
+
+Status: Complete — Added the read-only `config artifacts` catalog over finalized
+configuration, including effective dependencies and definition/family/party
+provenance; documented it and verified focused, full, and race test suites.
+
+Read: `internal/app/config_commands.go`, config diff/source rendering, the safe
+projection and family-origin catalogs, and `docs/cli.md`.
+
+1. Add `narratio config artifacts` to dispatch and help. Reuse ordinary pipeline,
+   profile, campaign, campaign-file, and secrets-directory parsing/resolution.
+   Resolve a campaign when families require it or when explicitly selected; a
+   pipeline with only ordinary artifacts can be inspected without a campaign.
+2. Render one stable record per effective concrete artifact in key order with
+   key, `narratio.artifact.<key>` source ID, defining source location(s), enabled
+   state, output path, and sorted effective dependencies. Family members additionally
+   identify family key, character ID, defining family source, and party source.
+3. Derive identity from the map key. A profile may contribute to an existing
+   artifact, so report contributing definition locations instead of inventing
+   a single filename owner. Use a deterministic escaped tabular text rendering;
+   include a header and no records for an empty catalog. Propagate writer errors.
+4. Use the same finalized state as show/sources/diff and runtime configuration.
+   No session download, adapter initialization, manifest access, or workspace
+   materialization is allowed. Secret bootstrap remains the agreed startup exception.
+
+Acceptance: CLI tests cover ordinary and generated artifacts, disabled entries,
+profile-added/overridden artifacts, empty catalog, no-campaign ordinary inspection,
+required-campaign failure for families, stable ordering, and no runtime side effects.
+Check identity stays the same after moving an artifact file without changing its key.
+Document the command in its canonical CLI owner and link to config layout examples.
+
+## Stage 10 — Finish the operator layout and v2 migration documentation
+
+Status: Complete
+
+Completion: Reworked the maintained bundle around an unprofiled production base,
+one-entry artifact and family imports, and a testing overlay; updated the example
+harness and inspection workflows; consolidated the v2 configuration guidance; and
+added the v2.0.0 operator migration checklist. Focused documentation/example tests,
+the offline inspection walkthrough, and the full normal/race suites pass.
+
+Read: roadmap “Artifact Definition Ergonomics” and “Compatibility and Migration”;
+`examples/README.md`, existing example bundles and their loader test,
+`docs/config.md`, `docs/operations.md`, `docs/release.md`, and internal configuration
+and artifact documentation.
+
+1. Convert maintained pipeline examples to a production base with explicit
+   `configkit.imports`, `artifacts/<key>.yml`, `artifact-families/<key>.yml`, and
+   `profiles/testing.yml`. Each recommended artifact/family file defines exactly
+   one explicit map entry. Fold old production overlays into the base.
+2. Include an input-only dependency example and at least two character-family
+   examples sharing a campaign-owned party roster. Keep supported inline artifacts
+   legal; file layout is guidance, not a filename-based validation requirement.
+3. Update the example validation harness to load root configurations and selected
+   testing profiles, not attempt to load each imported fragment as a full pipeline.
+   Use synthetic environment setup for explicit references. No example includes
+   real credentials or depends on local tools/cloud services to validate.
+4. Consolidate canonical documentation updates from earlier stages. Remove current
+   references to old composition/profile catalogs, default-profile configuration,
+   secrets.env_dir, credential `_env` fields, credential-free inspection guarantees,
+   paired diff snapshots, and removed platform support. Explain explicit-value and
+   null changes without promising reloadable inspection output.
+5. Add `docs/releases/v2.0.0.md` using the release-note structure, with a concrete
+   migration checklist linking to the canonical config/CLI docs and examples.
+   Include static S3-reference migration versus SDK credential discovery, new
+   inspection prerequisites, artifact input dependency inference, and platform
+   removals. Do not claim publication or edit older immutable release notes.
+
+Acceptance: `go test ./internal/doccheck` and
+`go test ./internal/config -run '^TestExamplesLoadAndValidate$'` pass. Walk through
+the new operator examples with config validate/show/sources/diff/artifacts offline.
+Their identities, family members, origins, and testing overrides agree.
+
+## Stage 11 — Verify the complete migration and remove remaining scaffolding
+
+Status: Complete
+
+Completion: Audited the accepted roadmap and the full change set from `1558c9c`;
+no legacy configuration loader, profile/composition traversal, presence store,
+custom secrets loader, or production YAML decoder remains. The migration removes a
+net 1,235 production and 843 test Go lines; remaining configuration code owns domain
+defaults, normalization, validation, expansion, provenance, and safe projections.
+Go 1.26.7 module hygiene, tests, race tests, vet, build, documentation, examples,
+and the v2.0.0 release-candidate checker pass. Linux amd64 was tested natively;
+Linux arm64 and macOS arm64 release assets were cross-built, with native macOS
+arm64 validation remaining best effort and outside the release gate.
+
+Read: the whole feature roadmap, completed stage notes, and the current test and
+release procedures. Review the diff from the commit introducing this plan.
+
+1. Audit every roadmap requirement against current code and executable evidence.
+   Remove temporary migration adapters and any residual generic YAML parser,
+   composer, profile traversal, explicit-presence store, old secrets loader,
+   decoder hooks, or unused tests. `gopkg.in/yaml.v3` may remain for output encoding
+   and unrelated external formats; it must not supply a second configuration loader.
+2. Confirm result-owned snapshots versus final expanded configuration are handled
+   consistently; every inspection/runtime path sees normalized inferred dependencies
+   and generated publish rules. Recheck secret safety and semantic resume boundaries.
+3. Inspect current platform conditions and asset lists. Retain historical notes but
+   no active Windows/macOS amd64 obligation. Compare owned production/test code
+   before and after; report removals and any remaining complexity with reasons.
+   Do not chase a LOC quota by removing useful domain checks or regressions.
+4. Run the repository gates with the maintained Go 1.26 toolchain:
+
+   ```sh
+   go test ./...
+   go test -race ./...
+   go vet ./...
+   go build ./...
+   go test ./internal/doccheck
+   go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+   ```
+
+5. Run `scripts/check-release-candidate.sh v2.0.0` after the release note and all
+   candidate changes exist. This also verifies module hygiene and all three
+   release assets with `GOWORK=off`. Resolve failures; do not bypass checks or
+   publish a tag. Record native versus cross-build evidence accurately.
+
+Acceptance: all gates pass and every roadmap outcome has direct evidence. No
+dependency on the sibling ConfigKit checkout remains. Record completion and any
+best-effort platform limitations, then hand off for a separate release decision.
+
+## Open Questions
+
+None required before implementation. The agreed roadmap and the explicit choices
+above resolve the migration decisions. Upstream credential-free inspection and
+native macOS arm64 CI remain deferred, not prerequisites for this plan.

docs/roadmap/remediation.md

@@ -0,0 +1,331 @@
+# ConfigKit Migration Review — Required Remediations
+
+Reviewed 2026-09-07 at `f7a40af`, against the changes after `1558c9c`, the
+[feature roadmap](configkit-adoption.md), and the completed
+[implementation plan](implementation.md).
+
+The main migration is implemented. The following findings are reproducible gaps
+in the intended behavior, not proposals to restore retired compatibility or
+filesystem restrictions. P2 denotes a correctness issue to fix before release;
+P3 denotes an operator-facing completeness issue also required by the accepted
+plan. No application code was changed during this review.
+
+## 1. Derive explicit publish destinations after family expansion
+
+Status: Resolved — P2
+
+Resolution: Missing destinations are now derived through one shared domain helper
+both during ConfigKit resolution and again after concrete family expansion. Explicit
+destinations remain unchanged, while unknown sources still fail read-only publish
+validation. Regression coverage verifies resolved runtime and effective inspection
+output for generated members, explicit overrides, and unknown sources.
+
+**Evidence:** `internal/config/pipeline_configkit.go:208` derives missing publish
+destinations during the ConfigKit resolver. A concrete family member does not
+exist yet, so the resolver ignores the failed derivation. After family expansion,
+`internal/config/artifact_families.go:141` calls the now-read-only publish
+validator, which rejects the still-empty destination. The previous implementation
+could derive it during that later validation.
+
+**Reproduction:** In an otherwise valid pipeline, define a family `meta` over
+`party.characters`, with output pattern `artifacts/{character_id}.md`. Select a
+valid campaign whose party contains `alice`, and add:
+
+```yaml
+publish:
+  outputs:
+    - source: narratio.artifact.meta_alice
+```
+
+`config validate --config <pipeline> --campaign-file <campaign>` fails with
+`pipeline.publish.outputs[0].dest is required when destination cannot be derived
+from "narratio.artifact.meta_alice"`. Adding `dest: artifacts/alice.md` makes the
+same configuration pass. The destination is therefore available and derivable;
+the lifecycle omits the required final normalization step.
+
+**Required change:** Complete publish destination resolution after concrete
+artifacts exist, before publish validation and final projection/digest generation.
+Share the domain derivation logic and keep validation read-only. Preserve
+explicit destinations and generated family publish-rule collision checks. Do not
+require redundant `dest` fields or revive mutation inside validators.
+
+**Acceptance:** Test an explicit publish source naming a generated member with
+omitted destination, an explicit destination override, and a genuinely unknown
+source. Validate both effective inspection and resolved runtime configuration.
+
+## 2. Stop re-defaulting an explicitly empty Notarius working directory
+
+Status: Resolved — P2
+
+Resolution: Removed both legacy blank-value fallbacks after ConfigKit resolution.
+An omitted working directory is still derived by the presence-aware resolver, an
+explicit relative directory is still rooted at the pipeline, and explicit empty or
+whitespace-only values remain empty and fail enabled-Notarius validation. Existing
+relocation-invariant digest coverage remains green.
+
+**Evidence:** ConfigKit's resolver correctly uses presence-aware defaulting at
+`internal/config/pipeline_configkit.go:180`. However,
+`internal/config/load.go:454` subsequently replaces any blank working directory
+with the config file's directory, irrespective of input presence.
+`internal/config/effective_digest.go` also applies a blank-value fallback while
+capturing logical paths. These retained legacy fallbacks bypass the new rule that
+defaults apply only to absent fields.
+
+**Reproduction:** Set `notarius.enabled: true`, a valid config path, pipeline ID
+and output map, and `notarius.working_directory: ""`. With
+`config_path: notarius/config.yml`, `config show` succeeds and reports
+`working_directory: notarius`; validation never sees the explicit invalid blank.
+
+**Required change:** Let the ConfigKit resolver own the absent-field default.
+The subsequent runtime and logical-path conversions must preserve explicit
+emptiness so enabled-Notarius validation can reject it. Remove duplicate
+value-based defaulting rather than adding another private presence mechanism.
+
+**Acceptance:** Omission still derives the expected directory; explicit empty
+and whitespace-only values fail when Notarius is enabled; a configured relative
+directory resolves from the pipeline root. Preserve digest relocation invariance.
+
+## 3. Enforce unique base definitions even when a profile removes them
+
+Status: Resolved — P2
+
+Resolution: Loading now discovers base declarations through ConfigKit's public
+historical-path enumeration and metadata history from the selected result,
+without separately resolving the base (see finding 7). Duplicate ordinary or
+family definitions therefore fail even when an overlay nulls their container or
+all of Scriptorium. A single base owner
+with a legitimate profile override remains supported.
+
+**Evidence:** `rejectDuplicateBaseDefinitions` at
+`internal/config/pipeline_configkit.go:354` enumerates only artifact/family keys
+remaining in the final typed configuration. It inspects history for those keys
+but never examines base declarations removed by an overlay.
+
+**Reproduction:** Import `first.yml` defining
+`scriptorium.artifacts.recap.prompt_id` and `second.yml` defining
+`scriptorium.artifacts.recap.output_path`. Base validation correctly rejects the
+two-file declaration. Add `profiles/clear.yml` containing:
+
+```yaml
+scriptorium:
+  artifacts: null
+```
+
+The same root with `--profile clear` passes. A selected profile can thus hide the
+base ownership conflict, contrary to the plan's unique base-definition rule.
+This concerns disjoint contributions to one artifact, not ConfigKit's generic
+conflicting-leaf checks, which still work.
+
+**Required change:** Discover the complete base declaration set through public
+ConfigKit metadata/history, including declarations displaced by an overlay, and
+validate ownership independently of final effective-key survival. Keep legitimate
+profile overrides and profile-only definitions supported. Do not reparse source
+YAML or add a second composer.
+
+**Acceptance:** Cover duplicate ordinary and family definitions hidden by a null
+artifact/family container or null Scriptorium section. Base and selected-profile
+loads must reject the same base conflict, while a single base declaration with a
+valid overlay remains accepted.
+
+## 4. Preserve null versus empty collections in effective YAML
+
+Status: Resolved — P2
+
+Resolution: Effective YAML is now emitted from the credential-safe JSON value
+projection through an output-only YAML-node conversion. Explicit nulls, empty maps,
+empty slices, booleans, strings, and exact JSON numbers retain their semantic types.
+Public show/diff tests cover null and empty maps and slices, while digest,
+credential-redaction, deterministic output, and exact-integer coverage remain intact.
+
+**Evidence:** `internal/config/effective_pipeline.go:154` marshals the typed
+projection directly with `yaml.Marshal`. The encoder emits nil slices as `[]`
+and nil maps as `{}`, while the JSON projection used by diff and digest preserves
+them as `null`. The shared input type does not guarantee shared output semantics.
+
+**Reproduction:** Set `audita.modules: null` in a valid base and
+`audita.modules: []` in `profiles/empty.yml`. Both `config show` invocations emit
+`modules: []`, but `config diff default empty` reports:
+
+```text
+changed audita.modules null []
+```
+
+The commands disagree about the effective value. The same renderer also supplies
+the resolved pipeline diagnostic written by prepare.
+
+**Required change:** Render YAML from a credential-safe projection that explicitly
+preserves nulls and empty collections, consistently with the semantic projection.
+Keep exact integers and safe credential reference names. An output-only conversion
+is appropriate; do not restore YAML decoding or a configuration parser. This
+requires truthful inspection, not a guarantee that inspection output is reloadable.
+
+**Acceptance:** Verify null versus empty maps and slices through the public show
+boundary, and agreement with diff/digest distinctions. Retain credential-leakage
+and exact-integer regression coverage.
+
+## 5. Complete provenance for generated artifacts and inferred dependencies
+
+Status: Resolved — P3
+
+Resolution: Generated catalog records now collect the complete base/profile family
+definition set from ConfigKit history. Effective source ownership for materialized
+dependency lists is rebuilt narrowly from retained explicit-list and input-edge
+origins, including mixed and generated-family dependencies. CLI tests verify a
+profiled family catalog agrees with generated source attribution and that inferred
+ordinary edges name their input contributor instead of the central default.
+
+Two new reporting paths lose available declaration information:
+
+- **Profile contributions to family members:**
+  `internal/config/artifact_catalog.go:52` fills a generated member's definitions
+  from the single `member.Source`. That source is chosen by the first matching
+  non-default family field in `artifactFamilySource`, not the full set of effective
+  contributors. Define family `meta` in the base and override its `prompt_id` in
+  `profiles/testing.yml`. `config artifacts --profile testing` reports only the
+  base file for `meta_alice`, even though `config sources` correctly attributes
+  that member's prompt to the testing overlay.
+- **Inferred ordinary dependencies:** With `second.inputs.prior.source` set to
+  `narratio.artifact.first` and no explicit dependency list, `config sources`
+  reports `scriptorium.artifacts.second.depends_on` as `default / default`.
+  `finalizeArtifactDependencies` stores the input-edge origin, but the effective
+  ownership projection retains the pre-inference default attribution from
+  `internal/config/effective_digest.go:55` and does not consult that origin.
+
+**Impact:** Operators inspecting a testing profile or following a dependency can
+be directed to the wrong file or told a user-derived value is a central default.
+Runtime ordering and generated artifact values were correct in these reproductions.
+
+**Required change:** Use ConfigKit's contributing locations for family definition
+sets, as for ordinary artifacts, and add narrow generated attribution for inferred
+dependency lists using their retained edge origins. Preserve family/character/party
+identity and deterministic ordering. Avoid introducing a second general provenance
+store or choosing a file based on the first alphabetically visited field.
+
+**Acceptance:** A profiled family catalog exposes both relevant files; an inferred
+edge points back to its input declaration; mixed explicit/inferred dependencies
+retain their relevant contributors. Verify catalog and sources agree after expansion.
+
+## 6. Surface successful secrets bootstrap reports, including skipped empty variables
+
+Status: Resolved — P3
+
+Resolution: Successful bootstrap reports now flow once through the command
+diagnostics channel, while direct internal command calls default that channel to
+discard. ConfigKit's value-free report makes skipped existing empty variables
+explicit and remains available before later typed-load failures; structured output
+stays isolated on stdout. Command-level coverage verifies successful reporting,
+skipped-empty reporting, value redaction, clean config-show output, and one report
+for a two-profile diff.
+
+**Evidence:** `internal/app/secrets_bootstrap.go:36` discards the report whenever
+`LoadSecretsDir` returns no error. Reports are only rendered when directory
+bootstrap itself fails. This does not implement the roadmap's safe bootstrap
+reporting boundary or Stage 6's warning presentation requirement.
+
+**Reproduction:** Place a synthetic key in `secrets/NARRATIO_REVIEW_KEY`, reference
+it from `audita.llm_api_key`, and invoke config validation with
+`NARRATIO_REVIEW_KEY=` in the environment. ConfigKit correctly skips the file
+because the variable already exists, then typed loading fails with `secret.empty`.
+Narratio shows only that load failure, omitting the bootstrap report's
+`skipped_existing` and `ExistingEmpty` information. A successful load likewise
+discards all bootstrap results.
+
+**Required change:** Retain and render or summarize the safe report once per
+invocation through a diagnostic output channel, including when a later typed load
+fails. Make skipped-existing-empty outcomes explicit. Preserve process environment
+precedence and the existing partial report on failure; keep values out of all
+diagnostics and keep structured/YAML stdout uncontaminated. Do not add bootstrap
+rollback, retries, or repeated directory loading.
+
+**Acceptance:** Command-level tests demonstrate safe success/report output and a
+useful skipped-empty diagnostic, without leaking a synthetic secret or polluting
+show output. A two-profile diff still bootstraps and reports only once.
+
+## 7. Do not resolve unused base credentials when loading a profile
+
+Status: Resolved — P2
+
+Resolution: Upgraded to released ConfigKit v0.1.2 and replaced the extra base load
+with `Metadata.HistoricalPaths()` enumeration of declaration paths in the selected
+result. Ownership checks still inspect base-layer history for both ordinary
+artifacts and families, even when overlays remove them. Regression tests verify
+that a profile can replace an unset production credential with a resolvable testing
+reference or remove it with null. Existing hidden-duplicate and legitimate-override
+tests remain intact. No local dependency replacement or second composer is needed.
+
+Follow-up review: 2026-09-07 at `3ce1981`. The original six reproductions are
+addressed, but the fix for finding 3 introduces this regression.
+
+**Evidence:** `internal/config/pipeline_configkit.go:49-57` loads the base again
+without a profile to check duplicate base definitions. This calls the complete
+typed ConfigKit loader, including secret resolution, domain resolution, and
+validation—not just declaration ownership inspection. Consequently, loading a
+valid testing profile also requires credentials from the unselected base.
+
+**Reproduction:** In an otherwise valid base, set `audita.llm_api_key` to
+`NARRATIO_REVIEW_PROD_KEY`. In `profiles/testing.yml`, replace it with
+`NARRATIO_REVIEW_TEST_KEY`. Leave the production variable unset and supply only
+the testing variable with a synthetic value. `config validate --config <root>
+--profile testing` fails with `secret.unset` at the base credential declaration.
+Supplying both variables makes it pass. A profile that instead sets
+`audita.llm_api_key: null` also fails when the unused production variable is
+unset. Both failures were reproduced through the current CLI.
+
+**Impact:** Operators cannot use an independent testing credential environment
+or remove a production credential requirement through a profile. Commands fail
+despite having all credentials required by their effective configuration. The
+extra full load also unnecessarily repeats configuration processing.
+
+**Required change:** Enforce unique base ownership without requiring the
+unselected base to succeed as a separately resolved runtime configuration.
+Inspect complete declaration history through ConfigKit's public metadata,
+including declarations displaced by overlays. If additional upstream metadata
+support is needed, address that explicitly rather than requiring unused secrets,
+suppressing arbitrary load failures, reparsing YAML, or restoring a second
+composer. Preserve the duplicate-definition protection added for finding 3.
+
+**Acceptance:** Add public-boundary regression tests for a profile replacing an
+unset production credential with an available testing credential, and for a
+profile removing that credential with null. Both must load successfully without
+the production variable. Retain failures for duplicate base ordinary/family
+definitions hidden by null overlays, along with legitimate profile override and
+profile-only definition coverage. Resolve secrets only for the selected effective
+configuration.
+
+## Follow-up verification
+
+After adopting ConfigKit v0.1.2 and completing finding 7, the full
+`scripts/check-release-candidate.sh v2.0.0` check passed: module hygiene,
+formatting/whitespace, uncached tests and race tests, vet, build, documentation,
+maintained examples, and Linux amd64/arm64 plus macOS arm64 release cross-builds.
+`go mod verify` also passed. The new credential regressions failed before the
+ownership-check fix and passed afterward. No local module replacements, release
+tags, or pushes were used.
+
+At `3ce1981`, `go test ./...`, `go test -race ./...`, `go vet ./...`, and
+`go build ./...` passed. The new finding was reproduced with a freshly built CLI,
+temporary configuration files, and synthetic credentials; no external adapters
+or cloud services were invoked. That review pass changed only this review record.
+
+## Original verification and scope
+
+The following checks passed at the reviewed commit:
+
+- `go test ./...`
+- `go test -race ./...`
+- `go vet ./...`
+- `go build ./...`
+- The release asset builder for `v2.0.0`, producing exactly Linux amd64, Linux
+  arm64, and macOS arm64 assets; the Linux amd64 binary reported the expected version.
+
+The existing full suites include documentation and maintained-example checks.
+The findings above were additionally reproduced through a freshly built CLI with
+isolated temporary YAML fixtures and synthetic credentials. No live downstream
+tools or cloud services were invoked. Linux arm64 and macOS arm64 were cross-built,
+not tested natively. No tag was created or pushed.
+
+The migration removes a net 1,235 production Go lines and 843 test Go lines versus
+`1558c9c` (physical diff lines, including comments and whitespace). ConfigKit owns
+generic loading/composition; Narratio retains domain resolution, orchestration,
+safe inspection, and adapter delivery. These findings require completing that
+design, not reversing the simplification.

docs/troubleshooting.md

@@ -0,0 +1,770 @@
+# Troubleshooting
+
+Operational diagnosis guide for common Narratio failures.
+
+## Config file not found
+
+Symptom:
+
+- command fails to resolve `pipeline.yml`, `campaign.yml`, or `session.yml`.
+
+Likely causes:
+
+- missing files in default search paths;
+- wrong campaign selection;
+- omitted explicit flags.
+
+Diagnostics:
+
+```bash
+narratio session plan 2026-04-04
+```
+
+Safe fix:
+
+- pass explicit `--config`, `--campaign` or `--campaign-file`, and `--session`.
+
+Relevant reference: [Configuration discovery](./config.md#discovery-and-selection).
+
+## Session template placeholders rejected
+
+Symptom:
+
+- load error says session file must be concrete or contains `{{ ... }}` placeholders.
+
+Likely cause:
+
+- using template content as runtime session config.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04 --session /path/session.yml
+```
+
+Safe fix:
+
+- generate concrete session YAML with `narratio session init`.
+
+Relevant reference: [Operations: Session Initialization](./operations.md#session-initialization).
+
+## Strict decode or schema validation failure
+
+Symptom:
+
+- unknown field / invalid value error during config load.
+
+Likely cause:
+
+- stale field name, typo, invalid enum, invalid duration/path format, or an
+  unsupported local manifest schema/state marker.
+
+Diagnostics:
+
+```bash
+narratio session plan 2026-04-04 --config /path/pipeline.yml --campaign-file /path/campaign.yml --session /path/session.yml
+```
+
+Safe fix:
+
+- align config with [Configuration](./config.md) and the
+  [maintained examples](../examples/README.md).
+- for a manifest schema error, move the unsupported local state aside and
+  regenerate it; restore a verified current committed snapshot or republish
+  healthy regenerated state when appropriate. Do not add version or stage
+  markers by hand.
+- if an older manifest is rejected for its retired cleanup-obligation field,
+  finish outstanding cleanup with the pre-upgrade binary. Otherwise verify the
+  remote commit, move aside or regenerate the incompatible manifest, and use
+  the cleanup stage. Do not treat manual JSON editing as the normal migration
+  path.
+
+Relevant reference: [Configuration](./config.md).
+
+## Party schema version rejected
+
+Symptom:
+
+- configuration loading reports that `party.schema_version` is required or
+  unsupported.
+
+Likely cause:
+
+- the campaign `party.yml` is an unversioned legacy document, or names a
+  schema other than `narratio.party.v1`.
+
+Safe fix:
+
+- replace the file with one exact, single-document
+  [`narratio.party.v1` party](./integrations/party.md); do not add a version
+  marker to an otherwise incompatible legacy structure. The versioned party is
+  the only configured roster authority.
+
+Relevant reference: [Canonical Party Input](./integrations/party.md).
+
+## Unexpected imported or profile value
+
+Symptom:
+
+- an effective configuration value differs from the root file, or a duplicate
+  ownership/configuration error is hard to locate.
+
+Diagnostics:
+
+```bash
+narratio config sources --config /path/pipeline.yml --profile testing
+```
+
+Add `--campaign` or `--campaign-file` when the pipeline has party-driven
+artifact families. The output identifies each effective logical field's root,
+import, profile, default, campaign, party, or family source without printing
+the field value or credential contents.
+
+Safe fix:
+
+- move a duplicated base field so it has one owner;
+- correct the selected profile or its overlay; or
+- correct the campaign party/family declaration that owns generated values.
+
+To review what would actually change before switching profiles, run `config
+diff` with the same pipeline and campaign selectors. It compares normalized
+effective values rather than YAML formatting or source-file layout.
+
+Relevant reference: [Configuration inspection](./config.md#read-only-effective-pipeline-inspection).
+
+## Audio mode conflict
+
+Symptom:
+
+- validation fails on session audio configuration.
+
+Likely cause:
+
+- configured both local and S3 session audio inputs.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- use local mode (`audio_dir` or `audio_files`) or S3 mode (`audio_s3.prefix`), not both.
+
+Relevant reference: [Session configuration](./config.md#session).
+
+## `--artifacts` selection error
+
+Symptom:
+
+- unknown artifact key or invalid `--artifacts` usage.
+
+Likely causes:
+
+- key not defined in `pipeline.scriptorium.artifacts`;
+- empty list entry (for example trailing comma);
+- `run-stage` used with non-`analyze`/`publish` target.
+
+Diagnostics:
+
+```bash
+narratio session artifacts 2026-04-04
+```
+
+Safe fix:
+
+- provide only configured keys and use `--artifacts` with supported commands/stages.
+
+Relevant reference: [CLI artifact selection](./cli.md).
+
+## Bounded run prerequisite is unusable
+
+Symptom:
+
+- `run` or `session plan` reports that a prerequisite stage is absent or has a
+  pending, running, failed, stale, or interrupted status before the selected
+  start.
+
+Likely cause:
+
+- `--from` excludes upstream work that has not reached the terminal
+  `succeeded` or `skipped` state in the session manifest.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session plan 2026-04-04 --from render --through analyze
+```
+
+Safe fix:
+
+- widen the bounded range to include the first reported stage, or recover that
+  stage explicitly with `run-stage` before retrying. The failed check does not
+  create a run-local directory or modify the manifest. Narratio does not
+  resume-validate excluded prefix stages, and stages after `--through` are not
+  prerequisites.
+
+If prerequisite statuses are terminal but a selected stage reports a missing,
+unsafe, or checksum-inconsistent artifact, repair the artifact at the stage
+that owns it; do not edit the manifest to bypass the selected stage's concrete
+input validation.
+
+Relevant reference: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior).
+
+## Notarius executable missing
+
+Symptom:
+
+- extraction fails while resolving or starting the Notarius executable.
+
+Likely causes:
+
+- `pipeline.notarius.binary` is not installed, executable, or on `PATH`;
+- a configured executable path is wrong.
+
+Safe fix:
+
+- install a compatible Notarius release or correct the binary setting, then
+  rerun extraction.
+
+Relevant references: [Notarius configuration](./config.md#notarius-output-entries)
+and [Notarius integration](./integrations/notarius.md).
+
+## Notarius exits nonzero
+
+Symptom:
+
+- extraction reports a Notarius exit error instead of a receipt.
+
+Diagnostics:
+
+- inspect `runs/{run_id}/extract/notarius.stderr.log`; stdout is reserved for
+  the receipt and is not merged with diagnostics.
+
+Safe fix:
+
+- correct the reported Notarius pipeline, input, provider, or configuration
+  failure and rerun extraction. Do not edit a staged output bundle into place.
+
+After a failed replacement, an older immutable bundle may still exist even
+though the current session manifest has no successful extraction payload. This
+is expected audit state, not a signal to relink the old bundle manually.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Prepared Notarius reference missing or inconsistent
+
+Symptom:
+
+- extraction or resume validation reports that a configured reference source is
+  unavailable, unsafe, empty, or checksum-inconsistent and recommends
+  `prepare --force`.
+
+Likely causes:
+
+- `prepare` has not run since the campaign/session stable input changed;
+- the configured source file is missing;
+- a prepared `inputs/` file or its manifest record was modified independently;
+- a spell-catalog binding exists without an effective `spell_catalog_file`.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- correct the campaign/session input path, then refresh canonical prepared
+  evidence before extraction:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Do not point Notarius directly at the original source path or edit the manifest
+checksum. Relevant references: [Notarius reference configuration](./config.md#notarius-reference-bindings)
+and [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Notarius reference selector or generated-handoff collision
+
+Symptom:
+
+- Notarius exits nonzero with an undeclared reference-slot, incompatible media,
+  or external/generated reference collision error.
+
+Likely causes:
+
+- a selector does not identify a slot declared by the selected Notarius target;
+- a prepared file does not satisfy that slot's Notarius media contract; or
+- a CLI binding attempts to replace a same-run generated D&D handoff.
+
+Safe fix:
+
+- compare external bindings with the selected Notarius pipeline's canonical
+  consumer documentation;
+- keep only campaign-owned external slots on the CLI; and
+- leave registry, scene, combat, and occurrence handoffs to Notarius pipeline
+  composition.
+
+Narratio validates selector structure and prepared evidence, while Notarius
+owns slot declarations, media compatibility, and generated-handoff conflicts.
+Relevant reference: [Notarius integration](./integrations/notarius.md).
+
+## Notarius receipt or index incompatible
+
+Symptom:
+
+- extraction rejects the receipt schema, pipeline identity, bundle/index path,
+  lane descriptor, or payload path even though Notarius exited successfully.
+
+Likely causes:
+
+- Narratio and Notarius versions disagree on their consumer contract;
+- the configured pipeline or lane constraints are stale;
+- output paths escape the bundle or traverse symlinks.
+
+Safe fix:
+
+- compare installed Notarius output with the canonical Notarius contracts,
+  including receipt `index_file: index.json` and index management names
+  `manifest.json`, `rejected.json`, `warnings.json`, and `diagnostics.json`; align
+  `pipeline.notarius` constraints and rerun. Do not bypass confinement or schema
+  checks.
+
+Relevant reference: [Notarius integration](./integrations/notarius.md).
+
+## Required Notarius lane rejected or missing
+
+Symptom:
+
+- extraction fails because a configured lane is rejected, missing, duplicated,
+  or incompatible, including after a zero exit.
+
+Safe fix:
+
+- inspect the Notarius diagnostic log and bundle rejection/warning information;
+- correct the Notarius module or the exact declared lane contract;
+- remove an output declaration only if downstream consumers genuinely no longer
+  require that source, then rerun extraction.
+
+Every configured output is required. Narratio does not promote a partial result.
+
+## Extraction resume invalidated
+
+Symptom:
+
+- a previously successful extraction runs again during ordinary continuation.
+
+Likely causes:
+
+- the executable/config path, pipeline ID, timeout, working directory, or
+  configured output contracts changed;
+- a configured prepared reference selector, source, path, checksum, or byte
+  size changed;
+- the durable bundle, index, lane set, provenance, regular-file status, or
+  checksum no longer validates.
+
+Safe fix:
+
+- allow the automatic rerun after verifying the current configuration. Treat
+  an unsafe path or bundle symlink error as filesystem corruption or tampering
+  and investigate it rather than replacing files manually.
+
+## Notarius transitive configuration changed
+
+Symptom:
+
+- Notarius profiles, prompts, modules, imported files, or references changed,
+  but Narratio still considers the previous extraction resumable.
+
+Safe fix:
+
+```bash
+narratio run-stage extract 2026-04-04 --force
+```
+
+Narratio fingerprints its invocation contract and prepared Narratio reference
+identities, not the contents of other transitive Notarius inputs. Always force
+extraction after changing those external inputs; downstream
+successful stages are then marked stale normally.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Analysis artifact evidence is not current
+
+Symptom:
+
+- ordinary continuation or `session plan` schedules one or more configured
+  artifacts even though a canonical output file exists; or
+- publish reports a configured artifact source unavailable.
+
+Likely causes:
+
+- the per-artifact record is stale, missing, failed, unselected, or malformed;
+- a configured prompt/profile, dependency, input identity, output path, or
+  effective variable changed; or
+- the recorded output is missing, unsafe, empty, or has a size/checksum that no
+  longer matches its manifest evidence.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session artifacts 2026-04-04
+narratio session plan 2026-04-04 --from analyze --through analyze
+```
+
+Safe fix:
+
+- investigate unexpected path or checksum changes as possible tampering;
+- otherwise let the selected analyze work rerun, or explicitly regenerate only
+  the affected targets; and
+- never edit the fingerprint/checksum in the manifest or copy an old file into
+  the canonical path as a substitute for current evidence.
+
+```bash
+narratio analyze 2026-04-04 --artifacts session_recap
+```
+
+Relevant references: [Operations: Artifact Selection](./operations.md#artifact-selection)
+and [Artifact Internals](./internal/artifacts.md#resolution-rules).
+
+## Analysis manifest checkpoint failed
+
+Symptom:
+
+- analyze reports that it could not save the session manifest after completing
+  an artifact.
+
+Likely causes:
+
+- the workspace is unwritable or out of space;
+- the manifest path or its parent directory has unsafe permissions or type; or
+- a transient local filesystem error interrupted atomic manifest replacement.
+
+Safe fix:
+
+- correct the reported filesystem or permission problem;
+- inspect `session artifacts` to identify the last durable current records; and
+- rerun analyze normally. Narratio reuses earlier successful checkpoints and
+  retries the failed or non-current artifact and its required dependents.
+
+Do not treat a newly created canonical artifact file as current when its
+checkpoint failed, and do not edit the manifest to advertise it. If the error
+recurs, preserve the manifest and run-local diagnostics before repairing the
+workspace.
+
+Relevant references: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior)
+and [Manifest Internals](./internal/manifest.md#persistence-semantics).
+
+## Built-in transcript or bounds source is unavailable
+
+Symptom:
+
+- analyze, extract, render, or publish says a built-in source is unavailable
+  even though a file exists at its usual session path.
+
+Likely cause:
+
+- the producing stage has not succeeded, its output lacks the exact registered
+  source, run ID, or checksum, its evidence is duplicated or stale, or the
+  durable bytes no longer match the manifest checksum.
+
+Safe fix:
+
+- do not copy or rename the file to make it discoverable. Rerun the producing
+  stage to regenerate both the bytes and current manifest evidence, then retry
+  the consumer.
+
+## An older manifest schema is rejected
+
+Symptom:
+
+- Narratio rejects a local session manifest because `schema_version` is missing
+  or unsupported; or
+- restore rejects a committed remote session manifest for the same reason.
+
+Likely cause:
+
+- the manifest predates the supported canonical state contract. Narratio
+  rejects it at the schema boundary rather than interpreting aggregate output
+  lists or incidental files as current authority.
+
+Safe fix:
+
+- for unsupported local state, move the old manifest aside and regenerate, or
+  restore a verified current committed snapshot when one is authoritative;
+- for unsupported committed remote state, regenerate healthy local state as
+  needed and republish the session.
+
+```bash
+narratio run 2026-04-04
+narratio publish 2026-04-04
+```
+
+After current records exist, inspect them and publish explicitly. Do not copy
+old files into canonical paths merely to influence selection; availability is
+manifest-owned.
+
+Relevant references: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior)
+and [Manifest Internals](./internal/manifest.md#analyze-owned-artifact-state).
+
+## Scriptorium private input changed without a rerun
+
+Symptom:
+
+- a prompt, profile, imported configuration file, executable, or other input
+  loaded privately by Scriptorium changed, but Narratio still considers an
+  artifact current.
+
+Likely cause:
+
+- analysis fingerprints cover Narratio-observable semantic identities, not
+  executable contents or arbitrary files and transitive configuration that
+  Scriptorium loads behind its configured paths and identifiers.
+
+Safe fix:
+
+- explicitly force the affected target after changing an unobserved private
+  input. Force applies to explicit targets; current prerequisites remain
+  reusable unless selected themselves.
+
+```bash
+narratio analyze 2026-04-04 --artifacts session_recap
+```
+
+Relevant reference: [Analyze Internals](./internal/stage-analyze.md#invariants).
+
+## Previous-session artifact input missing
+
+Symptom:
+
+- prepare/analyze fails due to missing required previous-session artifact cache input.
+
+Likely causes:
+
+- missing `session.previous_session_id`;
+- the selected previous-session commit does not declare the required canonical
+  source as a `published_output`; or
+- the declared published object is unavailable or fails identity validation.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+narratio session status 2026-04-04
+```
+
+Safe fix:
+
+```bash
+narratio session restore 2026-04-04
+```
+
+or rerun prepare after correcting session config:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
+## Session lock conflict (`.lock`)
+
+Symptom:
+
+- command fails acquiring session lock.
+
+Likely causes:
+
+- another process is running for the same session;
+- a process still holds the operating-system lock while it is shutting down.
+
+Diagnostics:
+
+```bash
+ls -l {workspace.root}/work/{campaign}/{session_id}/.lock
+ps aux | grep narratio
+```
+
+Safe fix:
+
+- wait for active process completion;
+- retry after an interrupted holder has exited; the kernel releases its lock
+  even though the `.lock` metadata file remains for inspection.
+
+Relevant reference: [Operations: Local State Layout](./operations.md#local-state-layout).
+
+## Restore conflict without `--force`
+
+Symptom:
+
+- restore fails with conflict count.
+
+Likely cause:
+
+- local durable files differ from remote restore sources.
+- a restore target or one of its ancestors is a symlink or another non-regular
+  filesystem entry.
+
+Diagnostics:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Safe fix:
+
+- review conflicts;
+- rerun with `--force` only when remote state should overwrite an eligible
+  regular local file; repair a symlink or other non-regular conflict first.
+
+If the filesystem changed after the plan was printed, rerun restore after
+repairing the destination; installation performs the same safety check again.
+
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
+## Restore current-state discovery failure
+
+Symptom:
+
+- restore cannot find current pointer or current manifest.
+
+Likely causes:
+
+- no committed publish current state;
+- storage credentials or connectivity failure.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session restore 2026-04-04 --dry-run
+```
+
+Safe fix:
+
+- resolve storage/auth issue;
+- republish from healthy local state if current pointer is missing.
+
+Relevant reference: [Operations: Publish Workflow](./operations.md#publish-workflow).
+
+## Publish output failure
+
+Symptom:
+
+- publish fails on missing required source, upload error, or commit write.
+
+Likely causes:
+
+- required source file not produced;
+- lock/state expectations mismatch;
+- remote storage failure.
+
+Diagnostics:
+
+```bash
+narratio session artifacts 2026-04-04 --remote
+narratio session status 2026-04-04
+narratio run-stage publish 2026-04-04 --force
+```
+
+Safe fix:
+
+- regenerate missing sources by rerunning prerequisite stages;
+- correct publish source/destination rules;
+- retry after storage failure is resolved.
+
+Relevant reference: [Publish configuration](./config.md#publish-configuration-summary).
+
+## Render markdown source missing
+
+Symptom:
+
+- analyze or publish fails because `narratio.transcript.final_markdown` or `narratio.transcript.final_trimmed_markdown` is unavailable.
+
+Likely causes:
+
+- render stage was not executed after transcript changes;
+- render stage failed before producing canonical markdown outputs.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+```
+
+Safe fix:
+
+- rerun render and then retry downstream stage(s):
+
+```bash
+narratio run-stage render 2026-04-04 --force
+narratio run-stage analyze 2026-04-04 --force
+```
+
+Relevant reference: [Operations: Stage Execution](./operations.md#stage-execution-and-continuation-behavior).
+
+## Secrets or storage credential failure
+
+Symptom:
+
+- object-store command fails at initialization/auth.
+
+Likely causes:
+
+- missing, unreadable, or invalid explicit `--secrets-dir`;
+- invalid entries in the conventional `secrets/` directory beside the root pipeline;
+- missing credential environment variables;
+- invalid S3 endpoint/bucket settings.
+
+Diagnostics:
+
+```bash
+ls -la /path/to/secrets_dir
+env | sed 's/=.*//' | grep -E 'OBJECT_STORAGE|AWS|AUDITA|SCRIPTORIUM'
+```
+
+Safe fix:
+
+- correct `--secrets-dir`, secret-file type/name, and operator-managed permissions;
+- provide required env vars;
+- keep secret values out of YAML.
+
+Relevant reference: [Secrets](./config.md#secrets-handling).
+
+## S3 audio prepare failure
+
+Symptom:
+
+- prepare fails listing/downloading session S3 audio.
+
+Likely causes:
+
+- incorrect `session.inputs.audio_s3.prefix`;
+- no matching `.flac` objects;
+- storage connectivity or permissions failure.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- verify prefix contents and storage access;
+- keep session audio mode consistent.
+
+Relevant reference: [Operations](./operations.md).
+
+## References
+
+- [docs/cli.md](./cli.md)
+- [docs/config.md](./config.md)
+- [docs/operations.md](./operations.md)
+- [docs/internal/stage-publish.md](./internal/stage-publish.md)

examples/README.md

@@ -0,0 +1,79 @@
+# Maintained Examples
+
+These files are safe, copyable starting points for Narratio configuration and
+input structure. Replace placeholder identifiers, storage names, integration
+URLs, and paths for the target environment. Field meanings and defaults belong
+in the [configuration reference](../docs/config.md).
+
+## Pipeline Configuration
+
+- [Minimal pipeline](pipeline.minimal.yml): campaign discovery plus the required
+  WhisperX URL.
+- [Production-shaped pipeline](pipeline.production.yml): S3 storage, publish,
+  external tools, and configured Scriptorium artifacts.
+- [Production-base/testing-overlay bundle](production-testing/pipeline.yml):
+  explicit `conf.d` imports, one ordinary artifact per `artifacts/<key>.yml`,
+  one family per `artifact-families/<key>.yml`, and the fixed
+  `profiles/testing.yml` overlay. Production values live in the base. The
+  bundle demonstrates an input-only artifact dependency, two character families
+  sharing the campaign party, and a testing-only disabled artifact. Inspect it
+  offline with:
+
+  ```sh
+  narratio config validate --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml
+  narratio config show --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config sources --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config artifacts --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config diff default testing --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml
+  ```
+
+  These inspection commands load configuration and resolve explicit credential
+  references, but do not load a session or create runtime state. This bundle is
+  intentionally local-storage and credential-reference free so the walkthrough
+  needs no tools or cloud services.
+- [Full annotated pipeline](pipeline.full.annotated.yml): every implemented
+  pipeline section with explanatory comments.
+- [Extraction subset pipeline](pipeline.extraction-subset.yml): a focused
+  Scriptorium artifact consuming only three declared Notarius lanes.
+
+The existing `internal/config` example test loads and validates each pipeline
+with the sample campaign and a compatible local- or S3-audio session.
+
+## Campaign And Session Configuration
+
+- [Sample campaign](campaigns/sample-campaign/campaign.yml), its
+  [session template](campaigns/sample-campaign/session.template.yml), and its
+  adjacent stable inputs provide a complete campaign directory shape.
+- [Local-audio session](session.local-audio.yml) and
+  [S3-audio session](session.s3-audio.yml) are concrete session files.
+- [Session template](session.template.yml) and the campaign-local equivalent
+  demonstrate the narrow placeholder syntax consumed by `session init`; they
+  are templates, not runtime session files.
+
+## Input Fixtures
+
+- [Speakers](speakers.yml), [autocorrect](autocorrect.yml), and
+  [glossary](glossary.yml) show the standalone input shapes.
+- The sample campaign references its local
+  [speakers](campaigns/sample-campaign/speakers.yml),
+  [autocorrect](campaigns/sample-campaign/autocorrect.yml),
+  [glossary](campaigns/sample-campaign/glossary.yml), and canonical
+  [party](campaigns/sample-campaign/party.yml) fixture, plus an optional
+  [spell-catalog overlay](campaigns/sample-campaign/spell_catalog.json) that
+  follows the Notarius v0.6 contract. Narratio derives the players projection
+  from this party source.
+- [Sample speaker audio](audio/sample-speaker.flac) is a text placeholder that
+  reserves the expected filename and directory shape. Replace it with a real
+  FLAC file before running transcription.
+
+The examples contain environment-variable names but no credential values. Set
+`OBJECT_STORAGE_KEY_ID`, `OBJECT_STORAGE_KEY`, and `AUDITA_LLM_API_KEY` to
+nonempty synthetic or real values before loading the production-shaped and full
+annotated pipelines. They use fictional campaign content and reserved example
+domains.
+
+For a deployed bundle, keep credential files out of source control in a
+`secrets/` directory beside the root `pipeline.yml`, one file per referenced
+environment-variable name, or select another directory with `--secrets-dir`.
+Static S3 credentials are optional: omit both S3 credential references to use
+the AWS SDK discovery chain.

examples/campaigns/sample-campaign/autocorrect.yml

@@ -0,0 +1 @@
+[]

examples/campaigns/sample-campaign/campaign.yml

@@ -0,0 +1,8 @@
+campaign_id: sample-campaign
+session_template_file: ./session.template.yml
+inputs:
+  speakers_file: ./speakers.yml
+  autocorrect_file: ./autocorrect.yml
+  glossary_file: ./glossary.yml
+  party_file: ./party.yml
+  spell_catalog_file: ./spell_catalog.json

examples/campaigns/sample-campaign/glossary.yml

@@ -0,0 +1 @@
+[]

examples/campaigns/sample-campaign/party.yml

@@ -0,0 +1,26 @@
+schema_version: narratio.party.v1
+
+characters:
+  arannis:
+    player:
+      name: Rowan Hale
+    character:
+      name: Arannis
+      alias:
+        - Ari
+        - The Grey Owl
+      classes:
+        - name: wizard
+          level: 8
+  brenna:
+    player:
+      name: Rowan Hale
+    character:
+      name: Brenna
+      alias:
+        - Shield of Dawn
+      classes:
+        - name: paladin
+          level: 6
+        - name: warlock
+          level: 2

examples/campaigns/sample-campaign/session.template.yml

@@ -0,0 +1,3 @@
+session_id: "{{ session_id }}"
+inputs:
+  audio_dir: ./audio

examples/campaigns/sample-campaign/speakers.yml

@@ -0,0 +1,5 @@
+match:
+  - speaker: "Example Speaker"
+    match:
+      - "Example_Speaker"
+      - "Example"

examples/campaigns/sample-campaign/spell_catalog.json

@@ -0,0 +1,18 @@
+{
+  "schema_version": "notarius.dnd.spell-catalog-overlay.v1",
+  "catalogs": [
+    {
+      "id": "narratio.sample-campaign",
+      "ruleset": "dnd-5e-2014",
+      "source": {
+        "title": "Narratio sample campaign spell names"
+      },
+      "spells": [
+        {
+          "name": "Aegis of Emberfall",
+          "aliases": ["Emberfall Aegis"]
+        }
+      ]
+    }
+  ]
+}

examples/pipeline.extraction-subset.yml

@@ -0,0 +1,59 @@
+# Purpose-specific extraction example: a Scriptorium session brief consumes
+# only the three Notarius lanes it needs.
+
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+
+whisperx:
+  transcribe_url: "https://transcription.example.com/transcribe"
+
+notarius:
+  enabled: true
+  binary: notarius
+  config_path: /usr/local/etc/notarius/config.yml
+  pipeline_id: dnd-session
+  timeout: 3h
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+  outputs:
+    npc_registry:
+      lane_id: npc-registry
+      media_type: application/json
+      schema_id: notarius.dnd.npc_registry
+      schema_version: v1
+      module_key: dnd/npc-registry
+    location_registry:
+      lane_id: location-registry
+      media_type: application/json
+      schema_id: notarius.dnd.location_registry
+      schema_version: v1
+      module_key: dnd/location-registry
+    scene_descriptions:
+      lane_id: scene-descriptions
+      media_type: application/json
+      schema_id: notarius.dnd.scene_descriptions
+      schema_version: v1
+      module_key: dnd/scene-descriptions
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  artifacts:
+    session_brief:
+      enabled: true
+      prompt_id: dnd.session_brief
+      output_path: artifacts/session_brief.md
+      inputs:
+        npcs:
+          source: narratio.extraction.npc_registry
+          required: true
+        locations:
+          source: narratio.extraction.location_registry
+          required: true
+        scenes:
+          source: narratio.extraction.scene_descriptions
+          required: true

examples/pipeline.full.annotated.yml

@@ -0,0 +1,270 @@
+# Full annotated pipeline example for implemented Narratio config fields.
+# Values are safe placeholders and must be adapted per environment.
+
+workspace:
+  # Optional: defaults to /var/lib/narratio.
+  root: /var/lib/narratio/workspace
+  # Optional: remove run-scoped workdir after successful publish commit.
+  cleanup_after_publish: false
+
+# Credential files, when used, belong in a sibling secrets/ directory or in the
+# directory selected with --secrets-dir; they are not pipeline configuration.
+
+storage:
+  # Defaults to "local". Use "s3" explicitly for publish + S3 audio workflows.
+  backend: s3
+  s3:
+    # Required when using S3 audio or S3 publish uploads.
+    bucket: my-dnd-archive
+    # Optional; defaults to "dnd".
+    root_prefix: dnd
+    # Optional region/endpoint settings.
+    region: us-east-1
+    endpoint: ""
+    force_path_style: false
+    # Optional; defaults shown explicitly.
+    access_key_id: OBJECT_STORAGE_KEY_ID
+    secret_access_key: OBJECT_STORAGE_KEY
+
+campaigns:
+  # Optional; defaults to /usr/local/share/narratio/campaigns.
+  root: /usr/local/share/narratio/campaigns
+  # Optional command default when --campaign is omitted.
+  default_campaign_id: sample-campaign
+
+spool:
+  # Optional; defaults to /var/spool/narratio.
+  root: /var/spool/narratio
+  # Optional cleanup of run-scoped spool audio after successful publish commit.
+  delete_audio_after_publish: false
+
+publish:
+  # Optional booleans; defaults are true.
+  enabled: true
+  upload_run: true
+  # Optional publish output rules; sources use Narratio artifact source IDs.
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+    - source: narratio.artifact.player_handout
+      dest: artifacts/player_handout.md
+      required: false
+    # Extraction lanes publish only when named explicitly; the bundle and index
+    # are never implicit publish sources.
+    - source: narratio.extraction.npc_registry
+      dest: artifacts/extraction/npc-registry.json
+      required: true
+
+whisperx:
+  # Required.
+  transcribe_url: "https://transcription.example.com/transcribe"
+  # Optional overrides; defaults shown explicitly.
+  language: en
+  timeout: 30m
+  retries: 3
+  retry_delay: 2s
+  concurrency: 2
+
+seriatim:
+  # Optional overrides; defaults shown explicitly.
+  binary: seriatim
+  timeout: 10m
+  output_schema: seriatim-intermediate
+  coalesce_gap: 3.0
+  report: true
+  env:
+    # Optional advanced tuning; set only when needed.
+    overlap_word_run_gap: 1.0
+    overlap_word_run_reorder_window: 1.0
+    backchannel_max_duration: 2.0
+    filler_max_duration: 1.25
+
+audita:
+  # Optional overrides; defaults shown explicitly where applicable.
+  binary: audita
+  timeout: 3h
+  llm_api_key: AUDITA_LLM_API_KEY
+  modules: [glossary, homophones, spoken_word, grammar]
+  base_url: ""
+  model: ""
+  total_llm_concurrency: 2
+  proposal_llm_concurrency: 1
+  validation_model: ""
+  validation_llm_concurrency: 1
+  transcript_description: ""
+  config_path: /usr/local/etc/audita/config.yml
+  output_schema: audita-v1
+  work_dir_retention: auto
+  report: true
+
+normalize:
+  # Optional; defaults shown explicitly.
+  output_path: transcripts/final.json
+  output_schema: seriatim-intermediate
+  report: true
+
+trim:
+  # Optional; defaults shown explicitly.
+  enabled: true
+  output_path: transcripts/final.trimmed.json
+  bounds:
+    prompt_id: dnd.session_bounds
+    profile_id: ""
+    transcript_input_name: transcript
+    output_path: artifacts/session_bounds.json
+    timeout: 10m
+    render_debug: false
+  seriatim:
+    report: false
+
+notarius:
+  # Optional structured extraction between trim and render.
+  enabled: true
+  binary: notarius
+  config_path: /usr/local/etc/notarius/config.yml
+  pipeline_id: dnd-session
+  timeout: 3h
+  working_directory: /usr/local/etc/notarius
+  # External campaign references use prepared Narratio source IDs. Omit an
+  # optional binding when the selected Notarius pipeline does not need it.
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+  # Each key creates source narratio.extraction.<key>. These constraints match
+  # the current Notarius D&D lane contracts; update them with Notarius.
+  outputs:
+    item_registry:
+      lane_id: item-registry
+      media_type: application/json
+      schema_id: notarius.dnd.item_registry
+      schema_version: v1
+      module_key: dnd/item-registry
+    npc_registry:
+      lane_id: npc-registry
+      media_type: application/json
+      schema_id: notarius.dnd.npc_registry
+      schema_version: v1
+      module_key: dnd/npc-registry
+    location_registry:
+      lane_id: location-registry
+      media_type: application/json
+      schema_id: notarius.dnd.location_registry
+      schema_version: v1
+      module_key: dnd/location-registry
+    scene_descriptions:
+      lane_id: scene-descriptions
+      media_type: application/json
+      schema_id: notarius.dnd.scene_descriptions
+      schema_version: v1
+      module_key: dnd/scene-descriptions
+    item_occurrences:
+      lane_id: item-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.item_occurrences
+      schema_version: v1
+      module_key: dnd/item-occurrences
+    spells:
+      lane_id: spells
+      media_type: application/json
+      schema_id: notarius.dnd.spells
+      schema_version: v1
+      module_key: dnd/spells
+    combat_turns:
+      lane_id: combat-turns
+      media_type: application/json
+      schema_id: notarius.dnd.combat_turns
+      schema_version: v1
+      module_key: dnd/combat-turns
+    npc_occurrences:
+      lane_id: npc-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.npc_occurrences
+      schema_version: v1
+      module_key: dnd/npc-occurrences
+    location_occurrences:
+      lane_id: location-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.location_occurrences
+      schema_version: v1
+      module_key: dnd/location-occurrences
+    enemy_events:
+      lane_id: enemy-events
+      media_type: application/json
+      schema_id: notarius.dnd.enemy_events
+      schema_version: v1
+      module_key: dnd/enemy-events
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  timeout: 10m
+  render_debug: false
+  artifacts:
+    # Configured artifact keys map to source IDs narratio.artifact.<key>.
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      profile_id: local-fast
+      output_path: artifacts/session_recap.md
+      timeout: 10m
+      inputs:
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+        previous_recap:
+          source: narratio.previous_session.artifact.session_recap
+          required: false
+        players:
+          source: narratio.input.players
+          required: true
+        party:
+          source: narratio.input.party
+          required: true
+        glossary:
+          source: narratio.input.glossary
+          required: false
+      vars:
+        session_id: true
+        session_date: true
+        campaign_name: true
+        previous_session_id: true
+        output_kind: session_recap
+
+    # Example dependent artifact:
+    # - depends_on entries use artifact keys.
+    # - narratio.artifact.<key> sources require matching depends_on membership.
+    player_handout:
+      enabled: true
+      depends_on:
+        - session_recap
+      prompt_id: dnd.player_handout
+      profile_id: local-fast
+      output_path: artifacts/player_handout.md
+      timeout: 10m
+      inputs:
+        recap:
+          source: narratio.artifact.session_recap
+          required: true
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+      vars:
+        session_id: true
+        campaign_name: true
+        output_kind: player_handout
+
+notification:
+  # No delivery provider is currently implemented.
+  mode: noop

examples/pipeline.minimal.yml

@@ -1,109 +1,6 @@
-workspace:
-  root: ./tmp/narratio-workspace
-
-storage:
-  backend: local
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
 
 whisperx:
   transcribe_url: "https://transcription.example.com/transcribe"
-  language: "en"
-  timeout: "30m"
-  retries: 3
-  retry_delay: "2s"
-  concurrency: 2
-
-seriatim:
-  binary: "seriatim"
-  timeout: "10m"
-  output_schema: "seriatim-intermediate"
-  coalesce_gap: 3.0
-  report: true
-  env:
-    overlap_word_run_gap: 1.0
-    overlap_word_run_reorder_window: 1.0
-    backchannel_max_duration: 2.0
-    filler_max_duration: 1.25
-
-audita:
-  binary: "audita"
-  timeout: "3h"
-  llm_api_key_env: "AUDITA_LLM_API_KEY"
-  modules:
-    - glossary
-    - homophones
-    - glossary
-    - spoken_word
-    - grammar
-    - homophones
-    - glossary
-  base_url: "https://openrouter.ai/api/v1"
-  model: "openrouter/google/gemma-4-31b-it"
-  llm_concurrency: 1
-  validation_model: ""
-  validation_llm_concurrency: 1
-  report: true
-
-normalize:
-  # Session-workdir-relative when not absolute.
-  output_path: "transcripts/normalized.json"
-  output_schema: "seriatim-intermediate"
-  report: true
-
-trim:
-  enabled: true
-  # Session-workdir-relative when not absolute.
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    # Empty means use prompt default profile.
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
-
-scriptorium:
-  binary: "scriptorium"
-  config_path: "/etc/scriptorium/config.yml"
-  timeout: "10m"
-  render_debug: false
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: "dnd.session_recap"
-      profile_id: "local-quality"
-      output_path: "artifacts/session_recap.md"
-      timeout: "10m"
-      # Optional per-artifact override of global scriptorium.render_debug.
-      # render_debug: true
-      inputs:
-        transcript:
-          # Available transcript sources:
-          # - trimmed_transcript (recommended for session_recap)
-          # - normalized_transcript (recommended for future full-session analysis)
-          # - processed_transcript (raw Audita-polished output)
-          source: "trimmed_transcript"
-          required: true
-        previous_recap:
-          source: "previous_session_artifact"
-          artifact: "session_recap"
-          # Optional: set when previous recap is available.
-          path: ""
-          required: false
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: "session_recap"
-
-analyzer:
-  timeout: 20m
-  artifacts:
-    output_dir: artifacts
-
-notification:
-  timeout: 10s

examples/pipeline.production.yml

@@ -0,0 +1,128 @@
+workspace:
+  root: /var/lib/narratio/workspace
+  cleanup_after_publish: true
+
+storage:
+  backend: s3
+  s3:
+    bucket: my-dnd-archive
+    root_prefix: dnd
+    region: us-east-1
+    access_key_id: OBJECT_STORAGE_KEY_ID
+    secret_access_key: OBJECT_STORAGE_KEY
+
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+
+spool:
+  root: /var/spool/narratio
+  delete_audio_after_publish: true
+
+publish:
+  enabled: true
+  upload_run: true
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+    - source: narratio.artifact.player_handout
+      dest: artifacts/player_handout.md
+      required: false
+
+whisperx:
+  transcribe_url: "https://transcription.example.com/transcribe"
+  language: en
+  timeout: 45m
+  retries: 3
+  retry_delay: 3s
+  concurrency: 2
+
+seriatim:
+  binary: seriatim
+  timeout: 10m
+  output_schema: seriatim-intermediate
+  coalesce_gap: 3.0
+  report: true
+
+audita:
+  binary: audita
+  timeout: 3h
+  llm_api_key: AUDITA_LLM_API_KEY
+  modules: [glossary, homophones, spoken_word, grammar]
+  output_schema: audita-v1
+  work_dir_retention: auto
+  total_llm_concurrency: 2
+  proposal_llm_concurrency: 1
+  validation_llm_concurrency: 1
+  report: true
+
+normalize:
+  output_path: transcripts/final.json
+  output_schema: seriatim-intermediate
+  report: true
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  timeout: 10m
+  render_debug: false
+  artifacts:
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      profile_id: local-fast
+      output_path: artifacts/session_recap.md
+      timeout: 10m
+      inputs:
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+        previous_recap:
+          source: narratio.previous_session.artifact.session_recap
+          required: false
+        players:
+          source: narratio.input.players
+          required: true
+        party:
+          source: narratio.input.party
+          required: true
+        glossary:
+          source: narratio.input.glossary
+          required: false
+      vars:
+        session_id: true
+        session_date: true
+        campaign_name: true
+        previous_session_id: true
+        output_kind: session_recap
+    player_handout:
+      enabled: true
+      depends_on:
+        - session_recap
+      prompt_id: dnd.player_handout
+      profile_id: local-fast
+      output_path: artifacts/player_handout.md
+      timeout: 10m
+      inputs:
+        recap:
+          source: narratio.artifact.session_recap
+          required: true
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+      vars:
+        session_id: true
+        output_kind: player_handout
+
+notification:
+  mode: noop

examples/production-testing/artifact-families/character_items.yml

@@ -0,0 +1,20 @@
+scriptorium:
+  artifact_families:
+    character_items:
+      enabled: true
+      for_each: party.characters
+      prompt_id: dnd.character_items
+      profile_id: production-placeholder
+      output_path_pattern: artifacts/characters/{character_id}/items.md
+      inputs:
+        character_meta:
+          source: narratio.member_artifact.character_meta
+          required: true
+      member_vars:
+        character_id: character_id
+        character_name: character.name
+        aliases: character.alias_summary
+      publish:
+        enabled: true
+        required: false
+        dest_pattern: artifacts/characters/{character_id}/items.md

examples/production-testing/artifact-families/character_meta.yml

@@ -0,0 +1,13 @@
+scriptorium:
+  artifact_families:
+    character_meta:
+      enabled: true
+      for_each: party.characters
+      prompt_id: dnd.character_meta
+      profile_id: production-placeholder
+      output_path_pattern: artifacts/characters/{character_id}/meta.md
+      member_vars:
+        character_id: character_id
+        character_name: character.name
+        player_name: player.name
+        class_summary: character.class_summary

examples/production-testing/artifacts/player_handout.yml

@@ -0,0 +1,11 @@
+scriptorium:
+  artifacts:
+    player_handout:
+      enabled: true
+      prompt_id: dnd.player_handout
+      profile_id: production-placeholder
+      output_path: artifacts/player-handout.md
+      inputs:
+        recap:
+          source: narratio.artifact.session_recap
+          required: true

examples/production-testing/artifacts/session_recap.yml

@@ -0,0 +1,11 @@
+scriptorium:
+  artifacts:
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      profile_id: production-placeholder
+      output_path: artifacts/session-recap.md
+      inputs:
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true

examples/production-testing/conf.d/platform.yml

@@ -0,0 +1,12 @@
+workspace:
+  root: ./workspace
+
+campaigns:
+  root: ../../campaigns
+  default_campaign_id: sample-campaign
+
+cache:
+  root: ./cache
+
+spool:
+  root: ./spool

examples/production-testing/conf.d/publish.yml

@@ -0,0 +1,7 @@
+publish:
+  enabled: true
+  upload_run: false
+  outputs:
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true

examples/production-testing/conf.d/storage.yml

@@ -0,0 +1,2 @@
+storage:
+  backend: local

examples/production-testing/conf.d/transcript.yml

@@ -0,0 +1,18 @@
+whisperx:
+  transcribe_url: https://transcription.example.com/transcribe
+  language: en
+
+seriatim:
+  binary: seriatim
+  output_schema: seriatim-intermediate
+
+audita:
+  binary: audita
+  modules: [glossary, grammar]
+  output_schema: audita-v1
+  model: narratio-production-model-placeholder
+  validation_model: narratio-production-validator-placeholder
+
+normalize:
+  output_path: transcripts/final.json
+  output_schema: seriatim-intermediate

examples/production-testing/pipeline.yml

@@ -0,0 +1,14 @@
+# Copyable production-base/testing-overlay entry point. ConfigKit expands only
+# these explicit path and glob imports; Narratio never scans directories itself.
+configkit:
+  imports:
+    - path: conf.d/platform.yml
+    - path: conf.d/storage.yml
+    - path: conf.d/transcript.yml
+    - path: conf.d/publish.yml
+    - glob: artifacts/*.yml
+    - glob: artifact-families/*.yml
+
+scriptorium:
+  binary: scriptorium
+  config_path: ./scriptorium/config.yml

examples/production-testing/profiles/testing.yml

@@ -0,0 +1,18 @@
+audita:
+  model: narratio-testing-model-placeholder
+  validation_model: narratio-testing-validator-placeholder
+
+scriptorium:
+  artifacts:
+    session_recap:
+      profile_id: testing-placeholder
+    player_handout:
+      profile_id: testing-placeholder
+    testing_notes:
+      enabled: false
+      output_path: artifacts/testing-notes.md
+  artifact_families:
+    character_meta:
+      profile_id: testing-placeholder
+    character_items:
+      profile_id: testing-placeholder

examples/session.local-audio.yml

@@ -0,0 +1,5 @@
+session_id: 2026-05-03
+date: 2026-05-03
+title: Sample Session
+inputs:
+  audio_dir: ./audio

examples/session.minimal.yml

@@ -1,10 +0,0 @@
-session_id: 2026-05-03
-campaign: sample-campaign
-date: 2026-05-03
-title: Sample Session
-inputs:
-  audio_dir: ./audio
-  speakers_file: ./speakers.yml
-  autocorrect_file: ./autocorrect.yml
-  glossary_file: ./glossary.yml
-

examples/session.s3-audio.yml

@@ -0,0 +1,6 @@
+session_id: 2026-05-03
+date: 2026-05-03
+title: Sample Session
+inputs:
+  audio_s3:
+    prefix: audio/

examples/session.template.yml

@@ -0,0 +1,3 @@
+session_id: "{{ session_id }}"
+inputs:
+  audio_dir: ./audio

examples/speakers.yml

@@ -1,5 +1,5 @@
 match:
-  - speaker: "Eric Rakestraw"
+  - speaker: "Example Speaker"
     match:
-      - "Eric_Rakestraw"
-      - "Eric"
+      - "Example_Speaker"
+      - "Example"

go.mod

@@ -1,5 +1,33 @@
 module gitea.maximumdirect.net/eric/narratio
 
-go 1.25.0
+go 1.26.0
 
-require gopkg.in/yaml.v3 v3.0.1
+require (
+	gitea.maximumdirect.net/eric/ConfigKit v0.1.2
+	github.com/aws/aws-sdk-go-v2/config v1.32.17
+	github.com/aws/aws-sdk-go-v2/credentials v1.19.16
+	github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0
+	github.com/aws/smithy-go v1.25.1
+	github.com/google/go-cmp v0.7.0
+	golang.org/x/sync v0.22.0
+	golang.org/x/sys v0.47.0
+	gopkg.in/yaml.v3 v3.0.1
+)
+
+require (
+	github.com/aws/aws-sdk-go-v2 v1.41.7 // indirect
+	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10 // indirect
+	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23 // indirect
+	github.com/aws/aws-sdk-go-v2/service/signin v1.0.11 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sso v1.30.17 // indirect
+	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 // indirect
+	go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect
+)

go.sum

@@ -1,3 +1,49 @@
+gitea.maximumdirect.net/eric/ConfigKit v0.1.2 h1:uZ5Mvf271SLItXEvUpQYfAvCOVYj9RaoG0auCcbCcSU=
+gitea.maximumdirect.net/eric/ConfigKit v0.1.2/go.mod h1:tYWpsrbLyu05r8+1MapWp5fN6ZrX9I8hU2xEPedYmYk=
+github.com/aws/aws-sdk-go-v2 v1.41.7 h1:DWpAJt66FmnnaRIOT/8ASTucrvuDPZASqhhLey6tLY8=
+github.com/aws/aws-sdk-go-v2 v1.41.7/go.mod h1:4LAfZOPHNVNQEckOACQx60Y8pSRjIkNZQz1w92xpMJc=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10 h1:gx1AwW1Iyk9Z9dD9F4akX5gnN3QZwUB20GGKH/I+Rho=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10/go.mod h1:qqY157uZoqm5OXq/amuaBJyC9hgBCBQnsaWnPe905GY=
+github.com/aws/aws-sdk-go-v2/config v1.32.17 h1:FpL4/758/diKwqbytU0prpuiu60fgXKUWCpDJtApclU=
+github.com/aws/aws-sdk-go-v2/config v1.32.17/go.mod h1:OXqUMzgXytfoF9JaKkhrOYsyh72t9G+MJH8mMRaexOE=
+github.com/aws/aws-sdk-go-v2/credentials v1.19.16 h1:r3RJBuU7X9ibt8RHbMjWE6y60QbKBiII6wSrXnapxSU=
+github.com/aws/aws-sdk-go-v2/credentials v1.19.16/go.mod h1:6cx7zqDENJDbBIIWX6P8s0h6hqHC8Avbjh9Dseo27ug=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23 h1:UuSfcORqNSz/ey3VPRS8TcVH2Ikf0/sC+Hdj400QI6U=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23/go.mod h1:+G/OSGiOFnSOkYloKj/9M35s74LgVAdJBSD5lsFfqKg=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23 h1:GpT/TrnBYuE5gan2cZbTtvP+JlHsutdmlV2YfEyNde0=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23/go.mod h1:xYWD6BS9ywC5bS3sz9Xh04whO/hzK2plt2Zkyrp4JuA=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23 h1:bpd8vxhlQi2r1hiueOw02f/duEPTMK59Q4QMAoTTtTo=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23/go.mod h1:15DfR2nw+CRHIk0tqNyifu3G1YdAOy68RftkhMDDwYk=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24 h1:OQqn11BtaYv1WLUowvcA30MpzIu8Ti4pcLPIIyoKZrA=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24/go.mod h1:X5ZJyfwVrWA96GzPmUCWFQaEARPR7gCrpq2E92PJwAE=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9 h1:FLudkZLt5ci0ozzgkVo8BJGwvqNaZbTWb3UcucAateA=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9/go.mod h1:w7wZ/s9qK7c8g4al+UyoF1Sp/Z45UwMGcqIzLWVQHWk=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15 h1:ieLCO1JxUWuxTZ1cRd0GAaeX7O6cIxnwk7tc1LsQhC4=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15/go.mod h1:e3IzZvQ3kAWNykvE0Tr0RDZCMFInMvhku3qNpcIQXhM=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23 h1:pbrxO/kuIwgEsOPLkaHu0O+m4fNgLU8B3vxQ+72jTPw=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23/go.mod h1:/CMNUqoj46HpS3MNRDEDIwcgEnrtZlKRaHNaHxIFpNA=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23 h1:03xatSQO4+AM1lTAbnRg5OK528EUg744nW7F73U8DKw=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23/go.mod h1:M8l3mwgx5ToK7wot2sBBce/ojzgnPzZXUV445gTSyE8=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0 h1:etqBTKY581iwLL/H/S2sVgk3C9lAsTJFeXWFDsDcWOU=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0/go.mod h1:L2dcoOgS2VSgbPLvpak2NyUPsO1TBN7M45Z4H7DlRc4=
+github.com/aws/aws-sdk-go-v2/service/signin v1.0.11 h1:TdJ+HdzOBhU8+iVAOGUTU63VXopcumCOF1paFulHWZc=
+github.com/aws/aws-sdk-go-v2/service/signin v1.0.11/go.mod h1:R82ZRExE/nheo0N+T8zHPcLRTcH8MGsnR3BiVGX0TwI=
+github.com/aws/aws-sdk-go-v2/service/sso v1.30.17 h1:7byT8HUWrgoRp6sXjxtZwgOKfhss5fW6SkLBtqzgRoE=
+github.com/aws/aws-sdk-go-v2/service/sso v1.30.17/go.mod h1:xNWknVi4Ezm1vg1QsB/5EWpAJURq22uqd38U8qKvOJc=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21 h1:+1Kl1zx6bWi4X7cKi3VYh29h8BvsCoHQEQ6ST9X8w7w=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21/go.mod h1:4vIRDq+CJB2xFAXZ+YgGUTiEft7oAQlhIs71xcSeuVg=
+github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 h1:F/M5Y9I3nwr2IEpshZgh1GeHpOItExNM9L1euNuh/fk=
+github.com/aws/aws-sdk-go-v2/service/sts v1.42.1/go.mod h1:mTNxImtovCOEEuD65mKW7DCsL+2gjEH+RPEAexAzAio=
+github.com/aws/smithy-go v1.25.1 h1:J8ERsGSU7d+aCmdQur5Txg6bVoYelvQJgtZehD12GkI=
+github.com/aws/smithy-go v1.25.1/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
+github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
+github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
+go.yaml.in/yaml/v4 v4.0.0-rc.6 h1:1h7H1ohdUh93/FyE4YaDa1Zh64K6VVbjF4K6WUxMtH4=
+go.yaml.in/yaml/v4 v4.0.0-rc.6/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0=
+golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=

internal/adapters/analyzer/fake.go

@@ -1,40 +0,0 @@
-package analyzer
-
-import "context"
-
-// NoopRunner is a deterministic no-op analyzer adapter.
-type NoopRunner struct{}
-
-// Run returns the requested output path with placeholder metadata.
-func (n *NoopRunner) Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error) {
-	if err := ctx.Err(); err != nil {
-		return AnalyzeResult{}, err
-	}
-	return AnalyzeResult{ArtifactPath: req.OutputPath, Metadata: map[string]any{"placeholder": true}}, nil
-}
-
-// FakeRunner captures analyze requests and returns deterministic responses.
-type FakeRunner struct {
-	Requests []AnalyzeRequest
-	Err      error
-	Result   AnalyzeResult
-}
-
-// Run records request and returns configured response.
-func (f *FakeRunner) Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error) {
-	if err := ctx.Err(); err != nil {
-		return AnalyzeResult{}, err
-	}
-	f.Requests = append(f.Requests, req)
-	if f.Err != nil {
-		return AnalyzeResult{}, f.Err
-	}
-	res := f.Result
-	if res.ArtifactPath == "" {
-		res.ArtifactPath = req.OutputPath
-	}
-	if res.Metadata == nil {
-		res.Metadata = map[string]any{"fake": true}
-	}
-	return res, nil
-}

internal/adapters/analyzer/fake_test.go

@@ -1,31 +0,0 @@
-package analyzer
-
-import (
-	"context"
-	"errors"
-	"testing"
-)
-
-func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
-	fake := &FakeRunner{}
-	req := AnalyzeRequest{ArtifactType: "session-log", OutputPath: "artifacts/session-log.md"}
-
-	res, err := fake.Run(context.Background(), req)
-	if err != nil {
-		t.Fatalf("Run() error = %v", err)
-	}
-	if len(fake.Requests) != 1 || fake.Requests[0].ArtifactType != "session-log" {
-		t.Fatalf("requests = %#v, want captured request", fake.Requests)
-	}
-	if res.ArtifactPath != req.OutputPath {
-		t.Fatalf("artifact path = %q, want %q", res.ArtifactPath, req.OutputPath)
-	}
-}
-
-func TestFakeRunnerError(t *testing.T) {
-	fake := &FakeRunner{Err: errors.New("boom")}
-	_, err := fake.Run(context.Background(), AnalyzeRequest{})
-	if err == nil {
-		t.Fatal("expected error, got nil")
-	}
-}

internal/adapters/analyzer/runner.go

@@ -1,28 +0,0 @@
-// Package analyzer declares the adapter contract for artifact analysis generation.
-package analyzer
-
-import "context"
-
-// TODO: implement analyzer integration once the analyzer contract is finalized.
-
-// Runner is the adapter boundary for analyzer invocations.
-type Runner interface {
-	Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error)
-}
-
-// AnalyzeRequest describes one analyzer artifact generation request.
-type AnalyzeRequest struct {
-	ArtifactType            string
-	ProcessedTranscriptPath string
-	ContextReferences       []string
-	OutputPath              string
-	GeneratedConfigPath     string
-	StdoutLogPath           string
-	StderrLogPath           string
-}
-
-// AnalyzeResult describes analyzer output.
-type AnalyzeResult struct {
-	ArtifactPath string
-	Metadata     map[string]any
-}

internal/adapters/audita/fake.go

@@ -4,10 +4,9 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"os"
 	"path/filepath"
 
-	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op audita adapter.
@@ -84,17 +83,17 @@ func materializePlaceholders(req PolishRequest) error {
 			"merged_transcript_path": req.MergedTranscriptPath,
 			"output_path":            req.OutputProcessedPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("audita noop/fake stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("audita noop/fake stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("audita noop/fake stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("audita noop/fake stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -117,14 +116,14 @@ func writeJSONIfRequested(path string, payload any) error {
 	if path == "" {
 		return nil
 	}
-	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+	if err := fileops.EnsureWorkspaceDirectory(filepath.Dir(path)); err != nil {
 		return fmt.Errorf("create parent directory %q: %w", filepath.Dir(path), err)
 	}
 	data, err := json.Marshal(payload)
 	if err != nil {
 		return fmt.Errorf("marshal placeholder json for %q: %w", path, err)
 	}
-	if err := subprocess.WriteFileAtomic(path, data, 0o644); err != nil {
+	if err := fileops.WriteFileAtomic(path, data, fileops.WorkspaceFileMode); err != nil {
 		return fmt.Errorf("write placeholder json %q: %w", path, err)
 	}
 	return nil

internal/adapters/audita/fake_test.go

@@ -14,7 +14,7 @@ func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := PolishRequest{
 		GeneratedConfigPath: filepath.Join(dir, "config", "audita.yml"),
-		OutputProcessedPath: filepath.Join(dir, "transcripts", "processed.json"),
+		OutputProcessedPath: filepath.Join(dir, "transcripts", "polished.json"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "audita.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "audita.stderr.log"),
 	}

internal/adapters/audita/generated_config.go

@@ -0,0 +1,19 @@
+package audita
+
+import (
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+	"gopkg.in/yaml.v3"
+)
+
+func writeGeneratedYAML(path string, payload any) error {
+	data, err := yaml.Marshal(payload)
+	if err != nil {
+		return fmt.Errorf("marshal generated YAML for %q: %w", path, err)
+	}
+	if err := fileops.WriteFileAtomic(path, data, fileops.WorkspaceFileMode); err != nil {
+		return fmt.Errorf("write generated YAML %q: %w", path, err)
+	}
+	return nil
+}

internal/adapters/audita/runner.go

@@ -6,8 +6,6 @@ import (
 	"time"
 )
 
-// TODO: implement a real Audita subprocess/service adapter.
-
 // Runner is the adapter boundary for audita polish invocations.
 type Runner interface {
 	Run(ctx context.Context, req PolishRequest) (PolishResult, error)
@@ -15,19 +13,15 @@ type Runner interface {
 
 // PolishRequest describes an audita invocation.
 type PolishRequest struct {
-	GeneratedConfigPath      string
-	MergedTranscriptPath     string
-	OutputProcessedPath      string
-	GlossaryPath             string
-	ReportPath               string
-	WorkDir                  string
-	Modules                  []string
-	BaseURL                  string
-	Model                    string
-	ValidationModel          string
-	ValidationLLMConcurrency *int
-	StdoutLogPath            string
-	StderrLogPath            string
+	GeneratedConfigPath  string
+	MergedTranscriptPath string
+	OutputProcessedPath  string
+	GlossaryPath         string
+	ReportPath           string
+	WorkDir              string
+	Modules              []string
+	StdoutLogPath        string
+	StderrLogPath        string
 }
 
 // PolishResult describes a polish output.

internal/adapters/audita/subprocess.go

@@ -5,23 +5,35 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/url"
-	"os"
 	"strconv"
 	"strings"
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
+// MaxProcessedOutputBytes bounds Audita's processed-transcript JSON result.
+const MaxProcessedOutputBytes int64 = 64 * 1024 * 1024
+
+// MaxReportOutputBytes bounds Audita's optional report JSON result.
+const MaxReportOutputBytes int64 = 16 * 1024 * 1024
+
 // SubprocessRunnerConfig defines deterministic settings for Audita CLI execution.
 type SubprocessRunnerConfig struct {
 	Binary                   string
 	Timeout                  time.Duration
-	LLMAPIKeyEnv             string
+	LLMAPIKey                string
+	LLMAPIKeyReference       string
 	Modules                  []string
 	BaseURL                  string
 	Model                    string
-	LLMConcurrency           *int
+	TranscriptDescription    string
+	ConfigPath               string
+	OutputSchema             string
+	WorkDirRetention         string
+	TotalLLMConcurrency      *int
+	ProposalLLMConcurrency   *int
 	ValidationModel          string
 	ValidationLLMConcurrency *int
 	Report                   bool
@@ -31,11 +43,17 @@ type SubprocessRunnerConfig struct {
 type SubprocessRunner struct {
 	binary                   string
 	timeout                  time.Duration
-	llmAPIKeyEnv             string
+	llmAPIKey                string
+	llmAPIKeyReference       string
 	modules                  []string
 	baseURL                  string
 	model                    string
-	llmConcurrency           *int
+	transcriptDescription    string
+	configPath               string
+	outputSchema             string
+	workDirRetention         string
+	totalLLMConcurrency      *int
+	proposalLLMConcurrency   *int
 	validationModel          string
 	validationLLMConcurrency *int
 	report                   bool
@@ -45,11 +63,17 @@ type SubprocessRunner struct {
 func NewSubprocessRunnerFromConfigValues(
 	binary string,
 	timeout string,
-	llmAPIKeyEnv string,
+	llmAPIKey string,
+	llmAPIKeyReference string,
 	modules []string,
 	baseURL string,
 	model string,
-	llmConcurrency *int,
+	transcriptDescription string,
+	configPath string,
+	outputSchema string,
+	workDirRetention string,
+	totalLLMConcurrency *int,
+	proposalLLMConcurrency *int,
 	validationModel string,
 	validationLLMConcurrency *int,
 	report bool,
@@ -64,11 +88,17 @@ func NewSubprocessRunnerFromConfigValues(
 	return NewSubprocessRunner(SubprocessRunnerConfig{
 		Binary:                   binary,
 		Timeout:                  parsedTimeout,
-		LLMAPIKeyEnv:             llmAPIKeyEnv,
+		LLMAPIKey:                llmAPIKey,
+		LLMAPIKeyReference:       llmAPIKeyReference,
 		Modules:                  modules,
 		BaseURL:                  baseURL,
 		Model:                    model,
-		LLMConcurrency:           llmConcurrency,
+		TranscriptDescription:    transcriptDescription,
+		ConfigPath:               configPath,
+		OutputSchema:             outputSchema,
+		WorkDirRetention:         workDirRetention,
+		TotalLLMConcurrency:      totalLLMConcurrency,
+		ProposalLLMConcurrency:   proposalLLMConcurrency,
 		ValidationModel:          validationModel,
 		ValidationLLMConcurrency: validationLLMConcurrency,
 		Report:                   report,
@@ -83,33 +113,42 @@ func NewSubprocessRunner(cfg SubprocessRunnerConfig) (*SubprocessRunner, error)
 	if cfg.Timeout <= 0 {
 		return nil, fmt.Errorf("audita timeout must be > 0")
 	}
-	if len(cfg.Modules) == 0 {
-		return nil, fmt.Errorf("audita modules must include at least one module")
+	if (cfg.LLMAPIKey == "") != (strings.TrimSpace(cfg.LLMAPIKeyReference) == "") {
+		return nil, fmt.Errorf("audita credential value and reference must be provided together")
 	}
 	for i, module := range cfg.Modules {
 		if strings.TrimSpace(module) == "" {
 			return nil, fmt.Errorf("audita module at index %d is empty", i)
 		}
 	}
-	if strings.TrimSpace(cfg.BaseURL) == "" {
-		return nil, fmt.Errorf("audita base url is required")
-	}
-	u, err := url.Parse(cfg.BaseURL)
-	if err != nil || u.Scheme == "" || u.Host == "" {
-		if err != nil {
-			return nil, fmt.Errorf("audita base url %q is invalid: %w", cfg.BaseURL, err)
+	if strings.TrimSpace(cfg.BaseURL) != "" {
+		u, err := url.Parse(cfg.BaseURL)
+		if err != nil || u.Scheme == "" || u.Host == "" {
+			if err != nil {
+				return nil, fmt.Errorf("audita base url %q is invalid: %w", cfg.BaseURL, err)
+			}
+			return nil, fmt.Errorf("audita base url %q is invalid", cfg.BaseURL)
 		}
-		return nil, fmt.Errorf("audita base url %q is invalid", cfg.BaseURL)
 	}
-	if strings.TrimSpace(cfg.Model) == "" {
-		return nil, fmt.Errorf("audita model is required")
+	if cfg.TotalLLMConcurrency != nil && *cfg.TotalLLMConcurrency <= 0 {
+		return nil, fmt.Errorf("audita total llm concurrency must be > 0 when provided")
 	}
-	if cfg.LLMConcurrency != nil && *cfg.LLMConcurrency <= 0 {
-		return nil, fmt.Errorf("audita llm concurrency must be > 0 when provided")
+	if cfg.ProposalLLMConcurrency != nil && *cfg.ProposalLLMConcurrency <= 0 {
+		return nil, fmt.Errorf("audita proposal llm concurrency must be > 0 when provided")
 	}
 	if cfg.ValidationLLMConcurrency != nil && *cfg.ValidationLLMConcurrency <= 0 {
 		return nil, fmt.Errorf("audita validation llm concurrency must be > 0 when provided")
 	}
+	switch strings.TrimSpace(cfg.OutputSchema) {
+	case "", "bare-segments", "audita-v1":
+	default:
+		return nil, fmt.Errorf("audita output schema must be one of: bare-segments, audita-v1")
+	}
+	switch strings.TrimSpace(cfg.WorkDirRetention) {
+	case "", "always", "auto", "never":
+	default:
+		return nil, fmt.Errorf("audita work dir retention must be one of: always, auto, never")
+	}
 
 	modules := make([]string, len(cfg.Modules))
 	for i, m := range cfg.Modules {
@@ -119,11 +158,17 @@ func NewSubprocessRunner(cfg SubprocessRunnerConfig) (*SubprocessRunner, error)
 	return &SubprocessRunner{
 		binary:                   strings.TrimSpace(cfg.Binary),
 		timeout:                  cfg.Timeout,
-		llmAPIKeyEnv:             strings.TrimSpace(cfg.LLMAPIKeyEnv),
+		llmAPIKey:                cfg.LLMAPIKey,
+		llmAPIKeyReference:       strings.TrimSpace(cfg.LLMAPIKeyReference),
 		modules:                  modules,
 		baseURL:                  strings.TrimSpace(cfg.BaseURL),
 		model:                    strings.TrimSpace(cfg.Model),
-		llmConcurrency:           cfg.LLMConcurrency,
+		transcriptDescription:    strings.TrimSpace(cfg.TranscriptDescription),
+		configPath:               strings.TrimSpace(cfg.ConfigPath),
+		outputSchema:             strings.TrimSpace(cfg.OutputSchema),
+		workDirRetention:         strings.TrimSpace(cfg.WorkDirRetention),
+		totalLLMConcurrency:      cfg.TotalLLMConcurrency,
+		proposalLLMConcurrency:   cfg.ProposalLLMConcurrency,
 		validationModel:          strings.TrimSpace(cfg.ValidationModel),
 		validationLLMConcurrency: cfg.ValidationLLMConcurrency,
 		report:                   cfg.Report,
@@ -152,41 +197,31 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 	}
 
 	reqModules := req.Modules
-	if len(reqModules) == 0 {
+	if reqModules == nil {
 		reqModules = append([]string(nil), r.modules...)
 	}
 	args := r.buildArgs(req, reqModules)
 
-	credentialEnvVar := strings.TrimSpace(r.llmAPIKeyEnv)
-	credentialPresent := false
+	credentialPresent := r.llmAPIKeyReference != ""
 	env := map[string]string{}
-	if credentialEnvVar != "" {
-		credential, ok := os.LookupEnv(credentialEnvVar)
-		if !ok || strings.TrimSpace(credential) == "" {
-			return PolishResult{}, fmt.Errorf("audita: required credential environment variable %s is not set", credentialEnvVar)
-		}
-		env["AUDITA_LLM_API_KEY"] = credential
-		credentialPresent = true
-	}
-	primaryConcurrencyViaEnv := false
-	if r.llmConcurrency != nil {
-		env["AUDITA_LLM_CONCURRENCY"] = strconv.Itoa(*r.llmConcurrency)
-		primaryConcurrencyViaEnv = true
+	if credentialPresent {
+		env["AUDITA_LLM_API_KEY"] = r.llmAPIKey
 	}
 
 	if req.GeneratedConfigPath != "" {
-		if err := r.writeInvocationConfig(req, args, reqModules, credentialPresent, primaryConcurrencyViaEnv); err != nil {
+		if err := r.writeInvocationConfig(req, args, reqModules, credentialPresent); err != nil {
 			return PolishResult{}, fmt.Errorf("write audita invocation config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 
 	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    r.binary,
-		Args:          args,
-		Timeout:       r.timeout,
-		EnvOverrides:  env,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:      r.binary,
+		Args:            args,
+		Timeout:         r.timeout,
+		EnvOverrides:    env,
+		DiagnosticOwner: "audita",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
 	})
 	if err != nil {
 		wrappedMessage := fmt.Sprintf(
@@ -196,7 +231,7 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 			req.StderrLogPath,
 		)
 		wrappedMessage = addSubprocessStreamHint(wrappedMessage, err)
-		return r.failureResult(req, reqModules, runRes, credentialPresent, primaryConcurrencyViaEnv), fmt.Errorf(
+		return r.failureResult(req, reqModules, runRes, credentialPresent), fmt.Errorf(
 			"%s: %w",
 			wrappedMessage,
 			err,
@@ -204,11 +239,11 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 	}
 
 	if err := validateProcessedOutput(req.OutputProcessedPath); err != nil {
-		return r.failureResult(req, reqModules, runRes, credentialPresent, primaryConcurrencyViaEnv), fmt.Errorf("validate audita processed output %q: %w", req.OutputProcessedPath, err)
+		return r.failureResult(req, reqModules, runRes, credentialPresent), fmt.Errorf("validate audita processed output %q: %w", req.OutputProcessedPath, err)
 	}
 	if r.report {
 		if err := validateJSONFile(req.ReportPath); err != nil {
-			return r.failureResult(req, reqModules, runRes, credentialPresent, primaryConcurrencyViaEnv), fmt.Errorf("validate audita report output %q: %w", req.ReportPath, err)
+			return r.failureResult(req, reqModules, runRes, credentialPresent), fmt.Errorf("validate audita report output %q: %w", req.ReportPath, err)
 		}
 	}
 
@@ -223,21 +258,25 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 		Duration:                runRes.Duration,
 		InvokedBinary:           r.binary,
 		Metadata: map[string]any{
-			"adapter":                          "audita_subprocess",
-			"modules":                          reqModules,
-			"base_url":                         r.baseURL,
-			"model":                            r.model,
-			"validation_model":                 r.validationModel,
-			"validation_llm_concurrency":       r.validationLLMConcurrency,
-			"credential_env_var":               r.llmAPIKeyEnv,
-			"credential_present":               credentialPresent,
-			"primary_llm_concurrency_via_env":  primaryConcurrencyViaEnv,
-			"primary_llm_concurrency_env_name": "AUDITA_LLM_CONCURRENCY",
+			"adapter":                    "audita_subprocess",
+			"modules":                    reqModules,
+			"base_url":                   r.baseURL,
+			"model":                      r.model,
+			"transcript_description":     r.transcriptDescription,
+			"config_path":                r.configPath,
+			"output_schema":              r.outputSchema,
+			"work_dir_retention":         r.workDirRetention,
+			"validation_model":           r.validationModel,
+			"total_llm_concurrency":      r.totalLLMConcurrency,
+			"proposal_llm_concurrency":   r.proposalLLMConcurrency,
+			"validation_llm_concurrency": r.validationLLMConcurrency,
+			"credential_env_var":         r.llmAPIKeyReference,
+			"credential_present":         credentialPresent,
 		},
 	}, nil
 }
 
-func (r *SubprocessRunner) failureResult(req PolishRequest, modules []string, runRes subprocess.RunResult, credentialPresent bool, primaryConcurrencyViaEnv bool) PolishResult {
+func (r *SubprocessRunner) failureResult(req PolishRequest, modules []string, runRes subprocess.RunResult, credentialPresent bool) PolishResult {
 	return PolishResult{
 		ProcessedTranscriptPath: req.OutputProcessedPath,
 		ReportPath:              req.ReportPath,
@@ -249,16 +288,20 @@ func (r *SubprocessRunner) failureResult(req PolishRequest, modules []string, ru
 		Duration:                runRes.Duration,
 		InvokedBinary:           r.binary,
 		Metadata: map[string]any{
-			"adapter":                          "audita_subprocess",
-			"modules":                          modules,
-			"base_url":                         r.baseURL,
-			"model":                            r.model,
-			"validation_model":                 r.validationModel,
-			"validation_llm_concurrency":       r.validationLLMConcurrency,
-			"credential_env_var":               r.llmAPIKeyEnv,
-			"credential_present":               credentialPresent,
-			"primary_llm_concurrency_via_env":  primaryConcurrencyViaEnv,
-			"primary_llm_concurrency_env_name": "AUDITA_LLM_CONCURRENCY",
+			"adapter":                    "audita_subprocess",
+			"modules":                    modules,
+			"base_url":                   r.baseURL,
+			"model":                      r.model,
+			"transcript_description":     r.transcriptDescription,
+			"config_path":                r.configPath,
+			"output_schema":              r.outputSchema,
+			"work_dir_retention":         r.workDirRetention,
+			"validation_model":           r.validationModel,
+			"total_llm_concurrency":      r.totalLLMConcurrency,
+			"proposal_llm_concurrency":   r.proposalLLMConcurrency,
+			"validation_llm_concurrency": r.validationLLMConcurrency,
+			"credential_env_var":         r.llmAPIKeyReference,
+			"credential_present":         credentialPresent,
 		},
 	}
 }
@@ -269,14 +312,38 @@ func (r *SubprocessRunner) buildArgs(req PolishRequest, modules []string) []stri
 		req.MergedTranscriptPath,
 		"--glossary", req.GlossaryPath,
 		"--output", req.OutputProcessedPath,
-		"--modules", strings.Join(modules, ","),
-		"--base-url", r.baseURL,
-		"--model", r.model,
 		"--work-dir", req.WorkDir,
 	}
+	if r.baseURL != "" {
+		args = append(args, "--base-url", r.baseURL)
+	}
+	if r.model != "" {
+		args = append(args, "--model", r.model)
+	}
+	if len(modules) > 0 {
+		args = append(args, "--modules", strings.Join(modules, ","))
+	}
 	if r.report {
 		args = append(args, "--report-json", req.ReportPath)
 	}
+	if r.transcriptDescription != "" {
+		args = append(args, "--transcript-description", r.transcriptDescription)
+	}
+	if r.configPath != "" {
+		args = append(args, "--config", r.configPath)
+	}
+	if r.outputSchema != "" {
+		args = append(args, "--output-schema", r.outputSchema)
+	}
+	if r.workDirRetention != "" {
+		args = append(args, "--work-dir-retention", r.workDirRetention)
+	}
+	if r.totalLLMConcurrency != nil {
+		args = append(args, "--total-llm-concurrency", strconv.Itoa(*r.totalLLMConcurrency))
+	}
+	if r.proposalLLMConcurrency != nil {
+		args = append(args, "--proposal-llm-concurrency", strconv.Itoa(*r.proposalLLMConcurrency))
+	}
 	if r.validationModel != "" {
 		args = append(args, "--validation-model", r.validationModel)
 	}
@@ -286,37 +353,39 @@ func (r *SubprocessRunner) buildArgs(req PolishRequest, modules []string) []stri
 	return args
 }
 
-func (r *SubprocessRunner) writeInvocationConfig(req PolishRequest, args []string, modules []string, credentialPresent bool, primaryConcurrencyViaEnv bool) error {
+func (r *SubprocessRunner) writeInvocationConfig(req PolishRequest, args []string, modules []string, credentialPresent bool) error {
 	payload := map[string]any{
-		"schema":                          "audita.generated.v1",
-		"binary":                          r.binary,
-		"args":                            args,
-		"timeout":                         r.timeout.String(),
-		"modules":                         modules,
-		"base_url":                        r.baseURL,
-		"model":                           r.model,
-		"validation_model":                r.validationModel,
-		"validation_llm_concurrency":      r.validationLLMConcurrency,
-		"report_enabled":                  r.report,
-		"merged_transcript_path":          req.MergedTranscriptPath,
-		"glossary_path":                   req.GlossaryPath,
-		"output_path":                     req.OutputProcessedPath,
-		"report_path":                     req.ReportPath,
-		"work_dir":                        req.WorkDir,
-		"credential_env_var":              r.llmAPIKeyEnv,
-		"credential_present":              credentialPresent,
-		"primary_llm_concurrency_via_env": primaryConcurrencyViaEnv,
+		"schema":                     "audita.generated.v1",
+		"binary":                     r.binary,
+		"args":                       args,
+		"timeout":                    r.timeout.String(),
+		"modules":                    modules,
+		"base_url":                   r.baseURL,
+		"model":                      r.model,
+		"transcript_description":     r.transcriptDescription,
+		"config_path":                r.configPath,
+		"output_schema":              r.outputSchema,
+		"work_dir_retention":         r.workDirRetention,
+		"validation_model":           r.validationModel,
+		"total_llm_concurrency":      r.totalLLMConcurrency,
+		"proposal_llm_concurrency":   r.proposalLLMConcurrency,
+		"validation_llm_concurrency": r.validationLLMConcurrency,
+		"report_enabled":             r.report,
+		"merged_transcript_path":     req.MergedTranscriptPath,
+		"glossary_path":              req.GlossaryPath,
+		"output_path":                req.OutputProcessedPath,
+		"report_path":                req.ReportPath,
+		"work_dir":                   req.WorkDir,
+		"credential_env_var":         r.llmAPIKeyReference,
+		"credential_present":         credentialPresent,
 	}
-	if r.llmConcurrency != nil {
-		payload["llm_concurrency"] = *r.llmConcurrency
-	}
-	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644)
+	return writeGeneratedYAML(req.GeneratedConfigPath, payload)
 }
 
 func validateProcessedOutput(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readAuditaResult(path, MaxProcessedOutputBytes, "processed transcript")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 
 	var payload map[string]any
@@ -345,9 +414,9 @@ func addSubprocessStreamHint(message string, runErr error) string {
 }
 
 func validateJSONFile(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readAuditaResult(path, MaxReportOutputBytes, "report")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 	var v any
 	if err := json.Unmarshal(data, &v); err != nil {
@@ -355,3 +424,11 @@ func validateJSONFile(path string) error {
 	}
 	return nil
 }
+
+func readAuditaResult(path string, limit int64, category string) ([]byte, error) {
+	data, err := fileops.ReadFileBounded(path, limit)
+	if err != nil {
+		return nil, fmt.Errorf("audita %s result exceeds or cannot be read within %d-byte limit: %w", category, limit, err)
+	}
+	return data, nil
+}

internal/adapters/audita/subprocess_test.go

@@ -6,16 +6,12 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
 func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
@@ -25,16 +21,23 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	wrapper := writeAuditaHelperWrapper(t)
-	llmConcurrency := 1
+	totalLLMConcurrency := 3
+	proposalLLMConcurrency := 2
 	validationLLMConcurrency := 2
 	runner, err := NewSubprocessRunner(SubprocessRunnerConfig{
 		Binary:                   wrapper,
 		Timeout:                  mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:             "OPENAI_KEY_SOURCE",
+		LLMAPIKey:                "super-secret",
+		LLMAPIKeyReference:       "OPENAI_KEY_SOURCE",
 		Modules:                  []string{"glossary", "homophones", "glossary"},
 		BaseURL:                  "https://openrouter.ai/api/v1",
 		Model:                    "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency:           &llmConcurrency,
+		TranscriptDescription:    "Campaign Session 42",
+		ConfigPath:               "/etc/audita/config.yml",
+		OutputSchema:             "audita-v1",
+		WorkDirRetention:         "auto",
+		TotalLLMConcurrency:      &totalLLMConcurrency,
+		ProposalLLMConcurrency:   &proposalLLMConcurrency,
 		ValidationModel:          "openrouter/google/gemma-4-31b-it",
 		ValidationLLMConcurrency: &validationLLMConcurrency,
 		Report:                   true,
@@ -42,13 +45,14 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	if err != nil {
 		t.Fatalf("NewSubprocessRunner() error = %v", err)
 	}
+	t.Setenv("OPENAI_KEY_SOURCE", "changed-after-resolution")
 
 	dir := t.TempDir()
 	req := PolishRequest{
 		GeneratedConfigPath:  filepath.Join(dir, "audita.generated.yml"),
-		MergedTranscriptPath: filepath.Join(dir, "merged.json"),
+		MergedTranscriptPath: filepath.Join(dir, "base.json"),
 		GlossaryPath:         filepath.Join(dir, "glossary.yml"),
-		OutputProcessedPath:  filepath.Join(dir, "processed.json"),
+		OutputProcessedPath:  filepath.Join(dir, "polished.json"),
 		ReportPath:           filepath.Join(dir, "audita.report.json"),
 		WorkDir:              filepath.Join(dir, "artifacts", "audita-work"),
 		StdoutLogPath:        filepath.Join(dir, "audita.stdout.log"),
@@ -93,11 +97,17 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 		"process", req.MergedTranscriptPath,
 		"--glossary", req.GlossaryPath,
 		"--output", req.OutputProcessedPath,
-		"--modules", "glossary,homophones,glossary",
+		"--work-dir", req.WorkDir,
 		"--base-url", "https://openrouter.ai/api/v1",
 		"--model", "openrouter/google/gemma-4-31b-it",
-		"--work-dir", req.WorkDir,
+		"--modules", "glossary,homophones,glossary",
 		"--report-json", req.ReportPath,
+		"--transcript-description", "Campaign Session 42",
+		"--config", "/etc/audita/config.yml",
+		"--output-schema", "audita-v1",
+		"--work-dir-retention", "auto",
+		"--total-llm-concurrency", "3",
+		"--proposal-llm-concurrency", "2",
 		"--validation-model", "openrouter/google/gemma-4-31b-it",
 		"--validation-llm-concurrency", "2",
 	}
@@ -107,8 +117,8 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	if rec.Env["AUDITA_LLM_API_KEY"] != "super-secret" {
 		t.Fatalf("AUDITA_LLM_API_KEY = %q, want propagated secret", rec.Env["AUDITA_LLM_API_KEY"])
 	}
-	if rec.Env["AUDITA_LLM_CONCURRENCY"] != "1" {
-		t.Fatalf("AUDITA_LLM_CONCURRENCY = %q, want 1", rec.Env["AUDITA_LLM_CONCURRENCY"])
+	if rec.Env["AUDITA_LLM_CONCURRENCY"] != "" {
+		t.Fatalf("AUDITA_LLM_CONCURRENCY = %q, want empty/omitted", rec.Env["AUDITA_LLM_CONCURRENCY"])
 	}
 
 	cfgData, err := os.ReadFile(req.GeneratedConfigPath)
@@ -120,51 +130,31 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	}
 }
 
-func TestSubprocessRunnerMissingConfiguredCredentialFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
-	llmConcurrency := 1
-	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "MISSING_AUDITA_KEY",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         false,
+func TestSubprocessRunnerRejectsIncompleteResolvedCredential(t *testing.T) {
+	_, err := NewSubprocessRunner(SubprocessRunnerConfig{
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKeyReference: "MISSING_AUDITA_KEY",
 	})
-	req := auditaReqForTest(t, false)
-
-	_, err := runner.Run(context.Background(), req)
-	if err == nil {
-		t.Fatal("expected error, got nil")
-	}
-	if !strings.Contains(err.Error(), "required credential environment variable MISSING_AUDITA_KEY is not set") {
-		t.Fatalf("error = %q, want missing credential guidance", err.Error())
+	if err == nil || !strings.Contains(err.Error(), "credential value and reference must be provided together") {
+		t.Fatalf("NewSubprocessRunner() error = %v, want credential pair error", err)
 	}
 }
 
 func TestSubprocessRunnerUnconfiguredCredentialEnvOmitsCredential(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         false,
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		Modules:   []string{"glossary"},
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
 	})
 	req := auditaReqForTest(t, false)
 
@@ -181,10 +171,7 @@ func TestSubprocessRunnerUnconfiguredCredentialEnvOmitsCredential(t *testing.T)
 	}
 }
 
-func TestSubprocessRunnerInheritsParentEnvironment(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
+func TestSubprocessRunnerOmitsUnspecifiedParentEnvironment(t *testing.T) {
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	t.Setenv("AUDITA_INHERITED_MARKER", "inherited-from-parent")
@@ -192,13 +179,13 @@ func TestSubprocessRunnerInheritsParentEnvironment(t *testing.T) {
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
-		LLMAPIKeyEnv: "",
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		Modules:   []string{"glossary"},
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
+		LLMAPIKey: "",
 	})
 	req := auditaReqForTest(t, false)
 	if _, err := runner.Run(context.Background(), req); err != nil {
@@ -206,30 +193,79 @@ func TestSubprocessRunnerInheritsParentEnvironment(t *testing.T) {
 	}
 
 	rec := readAuditaHelperRecord(t, recordPath)
-	if rec.Env["AUDITA_INHERITED_MARKER"] != "inherited-from-parent" {
-		t.Fatalf("AUDITA_INHERITED_MARKER = %q, want inherited-from-parent", rec.Env["AUDITA_INHERITED_MARKER"])
+	if rec.Env["AUDITA_INHERITED_MARKER"] != "" {
+		t.Fatalf("AUDITA_INHERITED_MARKER = %q, want omitted from the child environment", rec.Env["AUDITA_INHERITED_MARKER"])
+	}
+}
+
+func TestSubprocessRunnerOmitsModulesFlagWhenNotConfigured(t *testing.T) {
+	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
+	t.Setenv("AUDITA_HELPER_MODE", "success")
+	recordPath := filepath.Join(t.TempDir(), "record.json")
+	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
+
+	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
+	})
+	req := auditaReqForTest(t, false)
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	rec := readAuditaHelperRecord(t, recordPath)
+	for i := 0; i < len(rec.Args); i++ {
+		if rec.Args[i] == "--modules" {
+			t.Fatalf("args contained --modules unexpectedly: %#v", rec.Args)
+		}
+	}
+}
+
+func TestSubprocessRunnerOmitsBaseURLAndModelFlagsWhenNotConfigured(t *testing.T) {
+	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
+	t.Setenv("AUDITA_HELPER_MODE", "success")
+	recordPath := filepath.Join(t.TempDir(), "record.json")
+	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
+
+	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		Report:    false,
+	})
+	req := auditaReqForTest(t, false)
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	rec := readAuditaHelperRecord(t, recordPath)
+	for i := 0; i < len(rec.Args); i++ {
+		if rec.Args[i] == "--base-url" {
+			t.Fatalf("args contained --base-url unexpectedly: %#v", rec.Args)
+		}
+		if rec.Args[i] == "--model" {
+			t.Fatalf("args contained --model unexpectedly: %#v", rec.Args)
+		}
 	}
 }
 
 func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "fail")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -245,27 +281,26 @@ func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
 	if !strings.Contains(err.Error(), req.StdoutLogPath) || !strings.Contains(err.Error(), req.StderrLogPath) {
 		t.Fatalf("error = %q, want stdout/stderr log paths", err.Error())
 	}
+	if strings.Contains(err.Error(), "super-secret") {
+		t.Fatalf("failure diagnostic exposed credential value: %q", err.Error())
+	}
 }
 
 func TestSubprocessRunnerSubprocessFailureAddsStderrDescriptorHint(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "fail_bad_fd")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -278,24 +313,20 @@ func TestSubprocessRunnerSubprocessFailureAddsStderrDescriptorHint(t *testing.T)
 }
 
 func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "missing_output")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -308,24 +339,20 @@ func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "invalid_output")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -338,24 +365,20 @@ func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerSegmentsMissingFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "segments_missing")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -368,24 +391,20 @@ func TestSubprocessRunnerSegmentsMissingFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "invalid_report")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
-	llmConcurrency := 1
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:         writeAuditaHelperWrapper(t),
-		Timeout:        mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:   "OPENAI_KEY_SOURCE",
-		Modules:        []string{"glossary"},
-		BaseURL:        "https://openrouter.ai/api/v1",
-		Model:          "openrouter/google/gemma-4-31b-it",
-		LLMConcurrency: &llmConcurrency,
-		Report:         true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -398,21 +417,22 @@ func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerConstructorValidation(t *testing.T) {
-	_, err := NewSubprocessRunnerFromConfigValues("", "3h", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", nil, "", nil, true)
+	_, err := NewSubprocessRunnerFromConfigValues("", "3h", "synthetic-secret", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
 	if err == nil {
 		t.Fatal("expected binary validation error")
 	}
-	_, err = NewSubprocessRunnerFromConfigValues("audita", "bad", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", nil, "", nil, true)
+	_, err = NewSubprocessRunnerFromConfigValues("audita", "bad", "synthetic-secret", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
 	if err == nil {
 		t.Fatal("expected timeout parse error")
 	}
 	_, err = NewSubprocessRunner(SubprocessRunnerConfig{
-		Binary:       "audita",
-		Timeout:      mustParseAuditaDuration(t, "1s"),
-		LLMAPIKeyEnv: "AUDITA_LLM_API_KEY",
-		Modules:      []string{"glossary"},
-		BaseURL:      "://",
-		Model:        "openrouter/google/gemma-4-31b-it",
+		Binary:             "audita",
+		Timeout:            mustParseAuditaDuration(t, "1s"),
+		LLMAPIKey:          "synthetic-secret",
+		LLMAPIKeyReference: "AUDITA_LLM_API_KEY",
+		Modules:            []string{"glossary"},
+		BaseURL:            "://",
+		Model:              "openrouter/google/gemma-4-31b-it",
 	})
 	if err == nil {
 		t.Fatal("expected base url validation error")
@@ -516,7 +536,7 @@ func mustAuditaRunner(t *testing.T, cfg SubprocessRunnerConfig) *SubprocessRunne
 func auditaReqForTest(t *testing.T, withReport bool) PolishRequest {
 	t.Helper()
 	dir := t.TempDir()
-	merged := filepath.Join(dir, "merged.json")
+	merged := filepath.Join(dir, "base.json")
 	glossary := filepath.Join(dir, "glossary.yml")
 	writeAuditaTestFile(t, merged, `{"segments":[]}`)
 	writeAuditaTestFile(t, glossary, "terms: []\n")
@@ -524,7 +544,7 @@ func auditaReqForTest(t *testing.T, withReport bool) PolishRequest {
 		GeneratedConfigPath:  filepath.Join(dir, "audita.generated.yml"),
 		MergedTranscriptPath: merged,
 		GlossaryPath:         glossary,
-		OutputProcessedPath:  filepath.Join(dir, "processed.json"),
+		OutputProcessedPath:  filepath.Join(dir, "polished.json"),
 		WorkDir:              filepath.Join(dir, "artifacts", "audita-work"),
 		StdoutLogPath:        filepath.Join(dir, "audita.stdout.log"),
 		StderrLogPath:        filepath.Join(dir, "audita.stderr.log"),

internal/adapters/notarius/fake.go

@@ -0,0 +1,24 @@
+package notarius
+
+import "context"
+
+// FakeRunner is a configurable in-memory runner for stage tests.
+type FakeRunner struct {
+	Requests []RunRequest
+	Result   RunResult
+	Err      error
+}
+
+// Run records the request and returns the configured result or error.
+func (f *FakeRunner) Run(ctx context.Context, req RunRequest) (RunResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RunResult{}, err
+	}
+	copyRequest := req
+	copyRequest.References = append([]ReferenceBinding(nil), req.References...)
+	f.Requests = append(f.Requests, copyRequest)
+	if f.Err != nil {
+		return RunResult{}, f.Err
+	}
+	return f.Result, nil
+}

internal/adapters/notarius/runner.go

@@ -0,0 +1,160 @@
+// Package notarius declares the adapter contract for Notarius CLI invocations.
+package notarius
+
+import (
+	"context"
+	"time"
+)
+
+const ReceiptSchemaVersion = "notarius.run-result.v2"
+
+// Runner is the adapter boundary for a complete Notarius pipeline invocation.
+type Runner interface {
+	Run(ctx context.Context, req RunRequest) (RunResult, error)
+}
+
+// ReferenceBinding maps one normalized Notarius selector to an absolute
+// external reference path.
+type ReferenceBinding struct {
+	Selector string
+	Path     string
+}
+
+// RunRequest contains the resolved inputs and diagnostic destinations for one invocation.
+type RunRequest struct {
+	Binary           string
+	ConfigPath       string
+	PipelineID       string
+	InputPath        string
+	OutputRoot       string
+	WorkingDirectory string
+	ReceiptPath      string
+	LogPath          string
+	Timeout          time.Duration
+	References       []ReferenceBinding
+}
+
+// Receipt is the transport-neutral successful run receipt.
+type Receipt struct {
+	SchemaVersion             string
+	RunID                     string
+	PipelineID                string
+	OutputDirectory           string
+	IndexFile                 string
+	NormalizedOutputCount     int
+	RejectedOutputCount       int
+	WarningGroupCount         int
+	WarningOccurrenceCount    int
+	DiagnosticGroupCount      int
+	DiagnosticOccurrenceCount int
+	DiagnosticsTruncated      bool
+	ValidationStatus          string
+	ValidationSummaries       []ValidationSummary
+	DebugDirectory            string
+}
+
+// ValidationSummary retains the bounded outcome of one Notarius producer result.
+type ValidationSummary struct {
+	Stage                string
+	StepID               string
+	LaneID               string
+	ModuleKey            string
+	ChunkID              string
+	Status               string
+	RejectingValidators  []string
+	ReasonCodes          []string
+	IncompleteValidators []string
+	ProducerAttemptCount int
+	TerminalAction       string
+}
+
+// LaneDescriptor identifies one normalized lane payload discovered through the index.
+type LaneDescriptor struct {
+	LaneID        string
+	File          string
+	Path          string
+	MediaType     string
+	ModuleKey     string
+	SchemaID      string
+	SchemaName    string
+	SchemaVersion string
+}
+
+// PipelineDescriptor identifies a pipeline-wide artifact discovered through the index.
+type PipelineDescriptor struct {
+	ArtifactKind  string
+	File          string
+	Path          string
+	MediaType     string
+	SchemaID      string
+	SchemaName    string
+	SchemaVersion string
+}
+
+// Index describes the validated bundle-management and artifact paths.
+type Index struct {
+	Path            string
+	ManifestFile    string
+	ManifestPath    string
+	RejectedFile    string
+	RejectedPath    string
+	WarningsFile    string
+	WarningsPath    string
+	DiagnosticsFile string
+	DiagnosticsPath string
+	Lanes           []LaneDescriptor
+	ChunkMap        *PipelineDescriptor
+	EvidenceContext *PipelineDescriptor
+}
+
+// RejectionSummary retains structured rejection identity without free-form messages.
+type RejectionSummary struct {
+	Stage         string
+	StepID        string
+	LaneID        string
+	ModuleKey     string
+	ChunkID       string
+	ValidatorName string
+	ReasonCode    string
+}
+
+// WarningSummary retains structured warning identity without free-form messages.
+type WarningSummary struct {
+	Disposition     string
+	Category        string
+	ReasonCode      string
+	Origin          DiagnosticOrigin
+	OccurrenceCount int
+}
+
+// DiagnosticOrigin identifies the framework-owned pipeline location of a finding.
+type DiagnosticOrigin struct {
+	Stage        string
+	StepID       string
+	LaneID       string
+	ModuleKey    string
+	ValidatorKey string
+}
+
+// DiagnosticSummary retains bounded advisory or observation group metadata.
+type DiagnosticSummary struct {
+	Disposition     string
+	Category        string
+	ReasonCode      string
+	Origin          DiagnosticOrigin
+	OccurrenceCount int
+}
+
+// RunResult describes a successfully decoded and validated Notarius bundle.
+type RunResult struct {
+	Receipt     Receipt
+	Index       Index
+	BundleRoot  string
+	ReceiptPath string
+	LogPath     string
+	ExitCode    int
+	Duration    time.Duration
+	Rejections  []RejectionSummary
+	Warnings    []WarningSummary
+	Diagnostics []DiagnosticSummary
+}

internal/adapters/notarius/subprocess.go

@@ -0,0 +1,784 @@
+package notarius
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+	"gitea.maximumdirect.net/eric/narratio/internal/notariusref"
+	"gitea.maximumdirect.net/eric/narratio/internal/pathsafe"
+)
+
+const (
+	maxReceiptBytes          = 1 << 20
+	maxIndexBytes            = 4 << 20
+	maxSummaryBytes          = 4 << 20
+	canonicalIndexFile       = "index.json"
+	canonicalManifestFile    = "manifest.json"
+	canonicalRejectedFile    = "rejected.json"
+	canonicalWarningsFile    = "warnings.json"
+	canonicalDiagnosticsFile = "diagnostics.json"
+	warningsSchemaVersion    = "notarius.warnings.v2"
+	diagnosticsSchemaVersion = "notarius.diagnostics.v1"
+	maxWarningGroups         = 128
+	maxDiagnosticGroups      = 256
+	maxFindingSamples        = 3
+)
+
+type subprocessRun func(context.Context, subprocess.RunRequest) (subprocess.RunResult, error)
+
+// SubprocessRunner invokes Notarius through its public CLI.
+type SubprocessRunner struct {
+	run subprocessRun
+}
+
+// NewSubprocessRunner constructs a production Notarius subprocess runner.
+func NewSubprocessRunner() *SubprocessRunner {
+	return &SubprocessRunner{run: subprocess.Run}
+}
+
+// Run executes a complete Notarius pipeline and discovers its published bundle.
+func (r *SubprocessRunner) Run(ctx context.Context, req RunRequest) (RunResult, error) {
+	if r == nil || r.run == nil {
+		return RunResult{}, fmt.Errorf("notarius subprocess runner is nil")
+	}
+	references, err := validateRunRequest(req)
+	if err != nil {
+		return RunResult{}, err
+	}
+
+	args := []string{
+		"run", req.PipelineID,
+		"--config", req.ConfigPath,
+		"--input", req.InputPath,
+		"--output-dir", req.OutputRoot,
+	}
+	for _, reference := range references {
+		args = append(args, "--reference", reference.Selector+"="+reference.Path)
+	}
+	args = append(args, "--json")
+	processResult, err := r.run(ctx, subprocess.RunRequest{
+		Executable:      req.Binary,
+		Args:            args,
+		WorkingDir:      req.WorkingDirectory,
+		Timeout:         req.Timeout,
+		DiagnosticOwner: "notarius",
+		StdoutLogPath:   req.ReceiptPath,
+		StderrLogPath:   req.LogPath,
+	})
+	baseResult := RunResult{
+		ReceiptPath: req.ReceiptPath,
+		LogPath:     req.LogPath,
+		ExitCode:    processResult.ExitCode,
+		Duration:    processResult.Duration,
+	}
+	if err != nil {
+		return baseResult, fmt.Errorf("run notarius pipeline %q: %w", req.PipelineID, err)
+	}
+
+	receipt, err := loadReceipt(req.ReceiptPath, req.PipelineID)
+	if err != nil {
+		return baseResult, err
+	}
+	bundleRoot, err := validateBundleRoot(req.OutputRoot, receipt.OutputDirectory)
+	if err != nil {
+		return baseResult, err
+	}
+	indexPath, err := resolveRegularFile(bundleRoot, receipt.IndexFile)
+	if err != nil {
+		return baseResult, fmt.Errorf("resolve receipt index file: %w", err)
+	}
+	index, err := loadIndex(bundleRoot, indexPath)
+	if err != nil {
+		return baseResult, err
+	}
+	rejections, err := loadRejections(index.RejectedPath)
+	if err != nil {
+		return baseResult, err
+	}
+	warnings, err := loadWarnings(index.WarningsPath)
+	if err != nil {
+		return baseResult, err
+	}
+	diagnostics, diagnosticOccurrences, diagnosticsTruncated, err := loadDiagnostics(index.DiagnosticsPath)
+	if err != nil {
+		return baseResult, err
+	}
+	if receipt.NormalizedOutputCount != len(index.Lanes) || receipt.RejectedOutputCount != len(rejections) ||
+		receipt.WarningGroupCount != len(warnings) || receipt.DiagnosticGroupCount != len(diagnostics) {
+		return baseResult, fmt.Errorf("notarius receipt counts do not match published bundle")
+	}
+	warningOccurrences, err := sumWarningOccurrences(warnings)
+	if err != nil {
+		return baseResult, err
+	}
+	if receipt.WarningOccurrenceCount != warningOccurrences ||
+		receipt.DiagnosticOccurrenceCount != diagnosticOccurrences ||
+		receipt.DiagnosticsTruncated != diagnosticsTruncated {
+		return baseResult, fmt.Errorf("notarius receipt occurrence counts do not match published bundle")
+	}
+
+	baseResult.Receipt = receipt
+	baseResult.Index = index
+	baseResult.BundleRoot = bundleRoot
+	baseResult.Rejections = rejections
+	baseResult.Warnings = warnings
+	baseResult.Diagnostics = diagnostics
+	return baseResult, nil
+}
+
+func validateRunRequest(req RunRequest) ([]ReferenceBinding, error) {
+	if strings.TrimSpace(req.Binary) == "" {
+		return nil, fmt.Errorf("notarius binary is required")
+	}
+	if strings.TrimSpace(req.PipelineID) == "" {
+		return nil, fmt.Errorf("notarius pipeline id is required")
+	}
+	if req.Timeout <= 0 {
+		return nil, fmt.Errorf("notarius timeout must be positive")
+	}
+	for label, path := range map[string]string{
+		"config":            req.ConfigPath,
+		"input":             req.InputPath,
+		"output root":       req.OutputRoot,
+		"working directory": req.WorkingDirectory,
+		"receipt":           req.ReceiptPath,
+		"log":               req.LogPath,
+	} {
+		if strings.TrimSpace(path) == "" {
+			return nil, fmt.Errorf("notarius %s path is required", label)
+		}
+		if !filepath.IsAbs(path) {
+			return nil, fmt.Errorf("notarius %s path must be absolute", label)
+		}
+	}
+	if filepath.Clean(req.ReceiptPath) == filepath.Clean(req.LogPath) {
+		return nil, fmt.Errorf("notarius receipt and log paths must be different")
+	}
+	references := make([]ReferenceBinding, 0, len(req.References))
+	selectors := make(map[string]struct{}, len(req.References))
+	for index, binding := range req.References {
+		selector, err := notariusref.NormalizeSelector(binding.Selector)
+		if err != nil {
+			return nil, fmt.Errorf("notarius reference %d selector: %w", index, err)
+		}
+		if _, duplicate := selectors[selector]; duplicate {
+			return nil, fmt.Errorf("notarius reference selector %q is duplicated", selector)
+		}
+		selectors[selector] = struct{}{}
+		if strings.TrimSpace(binding.Path) == "" {
+			return nil, fmt.Errorf("notarius reference %q path is required", selector)
+		}
+		if !filepath.IsAbs(binding.Path) {
+			return nil, fmt.Errorf("notarius reference %q path must be absolute", selector)
+		}
+		references = append(references, ReferenceBinding{Selector: selector, Path: binding.Path})
+	}
+	if err := requireRegularFile(req.ConfigPath); err != nil {
+		return nil, fmt.Errorf("validate notarius config path: %w", err)
+	}
+	if err := requireRegularFile(req.InputPath); err != nil {
+		return nil, fmt.Errorf("validate notarius input path: %w", err)
+	}
+	if err := requireDirectory(req.OutputRoot); err != nil {
+		return nil, fmt.Errorf("validate notarius output root: %w", err)
+	}
+	if err := requireDirectory(req.WorkingDirectory); err != nil {
+		return nil, fmt.Errorf("validate notarius working directory: %w", err)
+	}
+	if err := validateLogDestination(req.ReceiptPath); err != nil {
+		return nil, fmt.Errorf("validate notarius receipt path: %w", err)
+	}
+	if err := validateLogDestination(req.LogPath); err != nil {
+		return nil, fmt.Errorf("validate notarius log path: %w", err)
+	}
+	return references, nil
+}
+
+type receiptDocument struct {
+	SchemaVersion             string                      `json:"schema_version"`
+	RunID                     string                      `json:"run_id"`
+	PipelineID                string                      `json:"pipeline_id"`
+	OutputDirectory           string                      `json:"output_directory"`
+	IndexFile                 string                      `json:"index_file"`
+	NormalizedOutputCount     *int                        `json:"normalized_output_count"`
+	RejectedOutputCount       *int                        `json:"rejected_output_count"`
+	WarningGroupCount         *int                        `json:"warning_group_count"`
+	WarningOccurrenceCount    *int                        `json:"warning_occurrence_count"`
+	DiagnosticGroupCount      *int                        `json:"diagnostic_group_count"`
+	DiagnosticOccurrenceCount *int                        `json:"diagnostic_occurrence_count"`
+	DiagnosticsTruncated      *bool                       `json:"diagnostics_truncated"`
+	ValidationStatus          string                      `json:"validation_status"`
+	ValidationSummaries       []validationSummaryDocument `json:"validation_summaries"`
+	DebugDirectory            string                      `json:"debug_directory"`
+}
+
+type validationSummaryDocument struct {
+	Stage                string   `json:"stage"`
+	StepID               string   `json:"step_id"`
+	LaneID               string   `json:"lane_id"`
+	ModuleKey            string   `json:"module_key"`
+	ChunkID              string   `json:"chunk_id"`
+	Status               string   `json:"status"`
+	RejectingValidators  []string `json:"rejecting_validators"`
+	ReasonCodes          []string `json:"reason_codes"`
+	IncompleteValidators []string `json:"incomplete_validators"`
+	ProducerAttemptCount *int     `json:"producer_attempt_count"`
+	TerminalAction       string   `json:"terminal_action"`
+}
+
+func validValidationStatus(value string) bool {
+	switch value {
+	case "approved", "rejected", "incomplete":
+		return true
+	default:
+		return false
+	}
+}
+
+func validateValidationSummaries(documents []validationSummaryDocument) ([]ValidationSummary, error) {
+	summaries := make([]ValidationSummary, 0, len(documents))
+	for _, document := range documents {
+		if document.Status != "complete" && document.Status != "rejected" && document.Status != "incomplete" {
+			return nil, fmt.Errorf("notarius validation summary status %q is invalid", document.Status)
+		}
+		if document.ProducerAttemptCount == nil || *document.ProducerAttemptCount <= 0 || !validTerminalAction(document.TerminalAction) {
+			return nil, fmt.Errorf("notarius validation summary is missing required fields")
+		}
+		summaries = append(summaries, ValidationSummary{
+			Stage: document.Stage, StepID: document.StepID, LaneID: document.LaneID,
+			ModuleKey: document.ModuleKey, ChunkID: document.ChunkID, Status: document.Status,
+			RejectingValidators:  append([]string(nil), document.RejectingValidators...),
+			ReasonCodes:          append([]string(nil), document.ReasonCodes...),
+			IncompleteValidators: append([]string(nil), document.IncompleteValidators...),
+			ProducerAttemptCount: *document.ProducerAttemptCount, TerminalAction: document.TerminalAction,
+		})
+	}
+	return summaries, nil
+}
+
+func validTerminalAction(value string) bool {
+	switch value {
+	case "accepted", "reject_output", "warn_continue", "fail_run":
+		return true
+	default:
+		return false
+	}
+}
+
+func loadReceipt(path, pipelineID string) (Receipt, error) {
+	var document receiptDocument
+	if err := decodeBoundedJSON(path, maxReceiptBytes, &document); err != nil {
+		return Receipt{}, fmt.Errorf("decode notarius receipt: %w", err)
+	}
+	if document.SchemaVersion != ReceiptSchemaVersion {
+		return Receipt{}, fmt.Errorf("unsupported notarius receipt schema version %q", document.SchemaVersion)
+	}
+	if strings.TrimSpace(document.RunID) == "" || strings.TrimSpace(document.PipelineID) == "" ||
+		strings.TrimSpace(document.OutputDirectory) == "" || strings.TrimSpace(document.ValidationStatus) == "" ||
+		document.NormalizedOutputCount == nil ||
+		document.RejectedOutputCount == nil || document.WarningGroupCount == nil ||
+		document.WarningOccurrenceCount == nil || document.DiagnosticGroupCount == nil ||
+		document.DiagnosticOccurrenceCount == nil || document.DiagnosticsTruncated == nil {
+		return Receipt{}, fmt.Errorf("notarius receipt is missing required fields")
+	}
+	if document.IndexFile != canonicalIndexFile {
+		return Receipt{}, fmt.Errorf("notarius receipt index_file %q is incompatible; want %q", document.IndexFile, canonicalIndexFile)
+	}
+	if document.PipelineID != pipelineID {
+		return Receipt{}, fmt.Errorf("notarius receipt pipeline id %q does not match requested pipeline %q", document.PipelineID, pipelineID)
+	}
+	if *document.NormalizedOutputCount < 0 || *document.RejectedOutputCount < 0 ||
+		*document.WarningGroupCount < 0 || *document.WarningOccurrenceCount < 0 ||
+		*document.DiagnosticGroupCount < 0 || *document.DiagnosticOccurrenceCount < 0 {
+		return Receipt{}, fmt.Errorf("notarius receipt counts must be non-negative")
+	}
+	if !validValidationStatus(document.ValidationStatus) {
+		return Receipt{}, fmt.Errorf("notarius receipt validation_status %q is invalid", document.ValidationStatus)
+	}
+	validationSummaries, err := validateValidationSummaries(document.ValidationSummaries)
+	if err != nil {
+		return Receipt{}, err
+	}
+	if !filepath.IsAbs(document.OutputDirectory) {
+		return Receipt{}, fmt.Errorf("notarius receipt output directory must be absolute")
+	}
+	if document.DebugDirectory != "" && !filepath.IsAbs(document.DebugDirectory) {
+		return Receipt{}, fmt.Errorf("notarius receipt debug directory must be absolute when present")
+	}
+	return Receipt{
+		SchemaVersion:             document.SchemaVersion,
+		RunID:                     document.RunID,
+		PipelineID:                document.PipelineID,
+		OutputDirectory:           filepath.Clean(document.OutputDirectory),
+		IndexFile:                 document.IndexFile,
+		NormalizedOutputCount:     *document.NormalizedOutputCount,
+		RejectedOutputCount:       *document.RejectedOutputCount,
+		WarningGroupCount:         *document.WarningGroupCount,
+		WarningOccurrenceCount:    *document.WarningOccurrenceCount,
+		DiagnosticGroupCount:      *document.DiagnosticGroupCount,
+		DiagnosticOccurrenceCount: *document.DiagnosticOccurrenceCount,
+		DiagnosticsTruncated:      *document.DiagnosticsTruncated,
+		ValidationStatus:          document.ValidationStatus,
+		ValidationSummaries:       validationSummaries,
+		DebugDirectory:            document.DebugDirectory,
+	}, nil
+}
+
+type indexDocument struct {
+	ManifestFile    string            `json:"manifest_file"`
+	OutputFiles     *[]laneDocument   `json:"output_files"`
+	RejectedFile    string            `json:"rejected_file"`
+	WarningsFile    string            `json:"warnings_file"`
+	DiagnosticsFile string            `json:"diagnostics_file"`
+	ChunkMap        *pipelineDocument `json:"chunk_map"`
+	EvidenceContext *pipelineDocument `json:"evidence_context"`
+}
+
+type laneDocument struct {
+	LaneID        string `json:"lane_id"`
+	File          string `json:"file"`
+	MediaType     string `json:"media_type"`
+	ModuleKey     string `json:"module_key"`
+	SchemaID      string `json:"schema_id"`
+	SchemaName    string `json:"schema_name"`
+	SchemaVersion string `json:"schema_version"`
+}
+
+type pipelineDocument struct {
+	ArtifactKind  string `json:"artifact_kind"`
+	File          string `json:"file"`
+	MediaType     string `json:"media_type"`
+	SchemaID      string `json:"schema_id"`
+	SchemaName    string `json:"schema_name"`
+	SchemaVersion string `json:"schema_version"`
+}
+
+func loadIndex(bundleRoot, indexPath string) (Index, error) {
+	var document indexDocument
+	if err := decodeBoundedJSON(indexPath, maxIndexBytes, &document); err != nil {
+		return Index{}, fmt.Errorf("decode notarius index: %w", err)
+	}
+	for _, field := range []struct {
+		name string
+		got  string
+		want string
+	}{
+		{name: "manifest_file", got: document.ManifestFile, want: canonicalManifestFile},
+		{name: "rejected_file", got: document.RejectedFile, want: canonicalRejectedFile},
+		{name: "warnings_file", got: document.WarningsFile, want: canonicalWarningsFile},
+		{name: "diagnostics_file", got: document.DiagnosticsFile, want: canonicalDiagnosticsFile},
+	} {
+		if field.got != field.want {
+			return Index{}, fmt.Errorf("notarius index %s %q is incompatible; want %q", field.name, field.got, field.want)
+		}
+	}
+	if document.OutputFiles == nil {
+		return Index{}, fmt.Errorf("notarius index is missing required output_files")
+	}
+
+	index := Index{
+		Path:            indexPath,
+		ManifestFile:    document.ManifestFile,
+		RejectedFile:    document.RejectedFile,
+		WarningsFile:    document.WarningsFile,
+		DiagnosticsFile: document.DiagnosticsFile,
+	}
+	var err error
+	if index.ManifestPath, err = resolveRegularFile(bundleRoot, index.ManifestFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius manifest file: %w", err)
+	}
+	if index.RejectedPath, err = resolveRegularFile(bundleRoot, index.RejectedFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius rejection file: %w", err)
+	}
+	if index.WarningsPath, err = resolveRegularFile(bundleRoot, index.WarningsFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius warning file: %w", err)
+	}
+	if index.DiagnosticsPath, err = resolveRegularFile(bundleRoot, index.DiagnosticsFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius diagnostics file: %w", err)
+	}
+
+	seenLanes := make(map[string]struct{}, len(*document.OutputFiles))
+	for _, lane := range *document.OutputFiles {
+		if strings.TrimSpace(lane.LaneID) == "" || strings.TrimSpace(lane.File) == "" {
+			return Index{}, fmt.Errorf("notarius lane descriptors require lane_id and file")
+		}
+		if _, exists := seenLanes[lane.LaneID]; exists {
+			return Index{}, fmt.Errorf("notarius index contains duplicate lane id %q", lane.LaneID)
+		}
+		seenLanes[lane.LaneID] = struct{}{}
+		path, err := resolveRegularFile(bundleRoot, lane.File)
+		if err != nil {
+			return Index{}, fmt.Errorf("resolve notarius lane %q file: %w", lane.LaneID, err)
+		}
+		index.Lanes = append(index.Lanes, LaneDescriptor{
+			LaneID: lane.LaneID, File: lane.File, Path: path, MediaType: lane.MediaType,
+			ModuleKey: lane.ModuleKey, SchemaID: lane.SchemaID, SchemaName: lane.SchemaName,
+			SchemaVersion: lane.SchemaVersion,
+		})
+	}
+	if document.ChunkMap != nil {
+		index.ChunkMap, err = resolvePipelineDescriptor(bundleRoot, "chunk_map", *document.ChunkMap)
+		if err != nil {
+			return Index{}, err
+		}
+	}
+	if document.EvidenceContext != nil {
+		index.EvidenceContext, err = resolvePipelineDescriptor(bundleRoot, "evidence_context", *document.EvidenceContext)
+		if err != nil {
+			return Index{}, err
+		}
+	}
+	return index, nil
+}
+
+func resolvePipelineDescriptor(bundleRoot, label string, document pipelineDocument) (*PipelineDescriptor, error) {
+	if strings.TrimSpace(document.ArtifactKind) == "" || strings.TrimSpace(document.File) == "" ||
+		strings.TrimSpace(document.MediaType) == "" || strings.TrimSpace(document.SchemaID) == "" ||
+		strings.TrimSpace(document.SchemaName) == "" || strings.TrimSpace(document.SchemaVersion) == "" {
+		return nil, fmt.Errorf("notarius %s descriptor is missing required fields", label)
+	}
+	path, err := resolveRegularFile(bundleRoot, document.File)
+	if err != nil {
+		return nil, fmt.Errorf("resolve notarius %s file: %w", label, err)
+	}
+	return &PipelineDescriptor{
+		ArtifactKind: document.ArtifactKind, File: document.File, Path: path,
+		MediaType: document.MediaType, SchemaID: document.SchemaID,
+		SchemaName: document.SchemaName, SchemaVersion: document.SchemaVersion,
+	}, nil
+}
+
+type rejectionDocument struct {
+	Rejected *[]struct {
+		Stage         string `json:"stage"`
+		StepID        string `json:"step_id"`
+		LaneID        string `json:"lane_id"`
+		ModuleKey     string `json:"module_key"`
+		ChunkID       string `json:"chunk_id"`
+		ValidatorName string `json:"validator_name"`
+		ReasonCode    string `json:"reason_code"`
+		Message       string `json:"message"`
+	} `json:"rejected"`
+}
+
+func loadRejections(path string) ([]RejectionSummary, error) {
+	var document rejectionDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, fmt.Errorf("decode notarius rejections: %w", err)
+	}
+	if document.Rejected == nil {
+		return nil, fmt.Errorf("notarius rejection document is missing rejected array")
+	}
+	summaries := make([]RejectionSummary, 0, len(*document.Rejected))
+	for _, item := range *document.Rejected {
+		if strings.TrimSpace(item.Stage) == "" || strings.TrimSpace(item.Message) == "" {
+			return nil, fmt.Errorf("notarius rejection entries require stage and message")
+		}
+		summaries = append(summaries, RejectionSummary{
+			Stage: item.Stage, StepID: item.StepID, LaneID: item.LaneID,
+			ModuleKey: item.ModuleKey, ChunkID: item.ChunkID,
+			ValidatorName: item.ValidatorName, ReasonCode: item.ReasonCode,
+		})
+	}
+	return summaries, nil
+}
+
+type findingGroupDocument struct {
+	Disposition     string                   `json:"disposition"`
+	Category        string                   `json:"category"`
+	ReasonCode      string                   `json:"reason_code"`
+	Origin          diagnosticOriginDocument `json:"origin"`
+	OccurrenceCount *int                     `json:"occurrence_count"`
+	Samples         *[]struct {
+		Scope      string `json:"scope"`
+		Message    string `json:"message"`
+		ChunkID    string `json:"chunk_id"`
+		ChunkIndex *int   `json:"chunk_index"`
+	} `json:"samples"`
+	OmittedSampleCount *int `json:"omitted_sample_count"`
+}
+
+type diagnosticOriginDocument struct {
+	Stage        string `json:"stage"`
+	StepID       string `json:"step_id"`
+	LaneID       string `json:"lane_id"`
+	ModuleKey    string `json:"module_key"`
+	ValidatorKey string `json:"validator_key"`
+}
+
+type warningDocument struct {
+	SchemaVersion   string                  `json:"schema_version"`
+	GroupCount      *int                    `json:"group_count"`
+	OccurrenceCount *int                    `json:"occurrence_count"`
+	Groups          *[]findingGroupDocument `json:"groups"`
+}
+
+func loadWarnings(path string) ([]WarningSummary, error) {
+	var document warningDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, fmt.Errorf("decode notarius warnings: %w", err)
+	}
+	if document.SchemaVersion != warningsSchemaVersion || document.GroupCount == nil ||
+		document.OccurrenceCount == nil || document.Groups == nil {
+		return nil, fmt.Errorf("notarius warning document is missing or incompatible required fields")
+	}
+	if *document.GroupCount < 0 || *document.GroupCount > maxWarningGroups || *document.OccurrenceCount < 0 ||
+		*document.GroupCount != len(*document.Groups) {
+		return nil, fmt.Errorf("notarius warning document counts are inconsistent")
+	}
+	summaries := make([]WarningSummary, 0, len(*document.Groups))
+	occurrences := 0
+	for _, group := range *document.Groups {
+		if err := validateFindingGroup(group); err != nil {
+			return nil, fmt.Errorf("notarius warning group: %w", err)
+		}
+		if group.Disposition != "warning" {
+			return nil, fmt.Errorf("notarius warning group disposition %q is invalid", group.Disposition)
+		}
+		if *group.OccurrenceCount > int(^uint(0)>>1)-occurrences {
+			return nil, fmt.Errorf("notarius warning occurrence count overflows")
+		}
+		occurrences += *group.OccurrenceCount
+		summaries = append(summaries, WarningSummary{
+			Disposition: group.Disposition, Category: group.Category, ReasonCode: group.ReasonCode,
+			Origin: diagnosticOrigin(group.Origin), OccurrenceCount: *group.OccurrenceCount,
+		})
+	}
+	if occurrences != *document.OccurrenceCount {
+		return nil, fmt.Errorf("notarius warning document occurrence count is inconsistent")
+	}
+	return summaries, nil
+}
+
+type diagnosticDocument struct {
+	SchemaVersion                string                  `json:"schema_version"`
+	GroupCount                   *int                    `json:"group_count"`
+	OccurrenceCount              *int                    `json:"occurrence_count"`
+	Truncated                    *bool                   `json:"truncated"`
+	UnrepresentedOccurrenceCount *int                    `json:"unrepresented_occurrence_count"`
+	Groups                       *[]findingGroupDocument `json:"groups"`
+}
+
+func loadDiagnostics(path string) ([]DiagnosticSummary, int, bool, error) {
+	var document diagnosticDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, 0, false, fmt.Errorf("decode notarius diagnostics: %w", err)
+	}
+	if document.SchemaVersion != diagnosticsSchemaVersion || document.GroupCount == nil ||
+		document.OccurrenceCount == nil || document.Truncated == nil ||
+		document.UnrepresentedOccurrenceCount == nil || document.Groups == nil {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document is missing or incompatible required fields")
+	}
+	if *document.GroupCount < 0 || *document.GroupCount > maxDiagnosticGroups || *document.OccurrenceCount < 0 ||
+		*document.UnrepresentedOccurrenceCount < 0 || *document.GroupCount != len(*document.Groups) {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document counts are inconsistent")
+	}
+	if !*document.Truncated && *document.UnrepresentedOccurrenceCount != 0 {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document has unrepresented occurrences without truncation")
+	}
+	summaries := make([]DiagnosticSummary, 0, len(*document.Groups))
+	representedOccurrences := 0
+	for _, group := range *document.Groups {
+		if err := validateFindingGroup(group); err != nil {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic group: %w", err)
+		}
+		if group.Disposition != "advisory" && group.Disposition != "observation" {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic group disposition %q is invalid", group.Disposition)
+		}
+		if *group.OccurrenceCount > int(^uint(0)>>1)-representedOccurrences {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic occurrence count overflows")
+		}
+		representedOccurrences += *group.OccurrenceCount
+		summaries = append(summaries, DiagnosticSummary{
+			Disposition: group.Disposition, Category: group.Category, ReasonCode: group.ReasonCode,
+			Origin: diagnosticOrigin(group.Origin), OccurrenceCount: *group.OccurrenceCount,
+		})
+	}
+	if *document.UnrepresentedOccurrenceCount > int(^uint(0)>>1)-representedOccurrences ||
+		representedOccurrences+*document.UnrepresentedOccurrenceCount != *document.OccurrenceCount {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document occurrence count is inconsistent")
+	}
+	return summaries, *document.OccurrenceCount, *document.Truncated, nil
+}
+
+func validateFindingGroup(group findingGroupDocument) error {
+	if strings.TrimSpace(group.Disposition) == "" || strings.TrimSpace(group.Category) == "" ||
+		strings.TrimSpace(group.ReasonCode) == "" || !validDiagnosticOriginStage(group.Origin.Stage) ||
+		!validDiagnosticCategory(group.Disposition, group.Category) ||
+		group.OccurrenceCount == nil || *group.OccurrenceCount <= 0 || group.Samples == nil ||
+		group.OmittedSampleCount == nil || *group.OmittedSampleCount < 0 {
+		return fmt.Errorf("missing required fields")
+	}
+	if len(*group.Samples) == 0 || len(*group.Samples) > maxFindingSamples ||
+		*group.OmittedSampleCount != *group.OccurrenceCount-len(*group.Samples) {
+		return fmt.Errorf("sample counts are inconsistent")
+	}
+	for _, sample := range *group.Samples {
+		if strings.TrimSpace(sample.Scope) == "" || strings.TrimSpace(sample.Message) == "" ||
+			(sample.ChunkIndex != nil && *sample.ChunkIndex < 0) {
+			return fmt.Errorf("samples require scope and message")
+		}
+	}
+	return nil
+}
+
+func validDiagnosticCategory(disposition, category string) bool {
+	switch disposition {
+	case "warning":
+		return category == "configuration" || category == "degradation" ||
+			category == "validation_incomplete" || category == "fallback"
+	case "advisory":
+		return category == "data_quality"
+	case "observation":
+		return category == "normalization"
+	default:
+		return false
+	}
+}
+
+func validDiagnosticOriginStage(stage string) bool {
+	switch stage {
+	case "references", "chunk", "extract", "merge", "normalize":
+		return true
+	default:
+		return false
+	}
+}
+
+func diagnosticOrigin(document diagnosticOriginDocument) DiagnosticOrigin {
+	return DiagnosticOrigin{
+		Stage: document.Stage, StepID: document.StepID, LaneID: document.LaneID,
+		ModuleKey: document.ModuleKey, ValidatorKey: document.ValidatorKey,
+	}
+}
+
+func sumWarningOccurrences(values []WarningSummary) (int, error) {
+	total := 0
+	for _, value := range values {
+		if value.OccurrenceCount > int(^uint(0)>>1)-total {
+			return 0, fmt.Errorf("notarius warning occurrence count overflows")
+		}
+		total += value.OccurrenceCount
+	}
+	return total, nil
+}
+
+func decodeBoundedJSON(path string, limit int64, destination any) error {
+	data, err := fileops.ReadFileBounded(path, limit)
+	if err != nil {
+		return fmt.Errorf("notarius JSON result exceeds or cannot be read within %d-byte limit: %w", limit, err)
+	}
+	if err := json.Unmarshal(data, destination); err != nil {
+		return err
+	}
+	return nil
+}
+
+func validateBundleRoot(outputRoot, bundleRoot string) (string, error) {
+	root := filepath.Clean(outputRoot)
+	bundle := filepath.Clean(bundleRoot)
+	relative, err := filepath.Rel(root, bundle)
+	if err != nil {
+		return "", fmt.Errorf("compare notarius output paths: %w", err)
+	}
+	if relative == "." || relative == ".." || strings.HasPrefix(relative, ".."+string(filepath.Separator)) {
+		return "", fmt.Errorf("notarius output directory %q is not beneath output root %q", bundleRoot, outputRoot)
+	}
+	if err := requireDirectoryTree(root, relative); err != nil {
+		return "", fmt.Errorf("validate notarius output directory: %w", err)
+	}
+	return bundle, nil
+}
+
+func resolveRegularFile(root, logicalPath string) (string, error) {
+	resolved, err := pathsafe.JoinSlashRelativeUnderRoot(root, logicalPath)
+	if err != nil {
+		return "", err
+	}
+	relative, err := filepath.Rel(root, resolved)
+	if err != nil {
+		return "", err
+	}
+	if err := requireRegularFileTree(root, relative); err != nil {
+		return "", err
+	}
+	return resolved, nil
+}
+
+func requireDirectoryTree(root, relative string) error {
+	if err := requireDirectory(root); err != nil {
+		return err
+	}
+	current := root
+	for _, component := range strings.Split(relative, string(filepath.Separator)) {
+		current = filepath.Join(current, component)
+		if err := requireDirectory(current); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func requireRegularFileTree(root, relative string) error {
+	components := strings.Split(relative, string(filepath.Separator))
+	if len(components) == 0 {
+		return fmt.Errorf("regular file path is required")
+	}
+	if err := requireDirectory(root); err != nil {
+		return err
+	}
+	current := root
+	for _, component := range components[:len(components)-1] {
+		current = filepath.Join(current, component)
+		if err := requireDirectory(current); err != nil {
+			return err
+		}
+	}
+	return requireRegularFile(filepath.Join(current, components[len(components)-1]))
+}
+
+func requireDirectory(path string) error {
+	info, err := os.Lstat(path)
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.IsDir() {
+		return fmt.Errorf("path %q must be a directory without symlinks", path)
+	}
+	return nil
+}
+
+func requireRegularFile(path string) error {
+	info, err := os.Lstat(path)
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.Mode().IsRegular() {
+		return fmt.Errorf("path %q must be a regular file without symlinks", path)
+	}
+	return nil
+}
+
+func validateLogDestination(path string) error {
+	if err := requireDirectory(filepath.Dir(path)); err != nil {
+		return err
+	}
+	info, err := os.Lstat(path)
+	if errors.Is(err, os.ErrNotExist) {
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.Mode().IsRegular() {
+		return fmt.Errorf("path %q must be absent or a regular file without symlinks", path)
+	}
+	return nil
+}

internal/adapters/notarius/subprocess_test.go

@@ -0,0 +1,716 @@
+package notarius
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	sharedsubprocess "gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+)
+
+func TestSubprocessRunnerBuildsExactInvocationAndDiscoversBundle(t *testing.T) {
+	req := validRunRequest(t)
+	var captured sharedsubprocess.RunRequest
+	runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		captured = processReq
+		writeValidBundleAndReceipt(t, req, true)
+		return sharedsubprocess.RunResult{ExitCode: 0, Duration: 2 * time.Second}, nil
+	}}
+
+	result, err := runner.Run(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	wantArgs := []string{
+		"run", "dnd-session", "--config", req.ConfigPath, "--input", req.InputPath,
+		"--output-dir", req.OutputRoot, "--json",
+	}
+	if !reflect.DeepEqual(captured.Args, wantArgs) {
+		t.Fatalf("subprocess args = %#v, want %#v", captured.Args, wantArgs)
+	}
+	if captured.Executable != req.Binary || captured.WorkingDir != req.WorkingDirectory || captured.Timeout != req.Timeout {
+		t.Fatalf("subprocess request = %#v", captured)
+	}
+	if captured.StdoutLogPath != req.ReceiptPath || captured.StderrLogPath != req.LogPath {
+		t.Fatalf("stream paths = stdout %q stderr %q", captured.StdoutLogPath, captured.StderrLogPath)
+	}
+	if captured.EnvOverrides != nil {
+		t.Fatalf("environment overrides = %#v, want inherited environment only", captured.EnvOverrides)
+	}
+	for _, arg := range captured.Args {
+		if arg == "--session-id" {
+			t.Fatal("subprocess args unexpectedly contain --session-id")
+		}
+	}
+
+	if result.Receipt.SchemaVersion != ReceiptSchemaVersion || result.Receipt.RunID != "notarius-run-1" {
+		t.Fatalf("receipt = %#v", result.Receipt)
+	}
+	if len(result.Receipt.ValidationSummaries) != 1 || result.Receipt.ValidationSummaries[0].LaneID != "npc-registry" ||
+		result.Receipt.ValidationSummaries[0].Status != "complete" {
+		t.Fatalf("validation summaries = %#v", result.Receipt.ValidationSummaries)
+	}
+	if len(result.Index.Lanes) != 1 || result.Index.Lanes[0].LaneID != "npc-registry" {
+		t.Fatalf("lanes = %#v", result.Index.Lanes)
+	}
+	if result.Index.ChunkMap == nil || result.Index.ChunkMap.ArtifactKind != "chunk_map" {
+		t.Fatalf("chunk map = %#v", result.Index.ChunkMap)
+	}
+	if result.Index.EvidenceContext == nil || result.Index.EvidenceContext.ArtifactKind != "evidence_context" {
+		t.Fatalf("evidence context = %#v", result.Index.EvidenceContext)
+	}
+	if len(result.Rejections) != 1 || result.Rejections[0].LaneID != "spells" || result.Rejections[0].ReasonCode != "invalid_spell" {
+		t.Fatalf("rejections = %#v", result.Rejections)
+	}
+	if len(result.Warnings) != 1 || result.Warnings[0].Category != "degradation" || result.Warnings[0].ReasonCode != "normalized_name" {
+		t.Fatalf("warnings = %#v", result.Warnings)
+	}
+	if len(result.Diagnostics) != 1 || result.Diagnostics[0].Category != "data_quality" || result.Diagnostics[0].ReasonCode != "low_confidence" {
+		t.Fatalf("diagnostics = %#v", result.Diagnostics)
+	}
+}
+
+func TestSubprocessRunnerBuildsOrderedReferenceArguments(t *testing.T) {
+	req := validRunRequest(t)
+	referenceRoot := t.TempDir()
+	req.References = []ReferenceBinding{
+		{Selector: " party ", Path: filepath.Join(referenceRoot, "party context=primary.json")},
+		{Selector: " npc-registry . extract . glossary ", Path: filepath.Join(referenceRoot, "glossary.json")},
+	}
+	originalReferences := append([]ReferenceBinding(nil), req.References...)
+	var captured sharedsubprocess.RunRequest
+	runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		captured = processReq
+		writeValidBundleAndReceipt(t, req, false)
+		return sharedsubprocess.RunResult{ExitCode: 0}, nil
+	}}
+
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	wantArgs := []string{
+		"run", req.PipelineID,
+		"--config", req.ConfigPath,
+		"--input", req.InputPath,
+		"--output-dir", req.OutputRoot,
+		"--reference", "party=" + req.References[0].Path,
+		"--reference", "npc-registry.extract.glossary=" + req.References[1].Path,
+		"--json",
+	}
+	if !reflect.DeepEqual(captured.Args, wantArgs) {
+		t.Fatalf("subprocess args = %#v, want %#v", captured.Args, wantArgs)
+	}
+	if !reflect.DeepEqual(req.References, originalReferences) {
+		t.Fatalf("Run() mutated caller references = %#v, want %#v", req.References, originalReferences)
+	}
+}
+
+func TestSubprocessRunnerRejectsInvalidReferencesBeforeLaunch(t *testing.T) {
+	tests := []struct {
+		name       string
+		references func(string) []ReferenceBinding
+		wantErr    string
+	}{
+		{
+			name: "invalid selector",
+			references: func(root string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "lane.prepare.party", Path: filepath.Join(root, "party.json")}}
+			},
+			wantErr: "selector",
+		},
+		{
+			name: "duplicate normalized selector",
+			references: func(root string) []ReferenceBinding {
+				return []ReferenceBinding{
+					{Selector: "lane.party", Path: filepath.Join(root, "party.json")},
+					{Selector: " lane . party ", Path: filepath.Join(root, "party-2.json")},
+				}
+			},
+			wantErr: "duplicated",
+		},
+		{
+			name: "empty path",
+			references: func(string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "party", Path: "  "}}
+			},
+			wantErr: "path is required",
+		},
+		{
+			name: "relative path",
+			references: func(string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "party", Path: "references/party.json"}}
+			},
+			wantErr: "path must be absolute",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.References = tt.references(t.TempDir())
+			started := false
+			runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				started = true
+				return sharedsubprocess.RunResult{}, nil
+			}}
+
+			_, err := runner.Run(context.Background(), req)
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("Run() error = %v, want containing %q", err, tt.wantErr)
+			}
+			if started {
+				t.Fatal("subprocess started after request validation failure")
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerUsesMinimalEnvironmentAndSeparatesStreams(t *testing.T) {
+	req := validRunRequest(t)
+	writeValidBundleAndReceipt(t, req, false)
+	receiptFixture := req.ReceiptPath + ".fixture"
+	data, err := os.ReadFile(req.ReceiptPath)
+	if err != nil {
+		t.Fatalf("ReadFile(receipt) error = %v", err)
+	}
+	if err := os.WriteFile(receiptFixture, data, 0o644); err != nil {
+		t.Fatalf("WriteFile(receipt fixture) error = %v", err)
+	}
+	if err := os.Remove(req.ReceiptPath); err != nil {
+		t.Fatalf("Remove(receipt) error = %v", err)
+	}
+
+	captureDir := filepath.Join(filepath.Dir(req.ReceiptPath), "capture")
+	if err := os.Mkdir(captureDir, 0o755); err != nil {
+		t.Fatalf("Mkdir(capture) error = %v", err)
+	}
+	script := writeShellScript(t, `#!/bin/sh
+pwd > "$NOTARIUS_CAPTURE_DIR/working-directory"
+printf '%s' "$NOTARIUS_INHERITED_VALUE" > "$NOTARIUS_CAPTURE_DIR/environment"
+printf 'diagnostic stream\n' >&2
+cat "$NOTARIUS_RECEIPT_FIXTURE"
+`)
+	req.Binary = script
+	t.Setenv("NOTARIUS_CAPTURE_DIR", captureDir)
+	t.Setenv("NOTARIUS_INHERITED_VALUE", "inherited-value")
+	t.Setenv("NOTARIUS_RECEIPT_FIXTURE", receiptFixture)
+
+	if _, err := NewSubprocessRunner().Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	assertTextFile(t, filepath.Join(captureDir, "working-directory"), req.WorkingDirectory+"\n")
+	assertTextFile(t, filepath.Join(captureDir, "environment"), "")
+	assertTextFile(t, req.LogPath, "diagnostic stream\n")
+	receiptBytes, err := os.ReadFile(req.ReceiptPath)
+	if err != nil {
+		t.Fatalf("ReadFile(receipt) error = %v", err)
+	}
+	if strings.Contains(string(receiptBytes), "diagnostic stream") {
+		t.Fatal("receipt contains stderr output")
+	}
+}
+
+func TestSubprocessRunnerReturnsProcessFailuresWithoutParsingStdout(t *testing.T) {
+	tests := []struct {
+		name       string
+		scriptBody string
+		timeout    time.Duration
+		cancel     bool
+		want       string
+	}{
+		{name: "nonzero", scriptBody: "printf '{malformed receipt'; printf 'failed\\n' >&2; exit 7\n", timeout: time.Second, want: "exit code 7"},
+		{name: "timeout", scriptBody: "sleep 5\n", timeout: 20 * time.Millisecond, want: "timed out"},
+		{name: "cancellation", scriptBody: "sleep 5\n", timeout: time.Second, cancel: true, want: "canceled"},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.Binary = writeShellScript(t, "#!/bin/sh\n"+test.scriptBody)
+			req.Timeout = test.timeout
+			ctx := context.Background()
+			if test.cancel {
+				cancelCtx, cancel := context.WithCancel(ctx)
+				ctx = cancelCtx
+				time.AfterFunc(20*time.Millisecond, cancel)
+			}
+			_, err := NewSubprocessRunner().Run(ctx, req)
+			if err == nil || !strings.Contains(err.Error(), test.want) {
+				t.Fatalf("Run() error = %v, want fragment %q", err, test.want)
+			}
+			if strings.Contains(err.Error(), "decode notarius receipt") {
+				t.Fatalf("Run() parsed stdout after process failure: %v", err)
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerReturnsSharedSubprocessErrorWithoutReadingReceipt(t *testing.T) {
+	req := validRunRequest(t)
+	if err := os.WriteFile(req.ReceiptPath, []byte("not json"), 0o644); err != nil {
+		t.Fatalf("WriteFile(receipt) error = %v", err)
+	}
+	wantErr := errors.New("process failed")
+	runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		return sharedsubprocess.RunResult{ExitCode: 9}, wantErr
+	}}
+	_, err := runner.Run(context.Background(), req)
+	if !errors.Is(err, wantErr) {
+		t.Fatalf("Run() error = %v, want wrapped process error", err)
+	}
+	if strings.Contains(err.Error(), "decode") {
+		t.Fatalf("Run() parsed receipt after failure: %v", err)
+	}
+}
+
+func TestLoadReceiptValidation(t *testing.T) {
+	root := t.TempDir()
+	valid := map[string]any{
+		"schema_version": ReceiptSchemaVersion, "run_id": "run-1", "pipeline_id": "pipeline-1",
+		"output_directory": filepath.Join(root, "outputs", "run-1"), "index_file": "index.json",
+		"normalized_output_count": 1, "rejected_output_count": 0,
+		"warning_group_count": 0, "warning_occurrence_count": 0,
+		"diagnostic_group_count": 0, "diagnostic_occurrence_count": 0,
+		"diagnostics_truncated": false, "validation_status": "approved", "future_field": true,
+	}
+	tests := []struct {
+		name      string
+		mutate    func(map[string]any)
+		raw       []byte
+		wantOK    bool
+		wantError string
+	}{
+		{name: "unknown fields tolerated", wantOK: true},
+		{name: "malformed", raw: []byte("{")},
+		{name: "unsupported version", mutate: func(v map[string]any) { v["schema_version"] = "notarius.run-result.v1" }},
+		{name: "missing field", mutate: func(v map[string]any) { delete(v, "run_id") }},
+		{name: "pipeline mismatch", mutate: func(v map[string]any) { v["pipeline_id"] = "other" }},
+		{name: "relative output", mutate: func(v map[string]any) { v["output_directory"] = "run-1" }},
+		{name: "negative count", mutate: func(v map[string]any) { v["warning_group_count"] = -1 }},
+		{name: "invalid validation status", mutate: func(v map[string]any) { v["validation_status"] = "valid" }},
+		{
+			name: "nested index", mutate: func(v map[string]any) { v["index_file"] = "nested/index.json" },
+			wantError: `index_file "nested/index.json"`,
+		},
+		{
+			name: "cleanable index", mutate: func(v map[string]any) { v["index_file"] = "./index.json" },
+			wantError: `index_file "./index.json"`,
+		},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			path := filepath.Join(root, strings.ReplaceAll(test.name, " ", "-")+".json")
+			values := cloneMap(valid)
+			if test.mutate != nil {
+				test.mutate(values)
+			}
+			if test.raw != nil {
+				if err := os.WriteFile(path, test.raw, 0o644); err != nil {
+					t.Fatalf("WriteFile() error = %v", err)
+				}
+			} else {
+				writeJSONFile(t, path, values)
+			}
+			_, err := loadReceipt(path, "pipeline-1")
+			if test.wantOK && err != nil {
+				t.Fatalf("loadReceipt() error = %v", err)
+			}
+			if !test.wantOK && err == nil {
+				t.Fatal("loadReceipt() error = nil, want validation failure")
+			}
+			if test.wantError != "" && !strings.Contains(err.Error(), test.wantError) {
+				t.Fatalf("loadReceipt() error = %v, want fragment %q", err, test.wantError)
+			}
+		})
+	}
+
+	oversized := filepath.Join(root, "oversized.json")
+	if err := os.WriteFile(oversized, []byte(strings.Repeat("x", maxReceiptBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized) error = %v", err)
+	}
+	if _, err := loadReceipt(oversized, "pipeline-1"); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadReceipt(oversized) error = %v", err)
+	}
+}
+
+func TestValidateBundleRootRejectsEscapesAndSymlinks(t *testing.T) {
+	root := t.TempDir()
+	outputRoot := filepath.Join(root, "output")
+	if err := os.Mkdir(outputRoot, 0o755); err != nil {
+		t.Fatalf("Mkdir(output root) error = %v", err)
+	}
+	validBundle := filepath.Join(outputRoot, "run-1")
+	if err := os.Mkdir(validBundle, 0o755); err != nil {
+		t.Fatalf("Mkdir(bundle) error = %v", err)
+	}
+	if _, err := validateBundleRoot(outputRoot, validBundle); err != nil {
+		t.Fatalf("validateBundleRoot(valid) error = %v", err)
+	}
+
+	outside := filepath.Join(root, "output-other")
+	if err := os.Mkdir(outside, 0o755); err != nil {
+		t.Fatalf("Mkdir(outside) error = %v", err)
+	}
+	for name, candidate := range map[string]string{"equal root": outputRoot, "escape": root, "prefix confusion": outside} {
+		t.Run(name, func(t *testing.T) {
+			if _, err := validateBundleRoot(outputRoot, candidate); err == nil {
+				t.Fatalf("validateBundleRoot(%q) error = nil", candidate)
+			}
+		})
+	}
+	symlink := filepath.Join(outputRoot, "linked")
+	if err := os.Symlink(outside, symlink); err != nil {
+		t.Skipf("Symlink() unavailable: %v", err)
+	}
+	if _, err := validateBundleRoot(outputRoot, symlink); err == nil {
+		t.Fatal("validateBundleRoot(symlink) error = nil")
+	}
+}
+
+func TestLoadIndexRejectsMalformedUnsafeAndUnsupportedDocuments(t *testing.T) {
+	tests := []struct {
+		name       string
+		indexValue any
+		prepare    func(*testing.T, string)
+		wantError  string
+	}{
+		{name: "malformed", indexValue: json.RawMessage(`{"manifest_file":`)},
+		{name: "unsupported output shape", indexValue: map[string]any{"manifest_file": "manifest.json", "output_files": map[string]any{}, "rejected_file": "rejected.json", "warnings_file": "warnings.json"}},
+		{name: "missing management path", indexValue: map[string]any{"output_files": []any{}, "rejected_file": "rejected.json", "warnings_file": "warnings.json"}},
+		{name: "renamed manifest", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["manifest_file"] = "metadata.json"
+			return value
+		}(), wantError: `manifest_file "metadata.json"`},
+		{name: "cleanable manifest", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["manifest_file"] = "./manifest.json"
+			return value
+		}(), wantError: `manifest_file "./manifest.json"`},
+		{name: "renamed rejections", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["rejected_file"] = "rejections.json"
+			return value
+		}(), wantError: `rejected_file "rejections.json"`},
+		{name: "renamed warnings", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["warnings_file"] = "diagnostics/warnings.json"
+			return value
+		}(), wantError: `warnings_file "diagnostics/warnings.json"`},
+		{name: "duplicate lane", indexValue: validIndexValue([]any{
+			map[string]any{"lane_id": "npc", "file": "lanes/npc.json"},
+			map[string]any{"lane_id": "npc", "file": "lanes/npc.json"},
+		})},
+		{name: "absolute logical path", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "/tmp/npc.json"}})},
+		{name: "lexical traversal", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "../outside.json"}})},
+		{name: "root prefix confusion", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "../bundle-other/npc.json"}})},
+		{name: "file symlink", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "lanes/npc.json"}}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Symlink(filepath.Join(bundle, "manifest.json"), filepath.Join(bundle, "lanes", "npc.json")); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		}},
+		{name: "directory symlink", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "linked/npc.json"}}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Symlink(filepath.Join(bundle, "lanes"), filepath.Join(bundle, "linked")); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		}},
+		{name: "missing management file", indexValue: validIndexValue([]any{}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Remove(filepath.Join(bundle, "manifest.json")); err != nil {
+				t.Fatalf("Remove(manifest) error = %v", err)
+			}
+		}},
+		{name: "incomplete pipeline descriptor", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["chunk_map"] = map[string]any{"artifact_kind": "chunk_map", "file": "chunk-map.json"}
+			return value
+		}()},
+		{name: "pipeline descriptor escape", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["evidence_context"] = map[string]any{
+				"artifact_kind": "evidence_context", "file": "../evidence.json", "media_type": "application/json",
+				"schema_id": "evidence", "schema_name": "Evidence", "schema_version": "v1",
+			}
+			return value
+		}()},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			bundle := createBundleSkeleton(t)
+			indexPath := filepath.Join(bundle, "index.json")
+			if raw, ok := test.indexValue.(json.RawMessage); ok {
+				if err := os.WriteFile(indexPath, raw, 0o644); err != nil {
+					t.Fatalf("WriteFile(index) error = %v", err)
+				}
+			} else {
+				writeJSONFile(t, indexPath, test.indexValue)
+			}
+			if test.prepare != nil {
+				test.prepare(t, bundle)
+			}
+			if _, err := loadIndex(bundle, indexPath); err == nil {
+				t.Fatal("loadIndex() error = nil, want failure")
+			} else if test.wantError != "" && !strings.Contains(err.Error(), test.wantError) {
+				t.Fatalf("loadIndex() error = %v, want fragment %q", err, test.wantError)
+			}
+		})
+	}
+
+	bundle := createBundleSkeleton(t)
+	oversizedIndex := filepath.Join(bundle, "index.json")
+	if err := os.WriteFile(oversizedIndex, []byte(strings.Repeat("x", maxIndexBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized index) error = %v", err)
+	}
+	if _, err := loadIndex(bundle, oversizedIndex); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadIndex(oversized) error = %v", err)
+	}
+}
+
+func TestLoadDiagnosticSummariesValidateBoundsAndTolerateUnknownFields(t *testing.T) {
+	root := t.TempDir()
+	rejectedPath := filepath.Join(root, "rejected.json")
+	warningsPath := filepath.Join(root, "warnings.json")
+	diagnosticsPath := filepath.Join(root, "diagnostics.json")
+	writeJSONFile(t, rejectedPath, map[string]any{"rejected": []any{map[string]any{
+		"stage": "validate", "lane_id": "spells", "reason_code": "invalid", "message": "do not retain this", "future": true,
+	}}, "future": true})
+	writeJSONFile(t, warningsPath, findingEnvelope(warningsSchemaVersion, []any{findingGroup("warning", "degradation", "bounded", "normalize", 2)}, 2, false, 0))
+	writeJSONFile(t, diagnosticsPath, findingEnvelope(diagnosticsSchemaVersion, []any{findingGroup("advisory", "data_quality", "low_confidence", "normalize", 3)}, 4, true, 1))
+	rejections, err := loadRejections(rejectedPath)
+	if err != nil || len(rejections) != 1 || rejections[0].ReasonCode != "invalid" {
+		t.Fatalf("loadRejections() = %#v, %v", rejections, err)
+	}
+	warnings, err := loadWarnings(warningsPath)
+	if err != nil || len(warnings) != 1 || warnings[0].Category != "degradation" || warnings[0].OccurrenceCount != 2 {
+		t.Fatalf("loadWarnings() = %#v, %v", warnings, err)
+	}
+	diagnostics, occurrences, truncated, err := loadDiagnostics(diagnosticsPath)
+	if err != nil || len(diagnostics) != 1 || occurrences != 4 || !truncated || diagnostics[0].Category != "data_quality" {
+		t.Fatalf("loadDiagnostics() = %#v, %d, %t, %v", diagnostics, occurrences, truncated, err)
+	}
+
+	for name, path := range map[string]string{"rejections": rejectedPath, "warnings": warningsPath, "diagnostics": diagnosticsPath} {
+		t.Run("malformed "+name, func(t *testing.T) {
+			if err := os.WriteFile(path, []byte("{"), 0o644); err != nil {
+				t.Fatalf("WriteFile() error = %v", err)
+			}
+			var err error
+			if name == "rejections" {
+				_, err = loadRejections(path)
+			} else if name == "warnings" {
+				_, err = loadWarnings(path)
+			} else {
+				_, _, _, err = loadDiagnostics(path)
+			}
+			if err == nil {
+				t.Fatal("summary decoder error = nil")
+			}
+		})
+	}
+	oversized := filepath.Join(root, "oversized.json")
+	if err := os.WriteFile(oversized, []byte(strings.Repeat("x", maxSummaryBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized) error = %v", err)
+	}
+	if _, err := loadWarnings(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadWarnings(oversized) error = %v", err)
+	}
+	if _, err := loadRejections(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadRejections(oversized) error = %v", err)
+	}
+	if _, _, _, err := loadDiagnostics(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadDiagnostics(oversized) error = %v", err)
+	}
+}
+
+func TestFakeRunnerCapturesRequestsAndHonorsContextAndError(t *testing.T) {
+	req := RunRequest{PipelineID: "pipeline", References: []ReferenceBinding{{Selector: "party", Path: "/references/party.json"}}}
+	want := RunResult{BundleRoot: "/bundle"}
+	fake := &FakeRunner{Result: want}
+	got, err := fake.Run(context.Background(), req)
+	if err != nil || !reflect.DeepEqual(got, want) || !reflect.DeepEqual(fake.Requests, []RunRequest{req}) {
+		t.Fatalf("Run() = %#v, %v; requests = %#v", got, err, fake.Requests)
+	}
+	req.References[0].Path = "/references/changed.json"
+	if fake.Requests[0].References[0].Path != "/references/party.json" {
+		t.Fatalf("fake retained aliased request references: %#v", fake.Requests[0].References)
+	}
+
+	wantErr := errors.New("configured failure")
+	fake.Err = wantErr
+	if _, err := fake.Run(context.Background(), req); !errors.Is(err, wantErr) {
+		t.Fatalf("Run(configured error) = %v", err)
+	}
+	canceled, cancel := context.WithCancel(context.Background())
+	cancel()
+	before := len(fake.Requests)
+	if _, err := fake.Run(canceled, req); !errors.Is(err, context.Canceled) || len(fake.Requests) != before {
+		t.Fatalf("Run(canceled) error = %v; requests = %d", err, len(fake.Requests))
+	}
+}
+
+func validRunRequest(t *testing.T) RunRequest {
+	t.Helper()
+	root := t.TempDir()
+	configPath := filepath.Join(root, "notarius.yml")
+	inputPath := filepath.Join(root, "input.json")
+	outputRoot := filepath.Join(root, "outputs")
+	workingDirectory := filepath.Join(root, "work")
+	diagnostics := filepath.Join(root, "diagnostics")
+	for _, directory := range []string{outputRoot, workingDirectory, diagnostics} {
+		if err := os.Mkdir(directory, 0o755); err != nil {
+			t.Fatalf("Mkdir(%q) error = %v", directory, err)
+		}
+	}
+	if err := os.WriteFile(configPath, []byte("pipelines: {}\n"), 0o644); err != nil {
+		t.Fatalf("WriteFile(config) error = %v", err)
+	}
+	if err := os.WriteFile(inputPath, []byte("{}\n"), 0o644); err != nil {
+		t.Fatalf("WriteFile(input) error = %v", err)
+	}
+	return RunRequest{
+		Binary: "notarius", ConfigPath: configPath, PipelineID: "dnd-session", InputPath: inputPath,
+		OutputRoot: outputRoot, WorkingDirectory: workingDirectory,
+		ReceiptPath: filepath.Join(diagnostics, "receipt.json"), LogPath: filepath.Join(diagnostics, "stderr.log"),
+		Timeout: time.Second,
+	}
+}
+
+func writeValidBundleAndReceipt(t *testing.T, req RunRequest, includeUnknown bool) {
+	t.Helper()
+	bundle := filepath.Join(req.OutputRoot, "notarius-run-1")
+	if err := os.MkdirAll(filepath.Join(bundle, "lanes"), 0o755); err != nil {
+		t.Fatalf("MkdirAll(bundle) error = %v", err)
+	}
+	for path, data := range map[string]string{
+		"manifest.json":         `{}`,
+		"lanes/npc.json":        `{}`,
+		"chunk-map.json":        `{}`,
+		"evidence-context.json": `{}`,
+	} {
+		if err := os.WriteFile(filepath.Join(bundle, filepath.FromSlash(path)), []byte(data), 0o644); err != nil {
+			t.Fatalf("WriteFile(%q) error = %v", path, err)
+		}
+	}
+	rejection := map[string]any{"stage": "validate", "lane_id": "spells", "reason_code": "invalid_spell", "message": strings.Repeat("external detail", 20)}
+	if includeUnknown {
+		rejection["future"] = true
+	}
+	writeJSONFile(t, filepath.Join(bundle, "rejected.json"), map[string]any{"rejected": []any{rejection}, "future": true})
+	writeJSONFile(t, filepath.Join(bundle, "warnings.json"), findingEnvelope(warningsSchemaVersion, []any{findingGroup("warning", "degradation", "normalized_name", "normalize", 2)}, 2, false, 0))
+	writeJSONFile(t, filepath.Join(bundle, "diagnostics.json"), findingEnvelope(diagnosticsSchemaVersion, []any{findingGroup("advisory", "data_quality", "low_confidence", "normalize", 3)}, 4, true, 1))
+	index := validIndexValue([]any{map[string]any{
+		"lane_id": "npc-registry", "file": "lanes/npc.json", "media_type": "application/json",
+		"module_key": "dnd/npc-registry", "schema_id": "notarius.dnd.npc_registry",
+		"schema_name": "NPCRegistry", "schema_version": "v1", "future": true,
+	}})
+	index["chunk_map"] = map[string]any{
+		"artifact_kind": "chunk_map", "file": "chunk-map.json", "media_type": "application/json",
+		"schema_id": "notarius.chunk_map", "schema_name": "ChunkMap", "schema_version": "v1", "future": true,
+	}
+	index["evidence_context"] = map[string]any{
+		"artifact_kind": "evidence_context", "file": "evidence-context.json", "media_type": "application/json",
+		"schema_id": "notarius.evidence_context", "schema_name": "EvidenceContext", "schema_version": "v1", "future": true,
+	}
+	index["future"] = true
+	writeJSONFile(t, filepath.Join(bundle, "index.json"), index)
+	receipt := map[string]any{
+		"schema_version": ReceiptSchemaVersion, "run_id": "notarius-run-1", "pipeline_id": req.PipelineID,
+		"output_directory": bundle, "index_file": "index.json", "normalized_output_count": 1,
+		"rejected_output_count": 1, "warning_group_count": 1, "warning_occurrence_count": 2,
+		"diagnostic_group_count": 1, "diagnostic_occurrence_count": 4,
+		"diagnostics_truncated": true, "validation_status": "rejected",
+		"validation_summaries": []any{map[string]any{
+			"stage": "normalize", "lane_id": "npc-registry", "status": "complete",
+			"producer_attempt_count": 1, "terminal_action": "accepted",
+		}},
+	}
+	if includeUnknown {
+		receipt["future"] = true
+	}
+	writeJSONFile(t, req.ReceiptPath, receipt)
+}
+
+func createBundleSkeleton(t *testing.T) string {
+	t.Helper()
+	bundle := filepath.Join(t.TempDir(), "bundle")
+	if err := os.MkdirAll(filepath.Join(bundle, "lanes"), 0o755); err != nil {
+		t.Fatalf("MkdirAll(bundle) error = %v", err)
+	}
+	for _, name := range []string{"manifest.json", "rejected.json", "warnings.json", "diagnostics.json", "lanes/npc.json", "chunk-map.json"} {
+		if err := os.WriteFile(filepath.Join(bundle, filepath.FromSlash(name)), []byte("{}"), 0o644); err != nil {
+			t.Fatalf("WriteFile(%q) error = %v", name, err)
+		}
+	}
+	return bundle
+}
+
+func validIndexValue(lanes []any) map[string]any {
+	return map[string]any{
+		"manifest_file": "manifest.json", "output_files": lanes,
+		"rejected_file": "rejected.json", "warnings_file": "warnings.json", "diagnostics_file": "diagnostics.json",
+	}
+}
+
+func findingGroup(disposition, category, reasonCode, origin string, occurrences int) map[string]any {
+	return map[string]any{
+		"disposition": disposition, "category": category, "reason_code": reasonCode,
+		"origin": map[string]any{"stage": origin, "lane_id": "npc-registry"}, "occurrence_count": occurrences,
+		"samples":              []any{map[string]any{"scope": "lane:npc-registry", "message": "external detail"}},
+		"omitted_sample_count": occurrences - 1,
+	}
+}
+
+func findingEnvelope(schema string, groups []any, occurrences int, truncated bool, unrepresented int) map[string]any {
+	value := map[string]any{
+		"schema_version": schema, "group_count": len(groups), "occurrence_count": occurrences,
+		"groups": groups,
+	}
+	if schema == diagnosticsSchemaVersion {
+		value["truncated"] = truncated
+		value["unrepresented_occurrence_count"] = unrepresented
+	}
+	return value
+}
+
+func writeJSONFile(t *testing.T, path string, value any) {
+	t.Helper()
+	data, err := json.Marshal(value)
+	if err != nil {
+		t.Fatalf("json.Marshal() error = %v", err)
+	}
+	if err := os.WriteFile(path, data, 0o644); err != nil {
+		t.Fatalf("WriteFile(%q) error = %v", path, err)
+	}
+}
+
+func writeShellScript(t *testing.T, body string) string {
+	t.Helper()
+	path := filepath.Join(t.TempDir(), "notarius-helper")
+	if err := os.WriteFile(path, []byte(body), 0o755); err != nil {
+		t.Fatalf("WriteFile(script) error = %v", err)
+	}
+	return path
+}
+
+func assertTextFile(t *testing.T, path, want string) {
+	t.Helper()
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("ReadFile(%q) error = %v", path, err)
+	}
+	if string(data) != want {
+		t.Fatalf("ReadFile(%q) = %q, want %q", path, string(data), want)
+	}
+}
+
+func cloneMap(source map[string]any) map[string]any {
+	result := make(map[string]any, len(source))
+	for key, value := range source {
+		result[key] = value
+	}
+	return result
+}

internal/adapters/scriptorium/fake.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"fmt"
 
-	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op scriptorium adapter.
@@ -142,7 +142,7 @@ func (f *FakeRunner) RenderArtifact(ctx context.Context, req RenderArtifactReque
 
 func materializeRunPlaceholders(req RunArtifactRequest) error {
 	if req.OutputPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("scriptorium noop/fake run artifact\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.OutputPath, []byte("scriptorium noop/fake run artifact\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write run output %q: %w", req.OutputPath, err)
 		}
 	}
@@ -154,17 +154,17 @@ func materializeRunPlaceholders(req RunArtifactRequest) error {
 			"prompt_id":   req.PromptID,
 			"output_path": req.OutputPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake run stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake run stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake run stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake run stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -173,7 +173,7 @@ func materializeRunPlaceholders(req RunArtifactRequest) error {
 
 func materializeRenderPlaceholders(req RenderArtifactRequest) error {
 	if req.OutputPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("{\"schema\":\"scriptorium.render.v1\",\"placeholder\":true}\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.OutputPath, []byte("{\"schema\":\"scriptorium.render.v1\",\"placeholder\":true}\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write render output %q: %w", req.OutputPath, err)
 		}
 	}
@@ -185,17 +185,17 @@ func materializeRenderPlaceholders(req RenderArtifactRequest) error {
 			"prompt_id":   req.PromptID,
 			"output_path": req.OutputPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake render stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake render stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake render stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake render stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}

internal/adapters/scriptorium/generated_config.go

@@ -0,0 +1,19 @@
+package scriptorium
+
+import (
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+	"gopkg.in/yaml.v3"
+)
+
+func writeGeneratedYAML(path string, payload any) error {
+	data, err := yaml.Marshal(payload)
+	if err != nil {
+		return fmt.Errorf("marshal generated YAML for %q: %w", path, err)
+	}
+	if err := fileops.WriteFileAtomic(path, data, fileops.WorkspaceFileMode); err != nil {
+		return fmt.Errorf("write generated YAML %q: %w", path, err)
+	}
+	return nil
+}

internal/adapters/scriptorium/subprocess.go

@@ -9,8 +9,12 @@ import (
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
+// MaxOutputFileBytes bounds one Scriptorium artifact result.
+const MaxOutputFileBytes int64 = 64 * 1024 * 1024
+
 // SubprocessRunner invokes Scriptorium through its public CLI.
 type SubprocessRunner struct{}
 
@@ -52,13 +56,17 @@ func (r *SubprocessRunner) RunArtifact(ctx context.Context, req RunArtifactReque
 		}
 	}
 
+	envOverrides, sensitiveNames := credentialEnvironment(req.APIKeyEnv)
 	runRes, runErr := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    req.Binary,
-		Args:          args,
-		WorkingDir:    req.WorkingDir,
-		Timeout:       req.Timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:        req.Binary,
+		Args:              args,
+		WorkingDir:        req.WorkingDir,
+		Timeout:           req.Timeout,
+		EnvOverrides:      envOverrides,
+		SensitiveEnvNames: sensitiveNames,
+		DiagnosticOwner:   "scriptorium",
+		StdoutLogPath:     req.StdoutLogPath,
+		StderrLogPath:     req.StderrLogPath,
 	})
 
 	result := ArtifactResult{
@@ -134,13 +142,17 @@ func (r *SubprocessRunner) RenderArtifact(ctx context.Context, req RenderArtifac
 		}
 	}
 
+	envOverrides, sensitiveNames := credentialEnvironment(req.APIKeyEnv)
 	runRes, runErr := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    req.Binary,
-		Args:          args,
-		WorkingDir:    req.WorkingDir,
-		Timeout:       req.Timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:        req.Binary,
+		Args:              args,
+		WorkingDir:        req.WorkingDir,
+		Timeout:           req.Timeout,
+		EnvOverrides:      envOverrides,
+		SensitiveEnvNames: sensitiveNames,
+		DiagnosticOwner:   "scriptorium",
+		StdoutLogPath:     req.StdoutLogPath,
+		StderrLogPath:     req.StderrLogPath,
 	})
 
 	result := ArtifactResult{
@@ -223,6 +235,15 @@ func validateCommonRunRequest(
 	return true, nil
 }
 
+func credentialEnvironment(apiKeyEnv string) (map[string]string, []string) {
+	name := strings.TrimSpace(apiKeyEnv)
+	if name == "" {
+		return nil, nil
+	}
+	value, _ := os.LookupEnv(name)
+	return map[string]string{name: value}, []string{name}
+}
+
 func buildRunArgs(req RunArtifactRequest) []string {
 	args := []string{"run", "--prompt", strings.TrimSpace(req.PromptID)}
 	if cfgPath := strings.TrimSpace(req.ConfigPath); cfgPath != "" {
@@ -321,18 +342,15 @@ func writeInvocationConfig(path string, payload invocationPayload) error {
 		"render_format":        payload.RenderFormat,
 		"render_prompt_logged": payload.RenderPromptStore,
 	}
-	return subprocess.WriteYAMLAtomic(path, data, 0o644)
+	return writeGeneratedYAML(path, data)
 }
 
 func validateNonEmptyOutput(path string) error {
-	info, err := os.Stat(path)
+	data, err := fileops.ReadFileBounded(path, MaxOutputFileBytes)
 	if err != nil {
-		return fmt.Errorf("stat file: %w", err)
+		return fmt.Errorf("scriptorium artifact output exceeds or cannot be read within %d-byte limit: %w", MaxOutputFileBytes, err)
 	}
-	if info.IsDir() {
-		return fmt.Errorf("path is a directory")
-	}
-	if info.Size() <= 0 {
+	if len(data) == 0 {
 		return fmt.Errorf("file is empty")
 	}
 	return nil

internal/adapters/scriptorium/subprocess_test.go

@@ -6,16 +6,12 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
 func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "run_success")
@@ -31,7 +27,7 @@ func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *tes
 		ConfigPath:          "/etc/scriptorium/config.yml",
 		PromptID:            "dnd.session_recap",
 		ProfileID:           "local-quality",
-		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "processed.json"), "other": filepath.Join(dir, "other.md")},
+		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "polished.json"), "other": filepath.Join(dir, "other.md")},
 		Vars:                map[string]string{"session_id": "2026-05-03", "campaign_name": "Icewind Dale"},
 		OutputPath:          filepath.Join(dir, "artifacts", "session_recap.md"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "scriptorium.run.stdout.log"),
@@ -95,9 +91,6 @@ func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *tes
 }
 
 func TestSubprocessRunnerRunExitCodeOneFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "fail_exit1")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -118,9 +111,6 @@ func TestSubprocessRunnerRunExitCodeOneFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRunExitCodeTwoReturnsValidationFailureAndPreservesOutput(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "fail_exit2_with_output")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -144,9 +134,6 @@ func TestSubprocessRunnerRunExitCodeTwoReturnsValidationFailureAndPreservesOutpu
 }
 
 func TestSubprocessRunnerRunMissingOutputOnSuccessFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "run_success_missing_output")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -166,9 +153,6 @@ func TestSubprocessRunnerRunMissingOutputOnSuccessFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderSuccess(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "render_success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
@@ -180,7 +164,7 @@ func TestSubprocessRunnerRenderSuccess(t *testing.T) {
 	req := RenderArtifactRequest{
 		Binary:              wrapper,
 		PromptID:            "dnd.session_recap",
-		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "processed.json")},
+		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "polished.json")},
 		OutputPath:          filepath.Join(dir, "artifacts", "session_recap.render.json"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "scriptorium.render.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "scriptorium.render.stderr.log"),
@@ -285,7 +269,7 @@ type scriptoriumHelperRecord struct {
 func runReqForTest(t *testing.T, binary string) RunArtifactRequest {
 	t.Helper()
 	dir := t.TempDir()
-	transcriptPath := filepath.Join(dir, "processed.json")
+	transcriptPath := filepath.Join(dir, "polished.json")
 	writeScriptoriumFile(t, transcriptPath, `{"segments":[]}`)
 	return RunArtifactRequest{
 		Binary:              binary,

internal/adapters/seriatim/fake.go

@@ -4,7 +4,7 @@ import (
 	"context"
 	"fmt"
 
-	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op seriatim adapter.
@@ -68,6 +68,26 @@ func (n *NoopRunner) Normalize(ctx context.Context, req NormalizeRequest) (Norma
 	}, nil
 }
 
+// Render returns the requested output path with placeholder metadata.
+func (n *NoopRunner) Render(ctx context.Context, req RenderRequest) (RenderResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RenderResult{}, err
+	}
+	if err := materializeRenderPlaceholders(req); err != nil {
+		return RenderResult{}, err
+	}
+	return RenderResult{
+		OutputRenderedPath:  req.OutputRenderedPath,
+		StdoutLogPath:       req.StdoutLogPath,
+		StderrLogPath:       req.StderrLogPath,
+		GeneratedConfigPath: req.GeneratedConfigPath,
+		InvokedBinary:       "noop",
+		Format:              req.Format,
+		Title:               req.Title,
+		Metadata:            map[string]any{"placeholder": true},
+	}, nil
+}
+
 // FakeRunner captures merge requests and returns deterministic responses.
 type FakeRunner struct {
 	Requests          []MergeRequest
@@ -79,6 +99,9 @@ type FakeRunner struct {
 	TrimRequests      []TrimRequest
 	TrimErr           error
 	TrimResult        TrimResult
+	RenderRequests    []RenderRequest
+	RenderErr         error
+	RenderResult      RenderResult
 }
 
 // Run records request and returns configured response.
@@ -195,9 +218,49 @@ func (f *FakeRunner) Normalize(ctx context.Context, req NormalizeRequest) (Norma
 	return res, nil
 }
 
+// Render records request and returns configured response.
+func (f *FakeRunner) Render(ctx context.Context, req RenderRequest) (RenderResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RenderResult{}, err
+	}
+	f.RenderRequests = append(f.RenderRequests, req)
+	if f.RenderErr != nil {
+		return RenderResult{}, f.RenderErr
+	}
+	if err := materializeRenderPlaceholders(req); err != nil {
+		return RenderResult{}, err
+	}
+	res := f.RenderResult
+	if res.OutputRenderedPath == "" {
+		res.OutputRenderedPath = req.OutputRenderedPath
+	}
+	if res.StdoutLogPath == "" {
+		res.StdoutLogPath = req.StdoutLogPath
+	}
+	if res.StderrLogPath == "" {
+		res.StderrLogPath = req.StderrLogPath
+	}
+	if res.GeneratedConfigPath == "" {
+		res.GeneratedConfigPath = req.GeneratedConfigPath
+	}
+	if res.InvokedBinary == "" {
+		res.InvokedBinary = "fake"
+	}
+	if res.Format == "" {
+		res.Format = req.Format
+	}
+	if res.Title == "" {
+		res.Title = req.Title
+	}
+	if res.Metadata == nil {
+		res.Metadata = map[string]any{"fake": true}
+	}
+	return res, nil
+}
+
 func materializePlaceholders(req MergeRequest) error {
 	if req.OutputMergedTranscriptPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputMergedTranscriptPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.OutputMergedTranscriptPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write merged transcript %q: %w", req.OutputMergedTranscriptPath, err)
 		}
 	}
@@ -208,22 +271,22 @@ func materializePlaceholders(req MergeRequest) error {
 			"input_transcript_paths": req.InputTranscriptPaths,
 			"output_path":            req.OutputMergedTranscriptPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
 	if req.ReportPath != "" {
-		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write report %q: %w", req.ReportPath, err)
 		}
 	}
@@ -232,7 +295,7 @@ func materializePlaceholders(req MergeRequest) error {
 
 func materializeTrimPlaceholders(req TrimRequest) error {
 	if req.OutputTrimmedPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputTrimmedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.OutputTrimmedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write trimmed transcript %q: %w", req.OutputTrimmedPath, err)
 		}
 	}
@@ -245,17 +308,17 @@ func materializeTrimPlaceholders(req TrimRequest) error {
 			"output_path":   req.OutputTrimmedPath,
 			"keep_selector": req.KeepSelector,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake trim stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake trim stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake trim stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake trim stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -264,7 +327,7 @@ func materializeTrimPlaceholders(req TrimRequest) error {
 
 func materializeNormalizePlaceholders(req NormalizeRequest) error {
 	if req.OutputNormalizedPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputNormalizedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.OutputNormalizedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write normalized transcript %q: %w", req.OutputNormalizedPath, err)
 		}
 	}
@@ -280,24 +343,60 @@ func materializeNormalizePlaceholders(req NormalizeRequest) error {
 		if req.ReportPath != "" {
 			payload["report_path"] = req.ReportPath
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake normalize stdout placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake normalize stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake normalize stderr placeholder\n"), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake normalize stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
 	if req.ReportPath != "" {
-		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), 0o644); err != nil {
+		if err := fileops.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write report %q: %w", req.ReportPath, err)
 		}
 	}
 	return nil
 }
+
+func materializeRenderPlaceholders(req RenderRequest) error {
+	if req.OutputRenderedPath != "" {
+		if err := fileops.WriteFileAtomic(req.OutputRenderedPath, []byte("# Transcript\n\nRendered markdown placeholder.\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write rendered transcript %q: %w", req.OutputRenderedPath, err)
+		}
+	}
+	if req.GeneratedConfigPath != "" {
+		payload := map[string]any{
+			"schema":              "seriatim.generated.v1",
+			"placeholder":         true,
+			"command":             "render",
+			"input_path":          req.InputTranscriptPath,
+			"output_path":         req.OutputRenderedPath,
+			"format":              req.Format,
+			"title":               req.Title,
+			"include_timestamps":  req.IncludeTimestamps,
+			"include_segment_ids": req.IncludeSegmentIDs,
+			"include_metadata":    req.IncludeMetadata,
+		}
+		if err := writeGeneratedYAML(req.GeneratedConfigPath, payload); err != nil {
+			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
+		}
+	}
+	if req.StdoutLogPath != "" {
+		if err := fileops.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake render stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
+		}
+	}
+	if req.StderrLogPath != "" {
+		if err := fileops.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake render stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
+		}
+	}
+	return nil
+}

internal/adapters/seriatim/fake_test.go

@@ -14,7 +14,7 @@ func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := MergeRequest{
 		GeneratedConfigPath:        filepath.Join(dir, "config", "seriatim.yml"),
-		OutputMergedTranscriptPath: filepath.Join(dir, "transcripts", "merged.json"),
+		OutputMergedTranscriptPath: filepath.Join(dir, "transcripts", "base.json"),
 		StdoutLogPath:              filepath.Join(dir, "logs", "seriatim.stdout.log"),
 		StderrLogPath:              filepath.Join(dir, "logs", "seriatim.stderr.log"),
 	}
@@ -57,8 +57,8 @@ func TestFakeRunnerTrimCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := TrimRequest{
 		GeneratedConfigPath: filepath.Join(dir, "config", "seriatim.trim.yml"),
-		InputTranscriptPath: filepath.Join(dir, "transcripts", "processed.json"),
-		OutputTrimmedPath:   filepath.Join(dir, "transcripts", "trimmed.json"),
+		InputTranscriptPath: filepath.Join(dir, "transcripts", "polished.json"),
+		OutputTrimmedPath:   filepath.Join(dir, "transcripts", "final.trimmed.json"),
 		KeepSelector:        "1-10",
 		StdoutLogPath:       filepath.Join(dir, "logs", "seriatim.trim.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "seriatim.trim.stderr.log"),
@@ -105,8 +105,8 @@ func TestFakeRunnerNormalizeCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := NormalizeRequest{
 		GeneratedConfigPath:  filepath.Join(dir, "config", "seriatim.normalize.yml"),
-		InputTranscriptPath:  filepath.Join(dir, "transcripts", "processed.json"),
-		OutputNormalizedPath: filepath.Join(dir, "transcripts", "normalized.json"),
+		InputTranscriptPath:  filepath.Join(dir, "transcripts", "polished.json"),
+		OutputNormalizedPath: filepath.Join(dir, "transcripts", "final.json"),
 		OutputSchema:         "seriatim-intermediate",
 		ReportPath:           filepath.Join(dir, "artifacts", "seriatim.normalize.report.json"),
 		StdoutLogPath:        filepath.Join(dir, "logs", "seriatim.normalize.stdout.log"),
@@ -148,3 +148,58 @@ func TestFakeRunnerNormalizeError(t *testing.T) {
 		t.Fatal("expected error, got nil")
 	}
 }
+
+func TestFakeRunnerRenderCapturesRequestAndReturnsPath(t *testing.T) {
+	fake := &FakeRunner{}
+	dir := t.TempDir()
+	req := RenderRequest{
+		GeneratedConfigPath: filepath.Join(dir, "config", "seriatim.render.yml"),
+		InputTranscriptPath: filepath.Join(dir, "transcripts", "final.trimmed.json"),
+		OutputRenderedPath:  filepath.Join(dir, "transcripts", "final.trimmed.md"),
+		Format:              "markdown",
+		Title:               "Session render",
+		IncludeTimestamps:   true,
+		IncludeSegmentIDs:   false,
+		IncludeMetadata:     true,
+		StdoutLogPath:       filepath.Join(dir, "logs", "seriatim.render.stdout.log"),
+		StderrLogPath:       filepath.Join(dir, "logs", "seriatim.render.stderr.log"),
+	}
+
+	res, err := fake.Render(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Render() error = %v", err)
+	}
+	if len(fake.RenderRequests) != 1 || fake.RenderRequests[0].GeneratedConfigPath == "" {
+		t.Fatalf("render requests = %#v, want captured request", fake.RenderRequests)
+	}
+	if res.OutputRenderedPath != req.OutputRenderedPath {
+		t.Fatalf("rendered path = %q, want %q", res.OutputRenderedPath, req.OutputRenderedPath)
+	}
+	if res.Format != req.Format {
+		t.Fatalf("format = %q, want %q", res.Format, req.Format)
+	}
+	if res.Title != req.Title {
+		t.Fatalf("title = %q, want %q", res.Title, req.Title)
+	}
+
+	cfgData, err := os.ReadFile(req.GeneratedConfigPath)
+	if err != nil {
+		t.Fatalf("read generated config: %v", err)
+	}
+	if !strings.Contains(string(cfgData), "command: render") {
+		t.Fatalf("generated config = %q, want render command marker", string(cfgData))
+	}
+	for _, path := range []string{req.StdoutLogPath, req.StderrLogPath, req.OutputRenderedPath} {
+		if _, err := os.Stat(path); err != nil {
+			t.Fatalf("expected file %q to exist: %v", path, err)
+		}
+	}
+}
+
+func TestFakeRunnerRenderError(t *testing.T) {
+	fake := &FakeRunner{RenderErr: errors.New("boom")}
+	_, err := fake.Render(context.Background(), RenderRequest{})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+}

internal/adapters/seriatim/generated_config.go

@@ -0,0 +1,19 @@
+package seriatim
+
+import (
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+	"gopkg.in/yaml.v3"
+)
+
+func writeGeneratedYAML(path string, payload any) error {
+	data, err := yaml.Marshal(payload)
+	if err != nil {
+		return fmt.Errorf("marshal generated YAML for %q: %w", path, err)
+	}
+	if err := fileops.WriteFileAtomic(path, data, fileops.WorkspaceFileMode); err != nil {
+		return fmt.Errorf("write generated YAML %q: %w", path, err)
+	}
+	return nil
+}