Close the completed roadmap documents

README.md

@@ -1,22 +1,42 @@
 # narratio
 
-Narratio is a stage-driven Go orchestrator for turning D&D session audio into polished 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.
 
-It runs a deterministic workflow across `prepare`, `transcribe`, `merge`, `polish`, `normalize`, `trim`, `render`, `analyze`, and `publish`, with manifest-driven continuation and restore support.
+It runs a deterministic workflow with manifest-driven continuation, remote
+publish, and restore support.
 
-```bash
+```sh
 narratio run 2026-04-04
 ```
 
-This requires resolvable `pipeline.yml`, `campaign.yml`, and concrete `session.yml` (or explicit config flags).
+This requires resolvable `pipeline.yml`, `campaign.yml`, and concrete
+`session.yml` files or their explicit command-line alternatives.
 
 ## Documentation
 
-- [CLI Reference](docs/cli.md)
-- [Configuration](docs/config.md)
-- [Operations](docs/operations.md)
-- [Troubleshooting](docs/troubleshooting.md)
-- [Internal Component Contracts](docs/internal/README.md)
-- [Development Guide](docs/policy/development.md)
-- [Architecture Principles](docs/policy/architecture.md)
-- [Maintained Examples](examples/)
+- [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.
+
+## Maintainer Documentation
+
+- [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.

docs/cli.md

@@ -75,7 +75,10 @@ narratio run <session_id> [--force] [--artifacts <name[,name...]>] [...common co
 Behavior:
 
 - evaluates full stage order;
-- skips already-succeeded stages unless `--force` is set;
+- runs `extract` between `trim` and `render`; 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;
 - continues interrupted or partially completed sessions by running non-succeeded stages;
 - writes session and run manifests.
 
@@ -93,6 +96,7 @@ Valid stage names:
 - `polish`
 - `normalize`
 - `trim`
+- `extract`
 - `render`
 - `analyze`
 - `publish`
@@ -135,14 +139,13 @@ narratio clean --all [--dry-run] [--clear-cache] [--config <pipeline.yml>]
 
 Behavior:
 
-- session mode removes:
-  - `{workspace.root}/work/{campaign}/{session_id}`
-  - `{spool.root}/{campaign}/{session_id}`
-- `--all` removes:
-  - `{workspace.root}/work/*`
-  - direct children under `{spool.root}`
+- 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 deletion scope and
+post-publish cleanup behavior.
+
 ### `session plan`
 
 ```bash
@@ -207,17 +210,11 @@ Behavior:
 
 - discovers committed remote current state;
 - plans local restores;
-- writes `reports/restore-latest.json` on execution;
+- writes an execution report;
 - blocks conflicting overwrites unless `--force` is set.
 
-Default restore scope:
-
-- `manifest.json`
-- `transcripts/**`
-- `artifacts/**`
-- `previous/**` when required by configured previous-session inputs
-
-`audio/**` is included only with `--include-audio`.
+See [Operations: Restore Workflow](./operations.md#restore-workflow) for the
+default restore scope, report location, and conflict-handling workflow.
 
 ### `session artifacts`
 
@@ -225,7 +222,10 @@ Default restore scope:
 narratio session artifacts <session_id> [--remote] [...common config flags]
 ```
 
-Lists effective built-in and configured artifact sources, publish rules, lock state, and optional remote published-state availability.
+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.
 
 ### `session locks`
 
@@ -237,10 +237,13 @@ narratio session locks remove <session_id> <source> [...common config flags]
 
 Behavior:
 
-- list mode merges static `pipeline.publish.locks` with remote `{session_prefix}/locks.yml`;
+- list mode reports the effective merge of static and remote locks;
 - add/remove mutate only remote locks;
 - static locks from pipeline config cannot be removed by CLI commands.
 
+See [Operations: Publish Locks](./operations.md#publish-locks) for lock storage
+and precedence.
+
 ## `--artifacts` Selection Rules
 
 - accepted on `run`, `run-stage`, `analyze`, and `publish`;
@@ -252,7 +255,9 @@ Effects:
 
 - filters analyze execution to selected configured artifacts;
 - filters publish rules that source `narratio.artifact.<name>`;
-- does not filter built-in transcript/bounds publish sources.
+- 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
 
@@ -279,3 +284,18 @@ Force publish only:
 ```bash
 narratio publish 2026-04-04
 ```
+
+## 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

@@ -118,6 +118,9 @@ 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`.
 - static locks (`pipeline.publish.locks`) merge with remote locks (`{session_prefix}/locks.yml`), with static locks taking precedence on duplicates.
 
@@ -196,6 +199,13 @@ Rules:
 | `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.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) |
@@ -209,7 +219,27 @@ Rules:
 | `pipeline.scriptorium.artifacts` | map | No | empty |
 | `pipeline.notification.backend` | string | No | empty |
 | `pipeline.notification.recipient` | string | No | empty |
-| `pipeline.notification.timeout` | duration | No | `30s` |
+| `pipeline.notification.timeout` | duration | No | empty |
+
+### 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
 
@@ -225,13 +255,15 @@ For each `pipeline.scriptorium.artifacts.<name>`:
 | `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 |
+| `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.
 
 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.artifact.<name>`, or `narratio.previous_session.artifact.<name>` |
+| `source` | string | Yes | built-in runtime source, prepared input source, `narratio.extraction.<name>`, `narratio.artifact.<name>`, or `narratio.previous_session.artifact.<name>` |
 | `artifact` | string | No | optional passthrough adapter field |
 | `path` | string | No | optional passthrough adapter field |
 | `required` | bool | No | optional input requirement |
@@ -272,10 +304,6 @@ Audio rules:
 
 ## Maintained Examples
 
-- `examples/pipeline.minimal.yml`
-- `examples/pipeline.production.yml`
-- `examples/pipeline.full.annotated.yml`
-- `examples/campaigns/sample-campaign/campaign.yml`
-- `examples/session.local-audio.yml`
-- `examples/session.s3-audio.yml`
-- `examples/session.template.yml`
+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 +1,43 @@
+# 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.
+
+## 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. |
+| Proposed or unimplemented behavior | [Roadmap](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. Run the repository-wide checks when a
+change affects shared contracts, application behavior, or maintained
+documentation examples:
+
+```sh
+go test ./...
+go vet ./...
+go build ./cmd/narratio
+```

docs/integrations/README.md

@@ -1,19 +1,35 @@
 # Integrations Index
 
 ## Audience
-Developers and coding agents changing Narratio's external integration boundaries.
+
+Operators, developers, and coding agents who need to understand Narratio's
+externally observable integration boundaries.
 
 ## Scope
-`docs/integrations/` is the implementation-level reference for downstream tool adapter contracts.
 
-These docs cover what Narratio expects from external tools and what each adapter guarantees back to stage code.
+`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.md`: transcript polishing adapter (`audita process`).
-- `seriatim.md`: merge/normalize/trim/render adapter (`seriatim`).
-- `scriptorium.md`: artifact run/render adapter (`scriptorium run|render`).
+
+- [Audita](./audita.md): transcript polishing (`audita process`).
+- [Notarius](./notarius.md): complete pipeline execution and safe JSON bundle
+  discovery (`notarius run`).
+- [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
-- `docs/config.md`: operator-facing configuration reference.
-- `docs/internal/adapters.md`: shared adapter boundary and runner wiring.
-- `docs/internal/stage-*.md`: stage-specific integration usage.
+
+- [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

@@ -3,16 +3,12 @@
 ## Purpose
 Define the Audita adapter contract used by the `polish` stage.
 
-## Adapter Boundary
-Interface:
-- `audita.Runner`
-- method: `Run(ctx, PolishRequest) (PolishResult, error)`
+## External Boundary
 
-Primary implementation:
-- `internal/adapters/audita/SubprocessRunner`
-
-Execution mode:
-- subprocess invocation of `audita process`
+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).
 
 ## Request Contract
 `PolishRequest` carries:
@@ -52,9 +48,12 @@ Failure results still include output/log/config/exit metadata for diagnostics.
 - Generated invocation YAML (`audita.generated.v1`) is emitted when requested.
 - Manifest writes are stage-owned; adapter itself is stateless.
 
-## Config Mapping
-Config fields consumed through runner/stage wiring are under `pipeline.audita.*`.
+## Configuration
+
+Operator-selected values are defined under `pipeline.audita.*` in the
+[configuration reference](../config.md#pipeline).
 
 Maintained example with Audita config:
-- `examples/pipeline.full.annotated.yml`
-- `examples/pipeline.production.yml`
+
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/notarius.md

@@ -0,0 +1,83 @@
+# 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 references:
+
+- [Subprocess consumer contract](https://gitea.maximumdirect.net/eric/notarius/src/branch/main/docs/consumers/subprocess.md)
+- [D&D pipeline and lane contracts](https://gitea.maximumdirect.net/eric/notarius/src/branch/main/docs/consumers/dnd-pipeline.md)
+- [Run-result receipt](https://gitea.maximumdirect.net/eric/notarius/src/branch/main/docs/integrations/run-result.md)
+- [JSON output bundle](https://gitea.maximumdirect.net/eric/notarius/src/branch/main/docs/integrations/json-output.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 and invokes:
+
+```text
+notarius run <pipeline_id> --config <config_path> --input <trimmed_json> --output-dir <staging_dir> --json
+```
+
+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 inherited environment
+apply to the subprocess.
+
+## Accepted Result
+
+Narratio currently accepts receipt schema `notarius.run-result.v1`. 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`, and
+`warnings.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, and lane-list fields remain mandatory. Narratio applies
+bounded reads to the receipt, index, rejection, and warning documents. 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.
+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.
+
+Rejection and warning summaries retain structured stage, scope, lane, and
+reason-code fields for diagnostics without exposing free-form external messages
+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/scriptorium.md

@@ -3,20 +3,17 @@
 ## Purpose
 Define the Scriptorium adapter contract used by `analyze` and trim-bounds generation in `trim`.
 
-## Adapter Boundary
-Interface:
-- `scriptorium.Runner`
-- methods:
-  - `RunArtifact(ctx, RunArtifactRequest)`
-  - `RenderArtifact(ctx, RenderArtifactRequest)`
+## External Boundary
 
-Primary implementation:
-- `internal/adapters/scriptorium/SubprocessRunner`
+Narratio invokes Scriptorium as a subprocess in these modes:
 
-Execution modes:
 - `scriptorium run`
 - `scriptorium render`
 
+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;
@@ -55,12 +52,17 @@ Render behavior:
 
 ## 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.
 
-## Config Mapping
-Config fields consumed through runner/stage wiring are under `pipeline.scriptorium.*` plus per-artifact settings under `pipeline.scriptorium.artifacts.*`.
+## 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:
-- `examples/pipeline.full.annotated.yml`
-- `examples/pipeline.production.yml`
+
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/seriatim.md

@@ -3,24 +3,19 @@
 ## Purpose
 Define the Seriatim adapter contract used by `merge`, `normalize`, `trim`, and `render`.
 
-## Adapter Boundary
-Interface:
-- `seriatim.Runner`
-- methods:
-  - `Run(ctx, MergeRequest)`
-  - `Normalize(ctx, NormalizeRequest)`
-  - `Trim(ctx, TrimRequest)`
-  - `Render(ctx, RenderRequest)`
+## External Boundary
 
-Primary implementation:
-- `internal/adapters/seriatim/SubprocessRunner`
+Narratio invokes Seriatim as a subprocess in these modes:
 
-Execution modes:
 - `seriatim merge`
 - `seriatim normalize`
 - `seriatim trim`
 - `seriatim render`
 
+The configured timeout and parent cancellation bound each invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
+
 ## Request/Result Contracts
 - `MergeRequest`/`MergeResult`: multi-input merge to base transcript, optional report.
 - `NormalizeRequest`/`NormalizeResult`: transcript normalization with explicit schema.
@@ -53,9 +48,13 @@ When report paths are provided/enabled, report files must parse as JSON.
 - generated invocation YAML (`seriatim.generated.v1`) is emitted when requested.
 - adapter does not write manifests or choose stage inputs.
 
-## Config Mapping
-Config fields consumed through runner/stage wiring are under `pipeline.seriatim.*` and `pipeline.render.*`.
+## Configuration
+
+Operator-selected values are defined under `pipeline.seriatim.*` and
+`pipeline.render.*` in the
+[configuration reference](../config.md#pipeline).
 
 Maintained examples with Seriatim config:
-- `examples/pipeline.full.annotated.yml`
-- `examples/pipeline.production.yml`
+
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/whisperx.md

@@ -0,0 +1,66 @@
+# 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.
+
+## 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 invalid 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/README.md

@@ -1,45 +0,0 @@
-# Internal Documentation Index
-
-## Audience
-Developers and coding agents changing Narratio internals.
-
-## Scope
-`docs/internal/` documents implemented internal contracts: stage boundaries, manifest/state behavior, artifact resolution, restore behavior, storage boundaries, and workspace invariants.
-
-User and operator behavior belongs in:
-- `docs/cli.md`
-- `docs/config.md`
-- `docs/operations.md`
-- `docs/troubleshooting.md`
-
-## Pipeline Stage Set
-Canonical stage order from `internal/stage.All()`:
-1. `prepare`
-2. `transcribe`
-3. `merge`
-4. `polish`
-5. `normalize`
-6. `trim`
-7. `render`
-8. `analyze`
-9. `publish`
-10. `notify` (placeholder)
-
-`notify` is currently a placeholder stage with optional notifier call behavior; it has no persisted pipeline outputs.
-
-## Internal Component Docs
-- `adapters.md`: external adapter boundaries and default runtime wiring.
-- `artifacts.md`: canonical source IDs, runtime catalog behavior, and resolution rules.
-- `manifest.md`: session and run manifest contracts.
-- `storage.md`: object-store interface and S3 implementation behavior.
-- `workspace.md`: local session layout, run-local layout, and cleanup guardrails.
-- `command-restore.md`: restore discovery, planning, execution, and reporting.
-- `stage-prepare.md`
-- `stage-transcribe.md`
-- `stage-merge.md`
-- `stage-polish.md`
-- `stage-normalize.md`
-- `stage-trim.md`
-- `stage-render.md`
-- `stage-analyze.md`
-- `stage-publish.md`

docs/internal/adapters.md

@@ -1,49 +1,78 @@
 # Internal: Adapters
 
 ## Purpose
-Define external integration boundaries and default adapter wiring used by app/stage orchestration.
+
+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
+
 `internal/app/runner.go` initializes default adapters when not injected:
+
 - WhisperX HTTP client from pipeline config.
 - Seriatim subprocess runner.
 - Audita subprocess runner.
 - Scriptorium subprocess runner.
+- Notarius subprocess runner when extraction is enabled.
 - Noop notifier (`notify.NoopSender`).
 - Object store only when required by selected stages/config.
 
-Object-store construction goes through `newCommandObjectStore`, which loads configured filesystem secrets before adapter initialization.
+Notarius is composed only when extraction is enabled; the extract stage owns
+receipt, bundle, and configured-lane policy rather than the adapter.
+
+Object-store construction goes through `newCommandObjectStore`, which loads
+configured filesystem secrets before adapter initialization.
 
 ## 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.
 
-## Test Surfaces
+## Implementation And Tests
+
+- Composition: `internal/app/runner.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

@@ -1,24 +1,38 @@
 # Internal: Artifacts
 
 ## Purpose
-Define canonical artifact IDs, runtime catalog behavior, source resolution rules, and shared current-state mechanics used by app and previous-cache code.
+
+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
 
-- `narratio.transcript.base` -> `transcripts/base.json` (`merge`)
-- `narratio.transcript.polished` -> `transcripts/polished.json` (`polish`)
-- `narratio.transcript.final` -> `transcripts/final.json` (`normalize`)
-- `narratio.transcript.final_trimmed` -> `transcripts/final.trimmed.json` (`trim`)
-- `narratio.transcript.final_markdown` -> `transcripts/final.md` (`render`)
-- `narratio.transcript.final_trimmed_markdown` -> `transcripts/final.trimmed.md` (`render`)
-- `narratio.bounds.session` -> `artifacts/session_bounds.json` (`trim`)
+The internal registry recognizes these stable built-in source IDs:
 
-## Configured and Previous-Session Sources
+- `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, canonical fallback,
+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>`
 
-Both formats are validated by strict source-policy rules.
+All formats are validated by strict source-policy rules. Extraction sources are
+registered only from `pipeline.notarius.outputs`; the Notarius index has no
+selectable source ID.
 
 ## Runtime Catalog
 
@@ -47,6 +61,15 @@ Configured sources (`narratio.artifact.*`):
 
 - resolve only through runtime catalog availability.
 
+Extraction sources (`narratio.extraction.*`):
+
+- use the shared registration and manifest hydration path in
+  `extraction_catalog.go`;
+- require a current successful extract record with the exact configured source,
+  compatible contract and Notarius provenance, a confined regular durable
+  payload, and matching checksum; and
+- are never inferred by scanning the Notarius bundle directory.
+
 Previous-session sources (`narratio.previous_session.artifact.*`):
 
 - resolve only from local `previous/` cache state;
@@ -72,7 +95,8 @@ Validation by content type:
 
 ## Current-State Helpers
 
-Artifacts package owns shared remote current-state loading mechanics used by restore, status/validate checks, and previous-cache planning.
+Artifacts package owns shared remote current-state loading mechanics used by
+restore, status and validation checks, and previous-cache planning.
 
 Core helpers:
 
@@ -107,9 +131,31 @@ Caller policy is intentionally outside artifacts helpers:
 - spool/cache paths;
 - S3 session/run/current-state key layout.
 
+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`
+- Current state: `internal/artifacts/current_state.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

@@ -1,21 +1,20 @@
 # Internal: Command Restore
 
 ## Purpose
-Define the implemented `narratio session restore` command contract:
 
-- committed remote current-state discovery;
-- deterministic restore planning;
-- safe local install semantics;
-- durable restore reporting.
+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 is split into explicit phases so remote authority, local conflict
+policy, and filesystem mutation can be tested independently.
 
 ## Discovery Contract
 
-Restore resolves remote committed state from the session publish current pointers:
-
-- `current/run_id.txt` (required, non-empty);
-- `current/manifest.json` (required, decodable).
-
-Current-state discovery uses shared artifacts-level mechanics and validates identity against the resolved request config:
+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.
@@ -34,26 +33,11 @@ Planner behavior:
 
 - remote list scope is the resolved session prefix;
 - remote-to-local mapping is traversal-safe;
-- actions are sorted deterministically by local relative path.
+- actions are sorted by local relative path and then remote key;
+- force converts differing local targets from conflicts to downloads.
 
-Restore scope from current remote state:
-
-- include `manifest.json`;
-- include `transcripts/**`;
-- include `artifacts/**`;
-- include `audio/**` only with `--include-audio`.
-
-Explicit exclusions from current remote state mapping:
-
-- `current/**`;
-- `runs/**`;
-- `logs/**`;
-- `reports/**`;
-- `config/**`;
-- `inputs/**`;
-- `previous/**`.
-
-Previous-cache files are planned separately through `previouscache.BuildPlan` when configured previous-session requirements exist.
+Previous-cache files are planned separately through `previouscache.BuildPlan`
+when configured previous-session requirements exist.
 
 ## Execution Contract
 
@@ -73,8 +57,9 @@ Audio restore path:
 
 ## Reporting Contract
 
-- `--dry-run`: prints summary only; no local writes.
-- non-dry-run: writes `reports/restore-latest.json`.
+- dry-run mode prints a summary and performs no local writes;
+- 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
@@ -82,3 +67,15 @@ Audio restore path:
 - restore uses committed remote current state as authority;
 - `current/run_id.txt` is the remote publish commit marker;
 - 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_workflow_test.go`

docs/internal/manifest.md

@@ -1,13 +1,15 @@
 # Internal: Manifest
 
 ## Purpose
-Define durable session state (`manifest.json`) and invocation state (`runs/{run_id}/manifest.json`) contracts.
+
+Explain the session-progress and invocation-audit models implemented by
+`internal/manifest`. Physical manifest placement belongs in
+[Operations](../operations.md#local-state-layout).
 
 ## Session Manifest
-Path:
-- `{workspace.root}/work/{campaign}/{session_id}/manifest.json`
 
-Primary model (`manifest.Manifest`):
+`manifest.Manifest` records:
+
 - 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`)
@@ -15,7 +17,8 @@ Primary model (`manifest.Manifest`):
 - durable `artifacts` records
 - per-stage `stages` map
 
-Stage status enum:
+The model admits these stage states:
+
 - `pending`
 - `running`
 - `succeeded`
@@ -25,10 +28,9 @@ Stage status enum:
 - `interrupted`
 
 ## Run Manifest
-Path:
-- `{workspace.root}/work/{campaign}/{session_id}/runs/{run_id}/manifest.json`
 
-Run model (`manifest.RunManifest`):
+`manifest.RunManifest` is created for each invocation and records:
+
 - invocation identity and `force` flag
 - requested stages
 - per-stage action (`run` or `skip`)
@@ -36,22 +38,63 @@ Run model (`manifest.RunManifest`):
 - overall run status (`running`, `succeeded`, `failed`)
 
 ## Persistence Semantics
+
 `manifest.LocalStore`:
+
 - validates loaded documents;
 - normalizes missing maps/stage records;
 - writes atomically via temp file + rename;
 - updates `updated_at` on save.
 
 ## Execution Semantics
-Runner updates both manifests per stage transition:
-- mark running
-- mark succeeded/failed/skipped
-- persist logs/generated config refs and metadata
+
+The application runner marks an executing stage running and then succeeded or
+failed in both manifests, persisting each transition. On success it records
+outputs, logs, generated configuration references, and metadata. 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 downstream session-stage records stale.
+
+Starting an execution clears the current session-stage record's prior outputs,
+logs, generated configuration references, and metadata. 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. Invocation run manifests remain immutable audit
+records of their own outcomes.
+
+A stage may explicitly return a skipped disposition and stable reason. The
+runner persists that outcome in both manifests, clears older outputs for the
+session-stage record along with older 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.
+
+When an already-succeeded stage is skipped, the invocation run manifest records
+the `skip` action and reason. The session manifest deliberately retains its
+existing succeeded record because it remains the cross-invocation progress
+authority. Stages with a resume validator, currently extraction, may reject an
+otherwise eligible skip when the recorded durable result is obsolete; the
+runner marks it stale and executes it.
 
 Session manifest is the authoritative stage-progress ledger across invocations.
 Run manifest is invocation-scoped audit state.
 
 ## Invariants
+
 - stage resume/skip decisions are session-manifest driven.
+- 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 downstream succeeded stages.
 - run manifest does not replace session manifest as progress authority.
+
+## Implementation And Tests
+
+- Models and transitions: `internal/manifest/manifest.go`,
+  `internal/manifest/run_manifest.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 +1,98 @@
+# 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.
+
+## 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, cleanup gates, and user-facing reporting. |
+| Configuration | `internal/config` | Strict YAML loading, discovery, 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, invocation audit state, 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` | Transcript and artifact data contracts plus narrow atomic filesystem 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).
+
+## 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. [`extract`](stage-extract.md)
+8. [`render`](stage-render.md)
+9. [`analyze`](stage-analyze.md)
+10. [`publish`](stage-publish.md)
+11. `notify` (placeholder)
+
+`notify` currently has optional notifier call behavior and no persisted pipeline
+outputs; its default collaborator is a no-op sender. The focused stage
+documents own implementation mechanics. The
+[CLI](../cli.md) and [Operations](../operations.md) own user-visible invocation
+and execution semantics.
+
+## Focused Documentation
+
+- [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 and run 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)
+- [`extract`](stage-extract.md)
+- [`render`](stage-render.md)
+- [`analyze`](stage-analyze.md)
+- [`publish`](stage-publish.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

@@ -1,24 +1,29 @@
 # Stage: analyze
 
 ## Purpose
+
 Execute selected configured Scriptorium artifacts in dependency order and materialize outputs.
 
 ## Inputs
+
 - configured artifacts from `pipeline.scriptorium.artifacts`
-- optional selected artifact filter (`--artifacts`)
+- optional selected artifact keys supplied through the stage environment
 - built-in/configured/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`
+- prepared stable inputs: `narratio.input.players`, `narratio.input.party`,
+  `narratio.input.glossary`
 - configured artifacts: `narratio.artifact.<key>`
 - previous-session cache: `narratio.previous_session.artifact.<key>`
 
 ## Outputs
+
 - one materialized output per executed configured artifact (`output_path`)
 - stage metadata describing selected/generated/reused artifacts
 
 ## Key Behavior
+
 - skips with metadata when Scriptorium config is missing or no executable artifacts remain.
 - builds runtime artifact catalog (built-ins + configured artifacts).
 - marks non-executable configured artifacts as reusable when output files already exist.
@@ -30,13 +35,26 @@ Supported source families:
 - 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.
+- 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.
 
 ## Invariants
+
 - `analyze` performs no remote storage calls for previous-session source resolution.
 - output provenance and metadata are deterministic per execution.
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#scriptorium-artifact-entries) owns artifact
+  fields and source-selection rules.
+- [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_test.go`

docs/internal/stage-extract.md

@@ -0,0 +1,84 @@
+# Internal: Extract Stage
+
+## Responsibility
+
+`extract` runs after `trim` and before `render`. 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 and fingerprints the Notarius invocation contract;
+3. creates a run-local staging directory and invokes the injected
+   `notarius.Runner`;
+4. validates the successful receipt, confined index, configured required lane
+   descriptors, and regular payload files;
+5. atomically promotes the complete bundle to its immutable durable location;
+6. records one non-selectable `notarius_index` output and one selectable
+   `notarius_lane` output per configured lane; and
+7. 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, and invocation fingerprint. Validation completes before
+promotion, so a rejected result cannot expose a partial durable bundle.
+
+Any executed extraction outcome that replaces a different effective outcome
+marks succeeded downstream stages stale. Repeating the same disabled self-skip
+with no outputs is stable and does not repeatedly invalidate downstream stages.
+
+## Resume Validation
+
+`internal/stage/extract_resume.go` permits a skip only when the existing stage
+record succeeded and still matches the current invocation fingerprint. The
+fingerprint covers the resolved executable and config paths, pipeline ID,
+timeout, working directory, and sorted configured output contracts.
+
+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.
+
+The fingerprint cannot observe files imported by Notarius configuration,
+profile contents, prompt/module definitions, or other transitive inputs.
+Operators must force extraction after changing any such input.
+
+## Failure Behavior
+
+Adapter startup, timeout, nonzero exit, receipt decoding, path confinement,
+index compatibility, required-lane rejection, payload inspection, checksum, or
+promotion 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 promotion.
+
+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 invocation manifest and immutable promoted bundle remain
+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`
+- 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

@@ -1,18 +1,22 @@
 # 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.
@@ -20,6 +24,14 @@ Normalize raw transcript inputs and merge into base transcript via Seriatim.
 - 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.
+
+## 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`

docs/internal/stage-normalize.md

@@ -1,16 +1,20 @@
 # 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 from manifest outputs/canonical fallback.
 - applies `pipeline.normalize` config or default normalize config.
 - runs Seriatim normalize with configured timeout/binary.
@@ -18,5 +22,13 @@ Normalize polished transcript into final transcript using Seriatim.
 - 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.
+
+## 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`

docs/internal/stage-polish.md

@@ -1,17 +1,21 @@
 # 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 from merge outputs/canonical fallback.
 - invokes Audita with configured model/module/runtime options.
 - validates processed transcript structure (`segments` array required).
@@ -19,5 +23,14 @@ Run Audita polishing on base transcript and produce polished transcript.
 - materializes canonical outputs; records logs/generated config and adapter metadata.
 
 ## Invariants
+
 - polished transcript schema validation is mandatory.
 - report output is config-gated.
+
+## 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`

docs/internal/stage-prepare.md

@@ -1,17 +1,18 @@
 # Stage: prepare
 
 ## Purpose
+
 Materialize canonical current-session inputs before processing stages.
 
 ## Inputs
-- resolved `campaign.yml`, `session.yml`, and pipeline config
+
+- resolved campaign, session, and pipeline configuration
 - stable input files (`speakers`, `autocorrect`, `glossary`, `players`, `party`)
-- audio source:
-  - local `audio_dir`/`audio_files`, or
-  - S3 `audio_s3.prefix`
+- 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`
@@ -26,6 +27,7 @@ Materialize canonical current-session inputs before processing stages.
 - deterministic `manifest.inputs` entries (checksums + provenance)
 
 ## Key Behavior
+
 - validates required config/store state.
 - enforces local audio vs S3 audio mutual exclusivity.
 - materializes S3 audio through spool/cache-aware logic.
@@ -39,6 +41,20 @@ Materialize canonical current-session inputs before processing stages.
 Required previous-session inputs fail when unavailable; optional missing inputs are skipped.
 
 ## Invariants
+
 - only `prepare` hydrates canonical `previous/` cache state.
-- managed previous artifacts are stored under `previous/artifacts/**` without duplicate `artifacts/artifacts/` nesting.
+- managed previous artifacts are stored under `previous/artifacts/**` without
+  duplicate `artifacts/artifacts/` nesting.
 - `manifest.inputs` ordering is deterministic (`kind`, `path`).
+
+## 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/audio/s3_audio_test.go`,
+  `internal/previouscache/*_test.go`

docs/internal/stage-publish.md

@@ -1,33 +1,45 @@
 # Stage: publish
 
 ## Purpose
+
 Upload run/session outputs to object storage and atomically advance remote current state.
 
 ## Inputs
-- successful prerequisite stages: `prepare`, `transcribe`, `merge`, `polish`, `normalize`, `trim`, `render`, `analyze`
-- run root `runs/{run_id}/**`
-- publish output rules (`pipeline.publish.outputs`)
+
+- successful preceding stages from the [canonical stage set](overview.md#pipeline-stage-set)
+- invocation-scoped run files
+- resolved publish output rules
 - effective publish locks (static + remote merged lock set)
-- local `previous/**` files when present
+- durable previous-session cache files when present
 
 ## Outputs
-- uploaded run files under remote `runs/{run_id}/...` (excluding `audio/**`)
-- uploaded selected publish outputs under session prefix
-- uploaded `previous/**` files under session prefix when present
-- uploaded `current/manifest.json`
-- uploaded `current/run_id.txt` written last
+
+- uploaded invocation record and selected publish outputs;
+- uploaded durable previous-session cache files when present;
+- updated remote current manifest; and
+- remote current-run commit marker, written last.
+
+Exact remote placement and the operator workflow belong in
+[Operations](../operations.md#publish-workflow).
 
 ## Key Behavior
+
 - stage can self-skip when publish disabled or run upload disabled.
 - validates prerequisite stage success and object-store availability.
-- collects deterministic run file list plus run `manifest.json`.
+- collects a deterministic run file list plus run `manifest.json`, excluding
+  `audio/**` and the run-local `extract/notarius-output/**` staging bundle.
+- keeps run-local Notarius receipt and stderr diagnostics eligible for the run
+  archive.
 - resolves publish output sources through runtime artifact catalog and manifest-aware resolution.
+- 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.
 - locked outputs are skipped intentionally (including required ones).
 - optional missing outputs are skipped; required missing unlocked outputs fail.
 - writes remote current manifest before current run pointer.
 
 ## Metadata Signals
+
 Includes counts/lists for:
 - run uploads
 - published output uploads
@@ -39,6 +51,24 @@ Includes counts/lists for:
 - `current_pointer_written`
 
 ## Invariants
+
 - `current/run_id.txt` is the remote commit marker and is written last.
-- run upload excludes `audio/**`.
+- run upload excludes `audio/**` and `extract/notarius-output/**`.
+- `extract/notarius.receipt.json` and `extract/notarius.stderr.log` remain
+  eligible run-record diagnostics.
 - publish locks are not overridden by `--force`.
+
+The commit boundary and cleanup gate are normative architecture invariants; see
+[Architecture](../policy/architecture.md#publish-commit-boundary).
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#publish-configuration-summary) owns output and
+  static-lock fields.
+- [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_test.go`, `internal/app/operator_helpers_test.go`,
+  `internal/app/post_publish_cleanup_test.go`

docs/internal/stage-render.md

@@ -1,17 +1,21 @@
 # Stage: render
 
 ## Purpose
+
 Render Markdown transcript artifacts from normalized JSON transcripts via Seriatim.
 
 ## 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 manifest-first, then canonical fallback.
 - writes run-local outputs first, then materializes canonical session outputs.
@@ -19,11 +23,20 @@ Render Markdown transcript artifacts from normalized JSON transcripts via Seriat
 - skips with stage metadata when `pipeline.render.enabled=false`.
 
 ## 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.
+
+## 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`

docs/internal/stage-transcribe.md

@@ -1,22 +1,36 @@
 # 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 speaker ID from `.flac` basename.
-- runs WhisperX with configured concurrency/retry settings.
+- dispatches WhisperX requests through a bounded worker pool.
 - validates each output as JSON.
 - writes run-local outputs then materializes canonical transcript outputs.
 
 ## Invariants
+
 - speaker basenames must be unique.
 - output path returned by adapter must match requested output path.
 - each successful output is validated before stage success.
+
+## 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`

docs/internal/stage-trim.md

@@ -1,16 +1,21 @@
 # Stage: trim
 
 ## Purpose
-Produce a final-trimmed transcript. By default, the stage generates bounds and applies a bounds-driven trim.
+
+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;
@@ -23,6 +28,16 @@ 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.
+
+## 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`

docs/internal/storage.md

@@ -1,10 +1,16 @@
 # Internal: Storage
 
 ## Purpose
-Document remote object-store contracts and S3 implementation behavior.
+
+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)`
 - `Download(ctx, key, localPath)`
 - `Upload(ctx, localPath, key, opts)`
@@ -14,15 +20,15 @@ Key invariant:
 - callers pass full bucket-relative keys;
 - storage implementations do not infer campaign/session/run prefixes.
 
-## Configuration
-`NewObjectStoreFromConfig` currently supports S3-backed stores from `pipeline.storage.*` config.
+## Composition
 
-S3 constructor behavior:
-- requires configured bucket;
-- uses region/endpoint/path-style options when set;
-- resolves credentials from configured env var names (with defaults).
+`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`.
 - `Download` writes local files with parent directory creation.
@@ -30,5 +36,13 @@ S3 constructor behavior:
 - `Exists` maps not-found responses to `false`.
 
 ## Invariants
+
 - storage layer is stateless regarding manifest/stage progression.
 - publish ordering semantics are owned by stage/app code, not storage adapters.
+
+## 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

@@ -1,57 +1,74 @@
 # Internal: Workspace
 
 ## Purpose
-Define local session layout, run-local stage layout, and cleanup guardrails.
 
-## Canonical Session Layout
-Session root:
-- `{workspace.root}/work/{campaign}/{session_id}`
+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).
 
-Core directories/files:
-- `inputs/`
-- `audio/`
-- `transcripts/`
-- `artifacts/`
-- `reports/`
-- `logs/`
-- `config/`
-- `current/`
-- `runs/`
-- `previous/`
-- `manifest.json`
-- `.lock`
+## Path Ownership
 
-`previous/` reserved files:
-- `previous/manifest.json`
-- `previous/artifacts/**`
+`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` and application cleanup helpers enforce confinement for
+relative destinations and deletion targets.
 
 ## Run-Local Stage Layout
-When run context is available, stages use:
-- `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/`
 
-Run-local outputs are materialized back into canonical session paths before stage success.
-`previous/**` writes are never redirected to run-local output paths.
+`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
+promotes the validated external bundle to the unique immutable Notarius bundle
+path supplied by `internal/artifacts`. `internal/fileops.PromoteDirectory`
+copies only regular files and directories to a same-filesystem temporary
+sibling. Source traversal uses confined directory handles and identity checks
+so replacing an inspected root, directory, or file is rejected rather than
+followed. The completed tree is atomically renamed without replacing an
+existing destination. Exact physical paths belong in
+[Operations](../operations.md#extraction-workflow).
 
 ## Locking
-`artifacts.LocalStore` enforces single-writer session lock via `.lock` file (`ErrLockConflict` on contention).
+
+`artifacts.LocalStore` enforces the single-writer session lock via `.lock`
+(`ErrLockConflict` on contention).
 
 ## Cleanup Semantics
+
 Automatic post-publish cleanup:
+
 - only runs when publish actually executed and succeeded;
 - requires `uploaded=true` and `current_pointer_written=true` metadata;
-- respects `pipeline.spool.delete_audio_after_publish` and `pipeline.workspace.cleanup_after_publish`;
+- consumes the resolved cleanup policy described in
+  [Configuration](../config.md);
 - refuses unsafe deletes (root delete, out-of-root delete, symlink paths).
 
-Manual clean command:
-- `clean <session_id>` removes session work and spool subtree.
-- `clean --all` removes all workspace work and spool children.
-- durable cache is preserved unless `--clear-cache` is requested.
+Manual cleanup uses the same scoped-target checks. 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.
+
+## 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 promotion: `internal/fileops/directory.go`
+- Cleanup confinement: `internal/app/cleanup_targets.go`,
+  `internal/app/post_publish_cleanup.go`
+- Tests: `internal/artifacts/paths_model_test.go`,
+  `internal/artifacts/local_test.go`, `internal/stage/run_local_test.go`,
+  `internal/fileops/directory_test.go`,
+  `internal/app/cleanup_targets_test.go`,
+  `internal/app/post_publish_cleanup_test.go`

docs/operations.md

@@ -75,16 +75,22 @@ Canonical stage order:
 4. `polish`
 5. `normalize`
 6. `trim`
-7. `render`
-8. `analyze`
-9. `publish`
-10. `notify`
+7. `extract`
+8. `render`
+9. `analyze`
+10. `publish`
+11. `notify`
 
 Execution rules:
 
 - succeeded stages are skipped unless `--force` is set;
 - `run` continues interrupted or partially completed sessions by running non-succeeded stages;
-- force rerunning a succeeded upstream stage marks succeeded downstream stages as `stale`.
+- forcing an upstream stage marks succeeded downstream stages as `stale` before
+  the replacement runs; and
+- an executed failure, changed self-skip, or success that replaces a different
+  effective upstream outcome also marks succeeded downstream stages stale. A
+  repeated self-skip with the same reason and no outputs is stable and does not
+  perpetually rerun downstream work.
 
 Single-stage execution:
 
@@ -101,7 +107,64 @@ Selection behavior:
 - validates names against `pipeline.scriptorium.artifacts`;
 - filters analyze execution to selected configured artifacts;
 - filters publish rules for `narratio.artifact.<name>` sources only;
-- does not suppress built-in transcript or bounds publish sources.
+- 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.
+
+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 run manifest and
+immutable bundle remain available for inspection, but downstream resolution
+requires a new current successful extraction record.
+
+Atomic Notarius bundle promotion is supported on Linux, macOS, and Windows.
+On other operating systems, extraction fails before copying the bundle into a
+temporary promotion tree because Narratio has no verified atomic no-replace
+directory primitive there. This is an extraction limitation, not a broader
+platform-support guarantee for every Narratio workflow.
+
+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-record upload excludes the complete
+`extract/notarius-output/**` subtree. The receipt and stderr files remain
+eligible run-record diagnostics. 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
+or durable output validation changes. It cannot fingerprint configuration
+files, profiles, prompts, modules, or references loaded transitively by
+Notarius. 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 Workflow
 
@@ -119,8 +182,10 @@ narratio run-stage publish 2026-04-04 --force
 
 Publish commit model:
 
-- uploads run files under `{session_prefix}/runs/{run_id}/`;
-- uploads configured published outputs;
+- uploads eligible run files under `{session_prefix}/runs/{run_id}/`, excluding
+  audio and the run-local Notarius staging bundle;
+- uploads configured published outputs, including only explicitly configured
+  extraction lanes;
 - uploads `previous/**` cache files when present;
 - writes `current/manifest.json`;
 - writes `current/run_id.txt` last.
@@ -198,6 +263,10 @@ Durable session paths:
 - `config/**`
 - `runs/**`
 
+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`

docs/policy/architecture.md

@@ -1,202 +1,241 @@
-# Narratio Architecture
+# Architecture
 
-## Purpose
+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.
 
-`narratio` is a Go orchestration application for processing D&D session audio into polished transcripts and generated session artifacts.
+## System Shape
 
-This document defines the development principles for the project. It is inward-facing: its audience is developers and LLM coding agents. It should guide future changes, not serve as a complete implementation reference.
+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.
 
-Implemented component details belong under `docs/internal/`.
+Narratio coordinates specialized external systems rather than reimplementing
+their domains:
 
-## Project Shape
+- 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 is a modular, stage-driven orchestrator.
+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/).
 
-It coordinates specialized downstream systems rather than reimplementing their domains:
+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.
+The implemented stage inventory belongs in the
+[Internal Overview](../internal/overview.md).
 
-- WhisperX handles transcription.
-- Seriatim handles deterministic transcript merge/normalization/trim behavior.
-- Audita handles transcript correction and polishing.
-- Scriptorium handles prompt execution and generated artifacts.
+Narratio is contract-first without being abstraction-heavy. Interfaces and
+extension points should protect demonstrated boundaries. New abstraction is not
+itself an architectural goal.
 
-Narratio owns orchestration, configuration loading, session/run state, local and remote path modeling, manifest persistence, stage sequencing, resume behavior, and publish semantics.
+## Ownership And Dependency Direction
 
-Narratio should remain explicit and comprehensible. It is not intended to become a generic workflow engine.
+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.
 
-## Core Principles
+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.
 
-### Modular and composable
+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.
 
-Code should be organized around clear responsibilities. Stages, adapters, config loading, manifest persistence, path construction, and storage behavior should remain separable and independently testable.
+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.
 
-### Hexagonal boundaries
+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.
 
-External systems should be isolated behind narrow adapters. Stage logic should depend on Narratio-level interfaces and data structures, not on external SDK types, subprocess argument construction, or transport-specific details.
+## Stage Boundaries
 
-### Standard library preference
+Each stage has one explicit responsibility and declares:
 
-Prefer the Go standard library. Add dependencies only when they provide substantial value, are necessary for an external integration, or are a widely used de facto standard.
-
-Accepted examples include a YAML library for configuration and the AWS SDK for S3-compatible storage.
-
-### Explicit orchestration
-
-The pipeline should remain stage-driven and explicit. New behavior should be added through clear stage, adapter, config, or manifest contracts rather than implicit side effects or generic workflow abstraction.
-
-## Stage Design
-
-Each stage should have a clear scope of responsibility.
-
-A stage should define:
-
-- its purpose;
 - required input state;
 - produced output state;
-- config fields it consumes;
+- configuration it consumes;
 - external adapters it uses;
-- manifest refs it reads or writes;
-- skip, force, and resume behavior;
-- failure behavior;
-- tests that protect its contract.
+- manifest references and metadata it reads or writes;
+- skip, force, invalidation, and resume behavior; and
+- failure behavior.
 
-Stages should avoid reaching across boundaries. If shared behavior is needed, prefer a helper or service with a narrow interface over duplicating ad hoc logic between stages.
+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.
 
-## Transactionality and Resume
+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 an upstream stage invalidates succeeded downstream work
+according to the canonical stage order.
 
-A stage should behave transactionally.
+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.
 
-A stage is complete only when its outputs have been written, validated, and recorded in the manifest. If a stage fails, Narratio should preserve enough local state for inspection, recovery, and resume.
+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.
 
-A failed or incomplete run must not be treated as successful. Later stages should depend on manifest-recorded success, not merely on incidental files existing on disk.
+## Manifest, Resume, And Restore
 
-## Manifest Model
+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.
 
-The manifest is the durable local ledger for a run.
+Resume and skip decisions are manifest-driven. Filesystem state may be
+inspected and validated, but file presence alone does not replace recorded
+stage state. Invocation-scoped run records provide an audit of one execution;
+they do not replace the session manifest as progress authority.
 
-It should record:
+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).
 
-- run identity;
-- stage status;
-- input and output refs;
-- logs and generated config refs;
-- checksums or provenance where useful;
-- non-secret adapter and publish metadata.
+## Configuration
 
-Resume behavior should be manifest-driven. Filesystem state may be inspected and validated, but it should not replace manifest stage state as the source of run progress.
+Configuration is strict, explicit, centralized, and operator-oriented.
 
-## Adapter Boundaries
+- YAML decoding rejects unknown fields.
+- Defaults are centralized and testable.
+- Empty configured values do not silently replace meaningful defaults.
+- Validation rejects invalid composition before stage execution where
+  practical.
+- 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.
 
-Adapters own external integration details.
+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).
 
-Expected boundaries:
+## Artifacts, Paths, And Storage
 
-- WhisperX HTTP details stay in the WhisperX adapter.
-- Seriatim CLI construction stays in the Seriatim adapter.
-- Audita CLI construction stays in the Audita adapter.
-- Scriptorium CLI construction stays in the Scriptorium adapter.
-- Object-storage details stay behind the storage adapter interface.
-- AWS SDK types stay inside the S3 storage implementation.
+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.
 
-Stage code should express intent in Narratio terms and call adapters through narrow contracts.
+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.
 
-## Configuration Philosophy
+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 configured post-publish
+gates, and it must preserve durable cache unless cache removal is explicitly
+requested.
 
-Configuration should be strict, explicit, and operator-friendly.
+Physical layout, retention, and operational lifecycle belong in
+[Operations](../operations.md). Logical external formats and durable integration
+contracts belong under [Integrations](../integrations/).
 
-Principles:
+## Publish Commit Boundary
 
-- YAML decoding should reject unknown fields.
-- Defaults should be centralized and testable.
-- Empty configured values should not silently override meaningful defaults.
-- Session templating should remain narrow and deterministic.
-- Template support should serve operator convenience, not become a general configuration language.
+Publish has one explicit remote commit boundary. A remote run becomes current
+only after Narratio has successfully uploaded the run record, required published
+outputs, `current/manifest.json`, and finally `current/run_id.txt`.
 
-Narratio should not become a secondary configuration system for downstream tools. Seriatim, Audita, and Scriptorium should own their runtime defaults wherever practical. Narratio should pass required stage-contract paths and explicit operator overrides.
+`current/run_id.txt` is the commit marker and must be written 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.
 
-## Path and Storage Discipline
+Automatic local cleanup is permitted only after a successful publish commit,
+only when explicitly configured, and only through the path-safety guardrails.
 
-Local and remote paths are part of Narratio’s application contract.
+## Security, Privacy, And Diagnostics
 
-Code should use centralized path helpers for workspace, spool, session, run, artifact, log, config, and publish/current paths. Stages should avoid reconstructing canonical paths through scattered string concatenation.
+Narratio handles private campaign material. Transcripts, prompts, generated
+artifacts, reports, logs, manifests, and diagnostic files are potentially
+sensitive.
 
-Storage backends should receive explicit bucket-relative keys. Storage implementations should not infer campaign, session, run, or root-prefix semantics.
+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. Secrets enter through configured environment
+variable names or secret-file references. Diagnostics should avoid transcript
+and prompt content unless a deliberate, bounded inspection mechanism requires
+it.
 
-## Publish Invariants
+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.
 
-Publish behavior must preserve a clear commit boundary.
+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).
 
-A remote run is current only after the publish stage has successfully uploaded the run record, required published outputs, `current/manifest.json`, and finally `current/run_id.txt`.
+## Determinism And Testability
 
-`current/run_id.txt` is the final remote commit marker and must be written last.
+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.
 
-Failed, incomplete, skipped, or uncommitted publish attempts must not be presented as current remote state. Local cleanup is permitted only after successful publish commit and only when explicitly configured.
+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).
 
-## Security and Privacy
+## Documentation And Decision Records
 
-Narratio handles private campaign material.
+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/`.
 
-Rules:
+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.
 
-- Do not store raw secrets in pipeline or session YAML.
-- Use environment variable names or secret-file references for secret handling.
-- Do not write raw secret values to manifests, logs, generated configs, or publish metadata.
-- Treat transcripts, generated artifacts, prompts, reports, and logs as potentially sensitive.
-- Avoid logging transcript or prompt content unless there is a deliberate diagnostic reason.
+## Architectural Non-Goals
 
-## Diagnostics
-
-Diagnostics should be durable and discoverable, but distinct from canonical outputs.
-
-Logs, reports, generated invocation/config files, and render-debug files support debugging. Transcript tiers and configured artifacts are pipeline products.
-
-Manifest refs should preserve that distinction.
-
-## Determinism
-
-Where practical, Narratio should prefer deterministic behavior:
-
-- stable local path layout;
-- stable remote key layout;
-- sorted upload order;
-- predictable generated config files;
-- repeatable command construction;
-- tests that do not depend on live external services.
-
-Run IDs and timestamps may be intentionally variable, but surrounding behavior should remain testable.
-
-## Testing Expectations
-
-Core behavior should be testable without live external services.
-
-Tests should cover:
-
-- config loading, defaults, and validation;
-- CLI parsing and command construction;
-- path helpers;
-- manifest transitions;
-- stage success, failure, skip, and resume behavior;
-- adapter command construction;
-- fake storage behavior;
-- publish commit ordering;
-- example config validity where practical.
-
-Live S3, WhisperX, LLM, or subprocess integration tests should be explicit integration tests, not required for ordinary unit test runs.
-
-## Documentation Expectations
-
-Documentation must follow `docs/documentation/policy.md`.
-
-Current behavior belongs in user-facing docs and `docs/internal/`. Future, planned, aspirational, experimental, or unimplemented work belongs only under `docs/roadmap/`.
-
-`docs/architecture.md` should remain concise and principle-focused. It should not duplicate the full config reference, CLI reference, operations guide, or internal stage documentation.
-
-## Non-Goals
-
-Narratio is not:
+Narratio does not aim to provide:
 
 - a generic DAG or workflow engine;
-- a replacement configuration layer for Seriatim, Audita, or Scriptorium;
-- a storage backend abstraction beyond the needs of this pipeline;
-- a place to embed raw secrets;
-- a place for stage logic to depend directly on AWS SDK types or downstream tool internals;
+- 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/development.md

@@ -1,94 +0,0 @@
-# Development Guide
-
-## Purpose
-Canonical contributor workflow and engineering conventions for implemented Narratio behavior.
-
-## Repository layout
-
-- `cmd/narratio/`: CLI entrypoint.
-- `internal/app/`: command handlers, run/stage orchestration, cleanup gates, secrets loading.
-- `internal/config/`: strict YAML loading, defaults, and validation.
-- `internal/stage/`: stage implementations and stage registry/order.
-- `internal/adapters/`: external boundary adapters (WhisperX, Seriatim, Audita, Scriptorium, storage, notify).
-- `internal/manifest/`: session/run manifest types and persistence.
-- `internal/artifacts/`: canonical local/remote path helpers and local artifact store.
-- `docs/`: canonical documentation set.
-- `examples/`: maintained config examples used by tests.
-
-## Build and test commands
-
-- Run focused CLI behavior checks:
-
-```bash
-go test ./internal/app -run TestExecute -v
-```
-
-- Run config example load/validate checks:
-
-```bash
-go test ./internal/config -run TestExamplesLoadAndValidate -v
-```
-
-- Run full test suite:
-
-```bash
-go test ./...
-```
-
-## Coding conventions
-
-- Keep orchestration explicit and stage-driven; do not introduce generic workflow/DAG abstractions.
-- Keep external-system details inside adapter packages; stages should consume Narratio-level contracts only.
-- Use centralized path helpers from `internal/artifacts` rather than ad hoc path concatenation.
-- Preserve manifest-driven state transitions (`running`, `succeeded`, `failed`, `skipped`, `stale`) as the source of run progress.
-- Keep user/operator docs implementation-accurate; planned work belongs only under `docs/roadmap/`.
-
-For design principles and invariants, see [docs/architecture.md](./architecture.md). For stage/adapter contracts, see [docs/internal/README.md](./internal/README.md).
-
-## Dependency policy
-
-- Prefer Go standard library where practical.
-- Add third-party dependencies only when they provide clear value for required behavior.
-- Keep dependency additions narrow to the boundary package that needs them.
-
-## Change playbooks
-
-### Add config fields
-
-1. Add fields to config structs in `internal/config`.
-2. Set defaults in `internal/config/defaults.go` when appropriate.
-3. Add validation rules in `internal/config/validate.go`.
-4. Add or update load/validate tests in `internal/config/*_test.go`.
-5. Update canonical config docs and examples:
-   - [docs/config.md](./config.md)
-   - relevant files under `examples/`
-
-### Add CLI flags or commands
-
-1. Update command parsing and behavior in `internal/app`.
-2. Add or update command tests (`TestExecute` and command-specific tests).
-3. Update [docs/cli.md](./cli.md) and, if operator workflow changes, [docs/operations.md](./operations.md).
-
-Remote-storage commands must obtain object storage through the app-level command object-store helper. Do not call `storage.NewObjectStoreFromConfig` directly from command handlers; the helper loads configured filesystem secrets before constructing the storage adapter.
-
-### Add or modify stages/adapters
-
-1. Implement stage behavior in `internal/stage` with clear input/output boundaries.
-2. Keep external transport/subprocess details in `internal/adapters`.
-3. Preserve manifest and publish-output semantics expected by runner and publish logic.
-4. Add/update stage and adapter tests.
-5. Update internal component contracts in `docs/internal/`.
-
-### Update examples
-
-1. Keep canonical examples only in `examples/`.
-2. Ensure examples load and validate through runtime config paths.
-3. Update `internal/config/load_validate_test.go` as needed.
-4. Update links in `docs/config.md` if example filenames change.
-
-### Update docs and roadmap
-
-1. Keep implemented behavior in canonical docs (`README`, `docs/*.md`, `docs/internal/`).
-2. Keep planned/unimplemented behavior only in `docs/roadmap/`.
-3. After completing roadmap items, remove or mark them complete in `docs/roadmap/documentation.md`.
-4. Run a link/path sweep before finalizing changes.

docs/policy/documentation.md

@@ -1,356 +1,148 @@
-# Go Project Documentation Policy
+# Documentation Policy
 
 ## Purpose
 
-Project documentation must help four audiences:
-
-1. users who need to run the application;
-2. administrators/operators who need to configure and operate it;
-3. developers who need to understand and change it safely;
-4. LLM coding agents that need clear scope, boundaries, and invariants.
-
-Docs should be accurate, concise, task-oriented, and organized by audience. Prefer links to canonical docs over repetition.
+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
 
-### 1. Keep docs concise
+### 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.
 
-Each document should cover a defined scope and only the essentials for that scope.
+### 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
 
-Avoid:
-- long background explanations;
-- repeated reference material;
-- implementation detail in user-facing docs;
-- aspirational language outside roadmap docs;
-- verbose examples where one minimal example is clearer.
-
-### 2. Document only implemented behavior outside roadmap files
-
-Unimplemented, planned, aspirational, experimental, or future work may be described only under:
-
-- `docs/roadmap/`
-
-No other documentation file, including `README.md`, should describe code, features, modules, stages, commands, config fields, or behaviors that do not currently exist.
-
-If a feature is partial, non-roadmap docs may describe only the implemented portion and its current boundary.
-
-### 3. Use canonical homes
-
-Each type of information should have one canonical location.
-
-Canonical homes:
-
-- project purpose and quickstart: `README.md`
-- development principles: `docs/architecture.md`
-- configuration reference: `docs/config.md`
-- CLI reference: `docs/cli.md`
-- operations and recovery: `docs/operations.md`
-- troubleshooting: `docs/troubleshooting.md`
-- implemented internals: `docs/internal/`
-- future work: `docs/roadmap/`
-- contributor workflow: `docs/development.md`
-- copyable examples: `examples/`
-
-Other files should summarize briefly and link to the canonical source.
-
-### 4. Keep examples real
-
-Examples should be valid, maintained, and free of secrets.
-
-Where practical:
-- example configs should load successfully;
-- example commands should match real CLI syntax;
-- important examples should be covered by tests.
-
-## Documentation Profiles
-
-All projects require:
-
-- `README.md`
-- `docs/architecture.md`
-
-Additional docs depend on the project.
-
-### Small library
-
-Recommended:
-- `docs/development.md`, if contributor conventions are non-obvious
-
-### Simple CLI
-
-Required:
-- `docs/cli.md`
-
-Recommended:
-- `docs/development.md`
-
-### Config-driven CLI
-
-Required:
-- `docs/cli.md`
-- `docs/config.md`
-
-Recommended:
-- `examples/`
-- `docs/development.md`
-
-### Stateful or operator-facing application
-
-Required:
-- `docs/cli.md`, if CLI-based
-- `docs/config.md`, if config-driven
-- `docs/operations.md`
-
-Recommended:
-- `docs/troubleshooting.md`
-- `examples/`
-- `docs/development.md`
-
-### Modular, staged, service-oriented, or orchestration application
-
-Required:
-- `docs/cli.md`, if CLI-based
-- `docs/config.md`, if config-driven
-- `docs/operations.md`
-- `docs/internal/`
-- `docs/development.md`
-
-Recommended:
-- `docs/troubleshooting.md`
-- validated examples under `examples/`
-
-## Required Documents
-
-### README.md
-
-**Audience:** users, administrators, operators
-
-The README is the outward-facing project orientation page.
-
-It should include, in order:
-
-1. concise description;
-2. elevator pitch;
-3. shortest useful command or usage example;
-4. links to targeted docs.
-
-The README should be short. It is not a manual.
-
-The “shortest useful command” means the simplest command that performs the project’s core use case. (It does not mean `app --help`.)
-
-### docs/architecture.md
-
-**Audience:** developers, LLM coding agents
-
-`docs/architecture.md` is required for every project.
-
-It is an inward-facing development policy document. It should describe how the project is intended to be built and changed.
-
-It should include:
-
-- project shape;
-- core design principles;
-- package and boundary philosophy;
-- state/persistence philosophy, if applicable;
-- external integration philosophy, if applicable;
-- error-handling and logging principles;
-- testing expectations;
-- documentation expectations;
-- architectural invariants;
-- explicit non-goals, if useful.
-
-For small projects, this file may be brief. It may simply state that the project is intentionally narrow, monolithic, and dependency-light.
-
-### docs/config.md
-
-**Audience:** administrators, operators, advanced users
-
-Required for applications with configuration files.
-
-It should include, in order:
-
-1. config file locations and discovery precedence;
-2. minimal working config;
-3. production-oriented config;
-4. full configuration reference;
-5. secrets handling, if applicable;
-6. links to maintained examples.
-
-The full configuration reference should be canonical.
-
-### docs/cli.md
-
-**Audience:** users, administrators, operators
-
-Required for CLI applications.
-
-It should include, in order:
-
-1. shortest useful command;
-2. command overview;
-3. complete flag reference;
-4. common workflows;
-5. diagnostic or recovery commands, if applicable.
-
-Explain when commands are useful, not just their syntax.
-
-### docs/operations.md
-
-**Audience:** administrators, operators
-
-Required for applications that maintain state, support resume behavior, run multiple stages, write durable artifacts, use remote storage, or require recovery procedures.
-
-It should cover:
-
-- normal workflow;
-- filesystem layout;
-- remote storage layout, if applicable;
-- logs and manifests;
-- resume/retry behavior;
-- cleanup behavior;
-- archive/backup behavior;
-- safe recovery procedures;
-- operational caveats.
-
-### docs/troubleshooting.md
-
-**Audience:** administrators, operators
-
-Recommended once recurring failure modes exist.
-
-Each entry should include:
-
-- symptom;
-- likely cause;
-- diagnostic command or inspection step;
-- safe fix;
-- relevant links.
-
-### docs/development.md
-
-**Audience:** developers, LLM coding agents
-
-Required for projects maintained by humans and LLM coding agents.
-
-It should include:
-
-- repository layout;
-- build/test commands;
-- coding conventions;
-- dependency policy;
-- how to add config fields;
-- how to add CLI flags;
-- how to add stages/modules/adapters, if applicable;
-- how to update examples;
-- documentation update expectations.
-
-### docs/internal/
-
-**Audience:** developers, LLM coding agents
-
-Required for modular, staged, service-oriented, or orchestration projects.
-
-This directory describes implemented internal components. It is not the roadmap.
-
-Use one file per major component where useful.
-
-Each component doc should include:
-
-1. purpose;
-2. inputs and outputs;
-3. boundaries;
-4. config fields used;
-5. external adapters used;
-6. state or manifest behavior, if applicable;
-7. skip/resume behavior, if applicable;
-8. failure behavior;
-9. tests to inspect before changing;
-10. architectural invariants.
-
-### docs/roadmap/
-
-**Audience:** maintainers, developers, LLM coding agents
-
-This is the only place for planned, future, aspirational, experimental, or unimplemented work.
-
-Roadmap docs should clearly distinguish:
-
-- proposed work;
-- accepted plans;
-- deferred ideas;
-- rejected ideas;
-- implementation prompts or task breakdowns, if useful.
-
-Roadmap docs should not be confused with current behavior.
-
-### docs/integrations/
-
-**Audience:** developers, LLM coding agents
-
-Required for projects that depend on external CLIs, APIs, services, protocols, or file formats where the integration contract is important to maintain.
-
-This directory contains concise, versioned reference notes for external integration contracts. It should document only the parts of the external system that this project actually uses.
-
-Use one file per integration where useful.
-
-## Examples Directory
-
-Projects with non-trivial configuration or workflows should include `examples/`.
-
-Useful examples include:
-
-- minimal working config;
-- production-oriented config;
-- full annotated config;
-- local development config;
-- remote/object-storage config;
-- minimal session/input file.
-
-Examples should be valid, maintained, tested when practical, and linked from relevant docs.
-
-## Security and Privacy
-
-Docs and examples must not include:
-
-- real API keys;
-- tokens;
-- passwords;
-- private keys;
-- private environment dumps;
-- sensitive user data;
-- raw private transcripts;
-- private infrastructure details unless intentionally public.
-
-Document secret-handling mechanisms, not actual secret values.
-
-## Maintenance Rules
-
-When docs change, verify the affected behavior.
-
-Where practical:
-
-- load example config files in tests;
-- test CLI examples or command parser behavior;
-- validate documented flags against real flags;
-- remove stale references;
-- update links after renames;
-- keep roadmap content out of non-roadmap docs.
-
-If documentation and code disagree, fix the documentation and/or open a roadmap item; do not leave aspirational behavior in current-behavior docs.
-
-Documentation is complete only when it matches the current code.
-
-## Documentation Change Checklist
-
-Before merging documentation changes, verify:
-
-- README is concise and orientation-focused.
-- `docs/architecture.md` describes development principles.
-- Future work appears only under `docs/roadmap/`.
-- User-facing docs avoid unnecessary internals.
-- Developer-facing docs preserve boundaries and invariants.
-- Config examples match the schema.
-- CLI examples match real commands and flags.
-- Defaults appear in the canonical config reference.
-- No secrets or private data are included.
-- Links are accurate.
+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. |
+| 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/troubleshooting.md

@@ -24,6 +24,8 @@ 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:
@@ -44,6 +46,8 @@ 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:
@@ -62,7 +66,10 @@ narratio session plan 2026-04-04 --config /path/pipeline.yml --campaign-file /pa
 
 Safe fix:
 
-- align config with [docs/config.md](./config.md) and maintained files under `examples/`.
+- align config with [Configuration](./config.md) and the
+  [maintained examples](../examples/README.md).
+
+Relevant reference: [Configuration](./config.md).
 
 ## Audio mode conflict
 
@@ -74,10 +81,18 @@ 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:
@@ -90,10 +105,159 @@ Likely causes:
 - 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).
+
+## 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).
+
+## Atomic Notarius promotion unsupported
+
+Symptom:
+
+- extraction fails with `atomic no-replace directory promotion is unsupported`
+  before a durable bundle or temporary promotion tree is created.
+
+Likely cause:
+
+- Narratio is running on an operating system other than Linux, macOS, or
+  Windows, where the required atomic no-replace directory primitive has not
+  been implemented and verified.
+
+Safe fix:
+
+- run extraction on Linux, macOS, or Windows. Do not replace the atomic commit
+  with a manual copy or move; the session manifest must never observe a partial
+  or overwritten bundle.
+
+This is an extraction-specific platform boundary, not a support statement for
+unrelated Narratio workflows. See
+[Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## 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`, and `warnings.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;
+- 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 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, not the contents of transitive
+Notarius inputs. Always force extraction after changing them; downstream
+successful stages are then marked stale normally.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
 ## Previous-session artifact input missing
 
 Symptom:
@@ -124,6 +288,8 @@ or rerun prepare after correcting session config:
 narratio run-stage prepare 2026-04-04 --force
 ```
 
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
 ## Session lock conflict (`.lock`)
 
 Symptom:
@@ -147,6 +313,8 @@ Safe fix:
 - wait for active process completion;
 - remove stale lock only after confirming no live process owns it.
 
+Relevant reference: [Operations: Local State Layout](./operations.md#local-state-layout).
+
 ## Restore conflict without `--force`
 
 Symptom:
@@ -168,6 +336,8 @@ Safe fix:
 - review conflicts;
 - rerun with `--force` only when remote state should overwrite local.
 
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
 ## Restore current-state discovery failure
 
 Symptom:
@@ -191,6 +361,8 @@ 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:
@@ -217,6 +389,8 @@ Safe fix:
 - 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:
@@ -232,7 +406,6 @@ Diagnostics:
 
 ```bash
 narratio session status 2026-04-04
-narratio run-stage render 2026-04-04 --force
 ```
 
 Safe fix:
@@ -244,6 +417,8 @@ 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:
@@ -269,6 +444,8 @@ Safe fix:
 - provide required env vars;
 - keep secret values out of YAML.
 
+Relevant reference: [Secrets](./config.md#secrets-handling).
+
 ## S3 audio prepare failure
 
 Symptom:
@@ -284,7 +461,7 @@ Likely causes:
 Diagnostics:
 
 ```bash
-narratio run-stage prepare 2026-04-04 --force
+narratio session validate 2026-04-04
 ```
 
 Safe fix:
@@ -292,6 +469,8 @@ Safe fix:
 - verify prefix contents and storage access;
 - keep session audio mode consistent.
 
+Relevant reference: [Operations](./operations.md).
+
 ## References
 
 - [docs/cli.md](./cli.md)

examples/README.md

@@ -0,0 +1,48 @@
+# 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.
+- [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),
+  [players](campaigns/sample-campaign/players.yml), and
+  [party](campaigns/sample-campaign/party.yml) fixtures.
+- [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. They
+use fictional campaign content and reserved example domains.

examples/campaigns/sample-campaign/speakers.yml

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

examples/pipeline.extraction-subset.yml

@@ -0,0 +1,55 @@
+# 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
+  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

@@ -60,6 +60,11 @@ publish:
     - 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.
@@ -123,6 +128,78 @@ trim:
   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
+  # 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

examples/speakers.yml

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

go.mod

@@ -7,6 +7,7 @@ require (
 	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
+	golang.org/x/sys v0.47.0
 	gopkg.in/yaml.v3 v3.0.1
 )
 

go.sum

@@ -34,6 +34,8 @@ github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 h1:F/M5Y9I3nwr2IEpshZgh1GeHpOIt
 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=
+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/notarius/fake.go

@@ -0,0 +1,22 @@
+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
+	}
+	f.Requests = append(f.Requests, req)
+	if f.Err != nil {
+		return RunResult{}, f.Err
+	}
+	return f.Result, nil
+}

internal/adapters/notarius/runner.go

@@ -0,0 +1,108 @@
+// Package notarius declares the adapter contract for Notarius CLI invocations.
+package notarius
+
+import (
+	"context"
+	"time"
+)
+
+const ReceiptSchemaVersion = "notarius.run-result.v1"
+
+// Runner is the adapter boundary for a complete Notarius pipeline invocation.
+type Runner interface {
+	Run(ctx context.Context, req RunRequest) (RunResult, error)
+}
+
+// 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
+}
+
+// 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
+	WarningCount          int
+	ValidationStatus      string
+	DebugDirectory        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
+	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 {
+	Scope      string
+	ReasonCode string
+}
+
+// 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
+}

internal/adapters/notarius/subprocess.go

@@ -0,0 +1,524 @@
+package notarius
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"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"
+)
+
+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")
+	}
+	if err := validateRunRequest(req); err != nil {
+		return RunResult{}, err
+	}
+
+	args := []string{
+		"run", req.PipelineID,
+		"--config", req.ConfigPath,
+		"--input", req.InputPath,
+		"--output-dir", req.OutputRoot,
+		"--json",
+	}
+	processResult, err := r.run(ctx, subprocess.RunRequest{
+		Executable:    req.Binary,
+		Args:          args,
+		WorkingDir:    req.WorkingDirectory,
+		Timeout:       req.Timeout,
+		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
+	}
+
+	baseResult.Receipt = receipt
+	baseResult.Index = index
+	baseResult.BundleRoot = bundleRoot
+	baseResult.Rejections = rejections
+	baseResult.Warnings = warnings
+	return baseResult, nil
+}
+
+func validateRunRequest(req RunRequest) error {
+	if strings.TrimSpace(req.Binary) == "" {
+		return fmt.Errorf("notarius binary is required")
+	}
+	if strings.TrimSpace(req.PipelineID) == "" {
+		return fmt.Errorf("notarius pipeline id is required")
+	}
+	if req.Timeout <= 0 {
+		return 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 fmt.Errorf("notarius %s path is required", label)
+		}
+		if !filepath.IsAbs(path) {
+			return fmt.Errorf("notarius %s path must be absolute", label)
+		}
+	}
+	if filepath.Clean(req.ReceiptPath) == filepath.Clean(req.LogPath) {
+		return fmt.Errorf("notarius receipt and log paths must be different")
+	}
+	if err := requireRegularFile(req.ConfigPath); err != nil {
+		return fmt.Errorf("validate notarius config path: %w", err)
+	}
+	if err := requireRegularFile(req.InputPath); err != nil {
+		return fmt.Errorf("validate notarius input path: %w", err)
+	}
+	if err := requireDirectory(req.OutputRoot); err != nil {
+		return fmt.Errorf("validate notarius output root: %w", err)
+	}
+	if err := requireDirectory(req.WorkingDirectory); err != nil {
+		return fmt.Errorf("validate notarius working directory: %w", err)
+	}
+	if err := validateLogDestination(req.ReceiptPath); err != nil {
+		return fmt.Errorf("validate notarius receipt path: %w", err)
+	}
+	if err := validateLogDestination(req.LogPath); err != nil {
+		return fmt.Errorf("validate notarius log path: %w", err)
+	}
+	return 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"`
+	WarningCount          *int   `json:"warning_count"`
+	ValidationStatus      string `json:"validation_status"`
+	DebugDirectory        string `json:"debug_directory"`
+}
+
+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.WarningCount == 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.WarningCount < 0 {
+		return Receipt{}, fmt.Errorf("notarius receipt counts must be non-negative")
+	}
+	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,
+		WarningCount:          *document.WarningCount,
+		ValidationStatus:      document.ValidationStatus,
+		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"`
+	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},
+	} {
+		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,
+	}
+	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)
+	}
+
+	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 warningDocument struct {
+	Warnings *[]struct {
+		Scope      string `json:"scope"`
+		ReasonCode string `json:"reason_code"`
+		Message    string `json:"message"`
+	} `json:"warnings"`
+}
+
+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.Warnings == nil {
+		return nil, fmt.Errorf("notarius warning document is missing warnings array")
+	}
+	summaries := make([]WarningSummary, 0, len(*document.Warnings))
+	for _, item := range *document.Warnings {
+		if strings.TrimSpace(item.ReasonCode) == "" || strings.TrimSpace(item.Message) == "" {
+			return nil, fmt.Errorf("notarius warning entries require reason_code and message")
+		}
+		summaries = append(summaries, WarningSummary{Scope: item.Scope, ReasonCode: item.ReasonCode})
+	}
+	return summaries, nil
+}
+
+func decodeBoundedJSON(path string, limit int64, destination any) error {
+	inspected, err := os.Lstat(path)
+	if err != nil {
+		return err
+	}
+	if inspected.Mode()&os.ModeSymlink != 0 || !inspected.Mode().IsRegular() {
+		return fmt.Errorf("path %q must be a regular file without symlinks", path)
+	}
+	file, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = file.Close() }()
+	opened, err := file.Stat()
+	if err != nil {
+		return err
+	}
+	if !opened.Mode().IsRegular() || !os.SameFile(inspected, opened) {
+		return fmt.Errorf("file %q changed before it could be read", path)
+	}
+	reader := io.LimitReader(file, limit+1)
+	data, err := io.ReadAll(reader)
+	if err != nil {
+		return err
+	}
+	if int64(len(data)) > limit {
+		return fmt.Errorf("file %q exceeds %d-byte limit", path, limit)
+	}
+	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 +1,572 @@
+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.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].Scope != "lane:npc-registry" || result.Warnings[0].ReasonCode != "normalized_name" {
+		t.Fatalf("warnings = %#v", result.Warnings)
+	}
+}
+
+func TestSubprocessRunnerInheritsEnvironmentAndSeparatesStreams(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"), "inherited-value")
+	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_count": 0,
+		"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.v2" }},
+		{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_count"] = -1 }},
+		{
+			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")
+	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, map[string]any{"warnings": []any{map[string]any{
+		"scope": "lane:spells", "reason_code": "bounded", "message": "do not retain this", "future": true,
+	}}, "future": true})
+	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].Scope != "lane:spells" {
+		t.Fatalf("loadWarnings() = %#v, %v", warnings, err)
+	}
+
+	for name, path := range map[string]string{"rejections": rejectedPath, "warnings": warningsPath} {
+		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 {
+				_, err = loadWarnings(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)
+	}
+}
+
+func TestFakeRunnerCapturesRequestsAndHonorsContextAndError(t *testing.T) {
+	req := RunRequest{PipelineID: "pipeline"}
+	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)
+	}
+
+	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)}
+	warning := map[string]any{"scope": "lane:npc-registry", "reason_code": "normalized_name", "message": strings.Repeat("external warning", 20)}
+	if includeUnknown {
+		rejection["future"] = true
+		warning["future"] = true
+	}
+	writeJSONFile(t, filepath.Join(bundle, "rejected.json"), map[string]any{"rejected": []any{rejection}, "future": true})
+	writeJSONFile(t, filepath.Join(bundle, "warnings.json"), map[string]any{"warnings": []any{warning}, "future": true})
+	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_count": 1, "validation_status": "rejected",
+	}
+	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", "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",
+	}
+}
+
+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/app/analyze_artifacts_commands_test.go

@@ -21,7 +21,7 @@ func TestExecuteRunStageArtifactsUnsupportedStageFails(t *testing.T) {
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
 	code := Execute(
-		[]string{"run-stage", "polish", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath, "--artifacts", "session_recap"},
+		[]string{"run-stage", "extract", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath, "--artifacts", "session_recap"},
 		&stdout,
 		&stderr,
 	)
@@ -130,6 +130,7 @@ func TestRunArtifactsWithSucceededAnalyzeSkipsUnlessForced(t *testing.T) {
 	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
 		seed.MarkStageSucceeded(stageName, time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), nil)
 	}
+	seed.MarkStageSkipped("extract", time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), "notarius_disabled")
 	if err := store.Save(context.Background(), manifestPath, seed); err != nil {
 		t.Fatalf("save manifest: %v", err)
 	}
@@ -143,7 +144,7 @@ func TestRunArtifactsWithSucceededAnalyzeSkipsUnlessForced(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "executed=0 skipped=10") {
+	if !strings.Contains(out.String(), "executed=1 skipped=11") {
 		t.Fatalf("output = %q, want all stages skipped", out.String())
 	}
 }

internal/app/commands.go

@@ -51,8 +51,3 @@ func Execute(args []string, stdout, stderr io.Writer) int {
 func printUsage(w io.Writer) {
 	fmt.Fprintf(w, "Usage: narratio <%s>\n", strings.Join(supportedCommands, "|"))
 }
-
-func placeholder(out io.Writer, command string) error {
-	_, err := fmt.Fprintf(out, "narratio %s: not yet implemented\n", command)
-	return err
-}

internal/app/commands_test.go

@@ -31,8 +31,8 @@ func TestExecuteValidCommands(t *testing.T) {
 		args    []string
 		wantOut string
 	}{
-		{name: "run", args: []string{"run", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "narratio run: session 2026-05-03; executed=10 skipped=0; manifest="},
-		{name: "session plan", args: []string{"session", "plan", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "prepare: skip\ntranscribe: skip\nmerge: skip\npolish: skip\nnormalize: skip\ntrim: skip\nrender: skip\nanalyze: skip\npublish: skip\nnotify: skip"},
+		{name: "run", args: []string{"run", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "narratio run: session 2026-05-03; executed=11 skipped=1; manifest="},
+		{name: "session plan", args: []string{"session", "plan", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "prepare: skip\ntranscribe: skip\nmerge: skip\npolish: skip\nnormalize: skip\ntrim: skip\nextract: run\nrender: skip\nanalyze: skip\npublish: skip\nnotify: skip"},
 		{name: "session status", args: []string{"session", "status", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "Session: 2026-05-03"},
 		{name: "run-stage", args: []string{"run-stage", "polish", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, wantOut: "narratio run-stage: stage=polish executed=0 skipped=1 force=false; manifest="},
 	}
@@ -335,7 +335,7 @@ func TestExecuteUsesDefaultPipelineConfigPathWhenConfigFlagOmitted(t *testing.T)
 	if code != 0 {
 		t.Fatalf("exit code = %d, want 0; stderr=%q", code, stderr.String())
 	}
-	if !strings.Contains(stdout.String(), "narratio run: session 2026-05-03; executed=10 skipped=0; manifest=") {
+	if !strings.Contains(stdout.String(), "narratio run: session 2026-05-03; executed=11 skipped=1; manifest=") {
 		t.Fatalf("stdout = %q, want successful run output", stdout.String())
 	}
 }

internal/app/extract_lifecycle_test.go

@@ -0,0 +1,416 @@
+package app
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/notarius"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+	"gitea.maximumdirect.net/eric/narratio/internal/stage"
+)
+
+type materializingNotariusRunner struct {
+	cfg               *config.NotariusConfig
+	requests          []notarius.RunRequest
+	failuresRemaining int
+}
+
+func (r *materializingNotariusRunner) Run(_ context.Context, req notarius.RunRequest) (notarius.RunResult, error) {
+	r.requests = append(r.requests, req)
+	if r.failuresRemaining > 0 {
+		r.failuresRemaining--
+		return notarius.RunResult{}, errors.New("notarius execution failed")
+	}
+	externalRunID := fmt.Sprintf("notarius-run-%d", len(r.requests))
+	bundle := filepath.Join(req.OutputRoot, externalRunID)
+	lanesDir := filepath.Join(bundle, "lanes")
+	if err := os.MkdirAll(lanesDir, 0o755); err != nil {
+		return notarius.RunResult{}, err
+	}
+	for path, content := range map[string]string{
+		filepath.Join(bundle, "index.json"):    `{"manifest_file":"manifest.json"}`,
+		filepath.Join(bundle, "manifest.json"): `{}`,
+		filepath.Join(bundle, "rejected.json"): `{"rejected":[]}`,
+		filepath.Join(bundle, "warnings.json"): `{"warnings":[]}`,
+		filepath.Join(lanesDir, "npcs.json"):   `{"npcs":[]}`,
+	} {
+		if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+			return notarius.RunResult{}, err
+		}
+	}
+	output := r.cfg.Outputs["npc_registry"]
+	return notarius.RunResult{
+		Receipt: notarius.Receipt{
+			SchemaVersion: notarius.ReceiptSchemaVersion, RunID: externalRunID,
+			PipelineID: req.PipelineID, OutputDirectory: bundle, IndexFile: "index.json",
+			NormalizedOutputCount: 1, ValidationStatus: "valid",
+		},
+		BundleRoot: bundle,
+		Index: notarius.Index{
+			Path: filepath.Join(bundle, "index.json"), RejectedPath: filepath.Join(bundle, "rejected.json"),
+			WarningsPath: filepath.Join(bundle, "warnings.json"),
+			Lanes: []notarius.LaneDescriptor{{
+				LaneID: output.LaneID, File: "lanes/npcs.json", Path: filepath.Join(lanesDir, "npcs.json"),
+				MediaType: output.MediaType, SchemaID: output.SchemaID,
+				SchemaVersion: output.SchemaVersion, ModuleKey: output.ModuleKey,
+			}},
+		},
+	}, nil
+}
+
+func TestExtractLifecycleDisabledThenEnabled(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, false)
+	plan, err := BuildSingleStagePlan("extract")
+	if err != nil {
+		t.Fatalf("BuildSingleStagePlan(extract) error = %v", err)
+	}
+
+	first, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("disabled executeStages() error = %v", err)
+	}
+	if len(first.Executed) != 1 || len(first.Skipped) != 1 || first.Skipped[0] != "extract" || len(runner.requests) != 0 {
+		t.Fatalf("disabled summary = %#v requests=%d", first, len(runner.requests))
+	}
+
+	cfg.Pipeline.Notarius.Enabled = true
+	second, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("enabled executeStages() error = %v", err)
+	}
+	if len(second.Executed) != 1 || len(second.Skipped) != 0 || len(runner.requests) != 1 {
+		t.Fatalf("enabled summary = %#v requests=%d", second, len(runner.requests))
+	}
+	loaded, err := (&manifest.LocalStore{}).Load(context.Background(), second.ManifestPath)
+	if err != nil {
+		t.Fatalf("Load() error = %v", err)
+	}
+	if loaded.Stages["extract"] == nil || loaded.Stages["extract"].Status != manifest.StatusSucceeded || len(loaded.Stages["extract"].Outputs) != 2 {
+		t.Fatalf("extract record = %#v, want succeeded manifest-ready outputs", loaded.Stages["extract"])
+	}
+}
+
+func TestExtractLifecycleChangedOutcomeRerunsSucceededDownstream(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, false)
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("disabled executeStages() error = %v", err)
+	}
+	if analyzeRuns != 1 || len(runner.requests) != 0 {
+		t.Fatalf("disabled run analyze=%d Notarius=%d, want 1 and 0", analyzeRuns, len(runner.requests))
+	}
+
+	cfg.Pipeline.Notarius.Enabled = true
+	summary, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("enabled executeStages() error = %v", err)
+	}
+	if analyzeRuns != 2 || len(runner.requests) != 1 {
+		t.Fatalf("enabled run analyze=%d Notarius=%d, want 2 and 1", analyzeRuns, len(runner.requests))
+	}
+	if len(summary.Executed) != 2 || len(summary.Skipped) != 0 {
+		t.Fatalf("enabled summary = %#v, want extract and analyze executed", summary)
+	}
+}
+
+func TestExtractLifecycleFailureInvalidatesAndOrdinaryRetryRerunsDownstream(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	runner.failuresRemaining = 1
+	markLifecycleStageSucceeded(t, cfg, "analyze")
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err == nil || !strings.Contains(err.Error(), "notarius execution failed") {
+		t.Fatalf("failed executeStages() error = %v", err)
+	}
+	failed := loadLifecycleManifest(t, cfg)
+	if failed.Stages["extract"].Status != manifest.StatusFailed || failed.Stages["analyze"].Status != manifest.StatusStale {
+		t.Fatalf("failed lifecycle extract=%#v analyze=%#v", failed.Stages["extract"], failed.Stages["analyze"])
+	}
+	if failed.Stages["analyze"].Error == nil || failed.Stages["analyze"].Error.Message != staleReasonFailure {
+		t.Fatalf("analyze stale reason = %#v, want %q", failed.Stages["analyze"].Error, staleReasonFailure)
+	}
+
+	summary, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("retry executeStages() error = %v", err)
+	}
+	if analyzeRuns != 1 || len(runner.requests) != 2 || len(summary.Executed) != 2 {
+		t.Fatalf("retry analyze=%d Notarius=%d summary=%#v", analyzeRuns, len(runner.requests), summary)
+	}
+}
+
+func TestExtractLifecycleForcedSelfSkipInvalidatesDownstream(t *testing.T) {
+	cfg, env, _ := extractionLifecycleFixture(t, false)
+	markLifecycleStageSucceeded(t, cfg, "analyze")
+	plan, _ := BuildSingleStagePlan("extract")
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env, Force: true}); err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	loaded := loadLifecycleManifest(t, cfg)
+	if loaded.Stages["extract"].Status != manifest.StatusSkipped || loaded.Stages["analyze"].Status != manifest.StatusStale {
+		t.Fatalf("forced self-skip extract=%#v analyze=%#v", loaded.Stages["extract"], loaded.Stages["analyze"])
+	}
+	if loaded.Stages["analyze"].Error == nil || loaded.Stages["analyze"].Error.Message != staleReasonForcedReplacement {
+		t.Fatalf("analyze stale reason = %#v, want %q", loaded.Stages["analyze"].Error, staleReasonForcedReplacement)
+	}
+}
+
+func TestExtractLifecycleForcedFailureInvalidatesDownstream(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	plan, _ := BuildSingleStagePlan("extract")
+
+	first, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("initial executeStages() error = %v", err)
+	}
+	succeeded := loadLifecycleManifest(t, cfg).Stages["extract"]
+	if succeeded == nil || succeeded.Status != manifest.StatusSucceeded || len(succeeded.Outputs) == 0 || len(succeeded.Logs) == 0 || len(succeeded.Metadata) == 0 {
+		t.Fatalf("initial extraction result = %#v, want succeeded result details", succeeded)
+	}
+
+	markLifecycleStageSucceeded(t, cfg, "analyze")
+	runner.failuresRemaining = 1
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env, Force: true}); err == nil {
+		t.Fatal("executeStages() error = nil, want forced extraction failure")
+	}
+	loaded := loadLifecycleManifest(t, cfg)
+	if loaded.Stages["extract"].Status != manifest.StatusFailed || loaded.Stages["analyze"].Status != manifest.StatusStale {
+		t.Fatalf("forced failure extract=%#v analyze=%#v", loaded.Stages["extract"], loaded.Stages["analyze"])
+	}
+	if loaded.Stages["analyze"].Error == nil || loaded.Stages["analyze"].Error.Message != staleReasonForcedReplacement {
+		t.Fatalf("analyze stale reason = %#v, want %q", loaded.Stages["analyze"].Error, staleReasonForcedReplacement)
+	}
+	failed := loaded.Stages["extract"]
+	if len(failed.Outputs) != 0 || len(failed.Logs) != 0 || len(failed.GeneratedConfigs) != 0 || len(failed.Metadata) != 0 {
+		t.Fatalf("failed replacement inherited extraction result details: %#v", failed)
+	}
+	historical, err := (&manifest.LocalStore{}).LoadRun(context.Background(), first.RunManifestPath)
+	if err != nil {
+		t.Fatalf("LoadRun(initial) error = %v", err)
+	}
+	historicalExtract := historical.Stages["extract"]
+	if historicalExtract == nil || historicalExtract.Status != manifest.StatusSucceeded || len(historicalExtract.Outputs) == 0 || len(historicalExtract.Logs) == 0 || len(historicalExtract.Metadata) == 0 {
+		t.Fatalf("historical extraction result = %#v, want preserved succeeded details", historicalExtract)
+	}
+	if _, err := os.Stat(succeeded.Outputs[0].LocalPath); err != nil {
+		t.Fatalf("durable extraction output was not preserved: %v", err)
+	}
+}
+
+func TestExtractLifecycleRepeatedSelfSkipPreservesSucceededDownstream(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, false)
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("first executeStages() error = %v", err)
+	}
+	second, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("second executeStages() error = %v", err)
+	}
+	if analyzeRuns != 1 || len(runner.requests) != 0 {
+		t.Fatalf("repeated disabled run analyze=%d Notarius=%d, want 1 and 0", analyzeRuns, len(runner.requests))
+	}
+	if len(second.Executed) != 1 || len(second.Skipped) != 2 {
+		t.Fatalf("second summary = %#v, want executed self-skip and skipped analyze", second)
+	}
+	loaded := loadLifecycleManifest(t, cfg)
+	if loaded.Stages["analyze"].Status != manifest.StatusSucceeded {
+		t.Fatalf("analyze = %#v, want succeeded", loaded.Stages["analyze"])
+	}
+}
+
+func TestExtractLifecycleSkipsCurrentResumableResult(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("first executeStages() error = %v", err)
+	}
+	resumed, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("resume executeStages() error = %v", err)
+	}
+	if len(resumed.Executed) != 0 || len(resumed.Skipped) != 2 || len(runner.requests) != 1 || analyzeRuns != 1 {
+		t.Fatalf("resume summary = %#v requests=%d analyze=%d", resumed, len(runner.requests), analyzeRuns)
+	}
+}
+
+func TestExtractLifecycleResumesAndRerunsObsoleteResults(t *testing.T) {
+	for _, test := range []struct {
+		name   string
+		mutate func(*testing.T, *config.Config, *manifest.Manifest)
+	}{
+		{name: "configuration changed", mutate: func(_ *testing.T, cfg *config.Config, _ *manifest.Manifest) {
+			output := cfg.Pipeline.Notarius.Outputs["npc_registry"]
+			output.SchemaVersion = "v2"
+			cfg.Pipeline.Notarius.Outputs["npc_registry"] = output
+		}},
+		{name: "payload missing", mutate: func(t *testing.T, _ *config.Config, m *manifest.Manifest) {
+			if err := os.Remove(m.Stages["extract"].Outputs[0].LocalPath); err != nil {
+				t.Fatalf("Remove() error = %v", err)
+			}
+		}},
+		{name: "payload tampered", mutate: func(t *testing.T, _ *config.Config, m *manifest.Manifest) {
+			if err := os.WriteFile(m.Stages["extract"].Outputs[0].LocalPath, []byte(`{"npcs":["tampered"]}`), 0o644); err != nil {
+				t.Fatalf("WriteFile() error = %v", err)
+			}
+		}},
+		{name: "record incompatible", mutate: func(_ *testing.T, _ *config.Config, m *manifest.Manifest) {
+			m.Stages["extract"].Outputs[0].Contract.SchemaID = "incompatible"
+		}},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			cfg, env, runner := extractionLifecycleFixture(t, true)
+			plan, _ := BuildSingleStagePlan("extract")
+			first, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+			if err != nil {
+				t.Fatalf("first executeStages() error = %v", err)
+			}
+			persisted, err := (&manifest.LocalStore{}).Load(context.Background(), first.ManifestPath)
+			if err != nil {
+				t.Fatalf("Load() error = %v", err)
+			}
+
+			test.mutate(t, cfg, persisted)
+			if err := (&manifest.LocalStore{}).Save(context.Background(), first.ManifestPath, persisted); err != nil {
+				t.Fatalf("Save(mutated) error = %v", err)
+			}
+			rerun, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+			if err != nil {
+				t.Fatalf("rerun executeStages() error = %v", err)
+			}
+			if len(rerun.Executed) != 1 || len(rerun.Skipped) != 0 || len(runner.requests) != 2 {
+				t.Fatalf("rerun summary = %#v requests=%d", rerun, len(runner.requests))
+			}
+		})
+	}
+}
+
+func TestExtractLifecycleUnsafeResumeErrorPreservesSuccess(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+	first, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("first executeStages() error = %v", err)
+	}
+	store := &manifest.LocalStore{}
+	persisted, err := store.Load(context.Background(), first.ManifestPath)
+	if err != nil {
+		t.Fatalf("Load() error = %v", err)
+	}
+	persisted.Stages["extract"].Outputs[0].LocalPath = filepath.Join(cfg.Pipeline.Workspace.Root, "outside.json")
+	if err := store.Save(context.Background(), first.ManifestPath, persisted); err != nil {
+		t.Fatalf("Save() error = %v", err)
+	}
+	before, _ := json.Marshal(map[string]*manifest.StageRecord{
+		"extract": persisted.Stages["extract"],
+		"analyze": persisted.Stages["analyze"],
+	})
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err == nil || !strings.Contains(err.Error(), "unsafe") {
+		t.Fatalf("executeStages() error = %v, want unsafe resume failure", err)
+	}
+	afterManifest, err := store.Load(context.Background(), first.ManifestPath)
+	if err != nil {
+		t.Fatalf("Load(after) error = %v", err)
+	}
+	after, _ := json.Marshal(map[string]*manifest.StageRecord{
+		"extract": afterManifest.Stages["extract"],
+		"analyze": afterManifest.Stages["analyze"],
+	})
+	if string(before) != string(after) || len(runner.requests) != 1 || analyzeRuns != 1 {
+		t.Fatalf("successful records changed: before=%s after=%s requests=%d analyze=%d", before, after, len(runner.requests), analyzeRuns)
+	}
+}
+
+func extractionLifecyclePlan(t *testing.T, analyzeRuns *int) []stage.Stage {
+	t.Helper()
+	plan, err := BuildSingleStagePlan("extract")
+	if err != nil {
+		t.Fatalf("BuildSingleStagePlan(extract) error = %v", err)
+	}
+	return append(plan, countingStage{name: "analyze", runs: analyzeRuns})
+}
+
+func loadLifecycleManifest(t *testing.T, cfg *config.Config) *manifest.Manifest {
+	t.Helper()
+	loaded, err := (&manifest.LocalStore{}).Load(context.Background(), manifestPathFor(cfg))
+	if err != nil {
+		t.Fatalf("Load() error = %v", err)
+	}
+	return loaded
+}
+
+func markLifecycleStageSucceeded(t *testing.T, cfg *config.Config, name string) {
+	t.Helper()
+	loaded := loadLifecycleManifest(t, cfg)
+	loaded.MarkStageSucceeded(name, time.Now().UTC(), nil)
+	if err := (&manifest.LocalStore{}).Save(context.Background(), manifestPathFor(cfg), loaded); err != nil {
+		t.Fatalf("Save() error = %v", err)
+	}
+}
+
+func extractionLifecycleFixture(t *testing.T, enabled bool) (*config.Config, *stage.Env, *materializingNotariusRunner) {
+	t.Helper()
+	cfg := testConfig(t)
+	root := cfg.Pipeline.Workspace.Root
+	binary := filepath.Join(root, "notarius")
+	configPath := filepath.Join(root, "notarius.yml")
+	workingDirectory := filepath.Join(root, "notarius-work")
+	if err := os.WriteFile(binary, []byte("#!/bin/sh\nexit 0\n"), 0o755); err != nil {
+		t.Fatalf("WriteFile(binary) error = %v", err)
+	}
+	if err := os.WriteFile(configPath, []byte("pipelines: {}\n"), 0o644); err != nil {
+		t.Fatalf("WriteFile(config) error = %v", err)
+	}
+	if err := os.Mkdir(workingDirectory, 0o755); err != nil {
+		t.Fatalf("Mkdir(working directory) error = %v", err)
+	}
+	cfg.Pipeline.Notarius = &config.NotariusConfig{
+		Enabled: enabled, Binary: binary, ConfigPath: configPath, PipelineID: "dnd-session",
+		Timeout: "45m", WorkingDirectory: workingDirectory,
+		Outputs: map[string]config.NotariusOutputConfig{
+			"npc_registry": {
+				LaneID: "npc-registry", MediaType: "application/json", SchemaID: "notarius.dnd.npc_registry",
+				SchemaVersion: "v1", ModuleKey: "dnd/npc-registry",
+			},
+		},
+	}
+	paths, err := artifacts.NewLocalStore(root).EnsureLayoutFor(cfg.Session.Campaign, cfg.Session.SessionID)
+	if err != nil {
+		t.Fatalf("EnsureLayoutFor() error = %v", err)
+	}
+	inputPath := filepath.Join(paths.ArtifactsDir, "trimmed.from-manifest.json")
+	if err := os.WriteFile(inputPath, []byte(`{"segments":[]}`), 0o644); err != nil {
+		t.Fatalf("WriteFile(input) error = %v", err)
+	}
+	m := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	m.Campaign = cfg.Session.Campaign
+	m.MarkStageSucceeded("trim", time.Now().UTC(), []manifest.ArtifactRecord{{
+		Kind: artifactmodel.TranscriptOutputKindFinalTrimmed, SourceID: artifactmodel.SourceTranscriptFinalTrimmed,
+		LocalPath: inputPath,
+	}})
+	if err := (&manifest.LocalStore{}).Save(context.Background(), paths.ManifestPath, m); err != nil {
+		t.Fatalf("Save(seed) error = %v", err)
+	}
+	runner := &materializingNotariusRunner{cfg: cfg.Pipeline.Notarius}
+	return cfg, &stage.Env{Notarius: runner}, runner
+}

internal/app/operator_artifact_rendering.go

@@ -10,9 +10,10 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactpolicy"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
 )
 
-func buildHelperArtifactCatalog(cfg *config.Config) (*artifacts.ArtifactCatalog, error) {
+func buildHelperArtifactCatalog(cfg *config.Config, m *manifest.Manifest) (*artifacts.ArtifactCatalog, error) {
 	catalog := artifacts.NewArtifactCatalog()
 	if err := catalog.RegisterBuiltIns(); err != nil {
 		return nil, err
@@ -26,6 +27,14 @@ func buildHelperArtifactCatalog(cfg *config.Config) (*artifacts.ArtifactCatalog,
 	if err := catalog.RegisterConfiguredArtifacts(configured, nil); err != nil {
 		return nil, err
 	}
+	extractionDefinitions := artifacts.ExtractionDefinitionsFromConfig(cfg.Pipeline.Notarius)
+	if err := catalog.RegisterExtractionArtifacts(extractionDefinitions); err != nil {
+		return nil, err
+	}
+	paths := artifacts.NewLocalStore(cfg.Pipeline.Workspace.Root).SessionPathsFor(cfg.Session.Campaign, cfg.Session.SessionID)
+	if cfg.Pipeline.Notarius != nil && cfg.Pipeline.Notarius.Enabled {
+		catalog.HydrateExtractionArtifacts(paths, m, extractionDefinitions)
+	}
 	return catalog, nil
 }
 
@@ -40,6 +49,14 @@ func writeArtifactList(out io.Writer, cfg *config.Config, catalog *artifacts.Art
 	for _, entry := range catalog.ListConfigured() {
 		writeArtifactLine(out, entry.SourceID, lockSet)
 	}
+	fmt.Fprintln(out, "Extraction:")
+	for _, entry := range catalog.ListExtraction() {
+		state := "unavailable"
+		if entry.Available {
+			state = "available"
+		}
+		writeExtractionArtifactLine(out, entry.SourceID, state, entry.Provenance, lockSet)
+	}
 	fmt.Fprintln(out, "Previous-session:")
 	for _, req := range artifacts.CollectPreviousArtifactRequirements(configuredScriptoriumArtifacts(cfg)) {
 		fmt.Fprintf(out, "- %s required=%t\n", artifactpolicy.PreviousSessionSourceID(req.Name), req.Required)
@@ -50,6 +67,17 @@ func writeArtifactList(out io.Writer, cfg *config.Config, catalog *artifacts.Art
 	}
 }
 
+func writeExtractionArtifactLine(out io.Writer, source, state, provenance string, lockSet map[string]config.PublishLockRule) {
+	parts := []string{source, "planned", state}
+	if strings.TrimSpace(provenance) != "" {
+		parts = append(parts, "provenance="+strings.TrimSpace(provenance))
+	}
+	if _, ok := lockSet[source]; ok {
+		parts = append(parts, "locked")
+	}
+	fmt.Fprintf(out, "- %s\n", strings.Join(parts, " "))
+}
+
 func writeArtifactLine(out io.Writer, source string, lockSet map[string]config.PublishLockRule) {
 	parts := []string{source}
 	if _, ok := lockSet[source]; ok {
@@ -103,7 +131,12 @@ func remotePublishedOutputAvailability(ctx context.Context, cfg *config.Config,
 
 func helperPublishedOutputDest(rule config.PublishOutputRule, catalog *artifacts.ArtifactCatalog) (string, bool, error) {
 	source := strings.TrimSpace(rule.Source)
-	normalized, err := artifactpolicy.ResolvePublishedDestination(source, rule.Dest, helperConfiguredOutputPathMap(catalog))
+	normalized, err := artifactpolicy.ResolvePublishedDestinationWithExtractions(
+		source,
+		rule.Dest,
+		helperConfiguredOutputPathMap(catalog),
+		helperExtractionOutputSet(catalog),
+	)
 	if err != nil {
 		return "", false, err
 	}
@@ -112,6 +145,19 @@ func helperPublishedOutputDest(rule config.PublishOutputRule, catalog *artifacts
 	return normalized, showDest, nil
 }
 
+func helperExtractionOutputSet(catalog *artifacts.ArtifactCatalog) map[string]struct{} {
+	out := map[string]struct{}{}
+	if catalog == nil {
+		return out
+	}
+	for _, entry := range catalog.ListExtraction() {
+		if strings.TrimSpace(entry.ExtractionKey) != "" {
+			out[entry.ExtractionKey] = struct{}{}
+		}
+	}
+	return out
+}
+
 func helperConfiguredOutputPathMap(catalog *artifacts.ArtifactCatalog) map[string]string {
 	out := map[string]string{}
 	if catalog == nil {

internal/app/operator_artifacts_list.go

@@ -22,11 +22,11 @@ func ArtifactsList(ctx context.Context, args []string, out io.Writer) error {
 	if strings.TrimSpace(flags.sessionID) == "" {
 		return fmt.Errorf("artifacts list: session_id is required")
 	}
-	cfg, store, locks, _, err := loadHelperContext(ctx, flags, remote)
+	cfg, store, locks, m, err := loadHelperContext(ctx, flags, remote)
 	if err != nil {
 		return fmt.Errorf("artifacts list: %w", err)
 	}
-	catalog, err := buildHelperArtifactCatalog(cfg)
+	catalog, err := buildHelperArtifactCatalog(cfg, m)
 	if err != nil {
 		return fmt.Errorf("artifacts list: %w", err)
 	}

internal/app/operator_helpers_test.go

@@ -8,8 +8,10 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
@@ -508,7 +510,7 @@ func TestExecuteLocksAddListAndRemoveUseRemoteLockStore(t *testing.T) {
 	if code != 0 {
 		t.Fatalf("locks remove exit code = %d, want 0; stderr=%q", code, stderr.String())
 	}
-	store, err := config.LoadPublishLockStoreBytes("locks.yml", fake.Objects[key].Data, nil)
+	store, err := config.LoadPublishLockStoreBytes("locks.yml", fake.Objects[key].Data, nil, nil)
 	if err != nil {
 		t.Fatalf("LoadPublishLockStoreBytes() error = %v", err)
 	}
@@ -772,6 +774,71 @@ func TestExecuteArtifactsListRemoteReportsPublishedAvailability(t *testing.T) {
 	}
 }
 
+func TestExecuteArtifactsListReportsExtractionLifecycleWithoutPayload(t *testing.T) {
+	workspaceRoot := t.TempDir()
+	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
+	addExtractionOutputToPipeline(t, pipelinePath)
+	addPublishOutputsToPipeline(t, pipelinePath, `
+  outputs:
+    - source: narratio.extraction.encounters
+      dest: artifacts/encounters.json
+      required: true
+`)
+	lanePath := writeOperatorExtractionManifest(t, workspaceRoot)
+	fake := &storage.FakeBackend{}
+	publishedKey := artifacts.S3PublishedOutputKey(
+		artifacts.S3SessionPrefix("dnd", "sample-campaign", "2026-05-03"),
+		"artifacts/encounters.json",
+	)
+	fake.SeedObject(storage.FakeObject{Key: publishedKey, Data: []byte(`{"secret":"DO_NOT_PRINT"}`)})
+	var storeInitCalls int
+	restoreAppConfigTestGlobals(t, fake, &storeInitCalls, []string{sessionPath})
+
+	var stdout bytes.Buffer
+	var stderr bytes.Buffer
+	code := Execute([]string{
+		"session", "artifacts", "2026-05-03",
+		"--config", pipelinePath,
+		"--campaign-file", campaignPath,
+		"--session", sessionPath,
+		"--remote",
+	}, &stdout, &stderr)
+	if code != 0 {
+		t.Fatalf("exit code = %d, want 0; stderr=%q", code, stderr.String())
+	}
+	out := stdout.String()
+	for _, want := range []string{
+		"Extraction:",
+		"narratio.extraction.encounters planned available provenance=manifest.current_extract_run",
+		"narratio.extraction.encounters dest=artifacts/encounters.json remote=published",
+	} {
+		if !strings.Contains(out, want) {
+			t.Fatalf("stdout = %q, want %q", out, want)
+		}
+	}
+	if strings.Contains(out, "DO_NOT_PRINT") {
+		t.Fatalf("operator output exposed extraction payload: %q", out)
+	}
+
+	if err := os.Remove(lanePath); err != nil {
+		t.Fatal(err)
+	}
+	stdout.Reset()
+	stderr.Reset()
+	code = Execute([]string{
+		"session", "artifacts", "2026-05-03",
+		"--config", pipelinePath,
+		"--campaign-file", campaignPath,
+		"--session", sessionPath,
+	}, &stdout, &stderr)
+	if code != 0 {
+		t.Fatalf("unavailable exit code = %d, want 0; stderr=%q", code, stderr.String())
+	}
+	if !strings.Contains(stdout.String(), "narratio.extraction.encounters planned unavailable") {
+		t.Fatalf("stdout = %q, want unavailable extraction state", stdout.String())
+	}
+}
+
 func TestExecuteArtifactsListRemoteUsesPublishOutputDestinations(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
@@ -1006,7 +1073,7 @@ func TestExecutePublishLoadsRemoteLocks(t *testing.T) {
 	restoreAppConfigTestGlobals(t, fake, &storeInitCalls, []string{sessionPath})
 
 	workRoot := filepath.Join(workspaceRoot, "work", "sample-campaign", "2026-05-03")
-	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze"} {
+	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze"} {
 		// The publish stage only checks the manifest statuses and source files.
 		_ = stageName
 	}
@@ -1041,6 +1108,72 @@ func addPublishOutputsToPipeline(t *testing.T, pipelinePath, publishYAML string)
 	}
 }
 
+func addExtractionOutputToPipeline(t *testing.T, pipelinePath string) {
+	t.Helper()
+	data, err := os.ReadFile(pipelinePath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	data = append(data, []byte(`notarius:
+  enabled: true
+  config_path: notarius.yml
+  pipeline_id: campaign.extract
+  outputs:
+    encounters:
+      lane_id: encounters
+      media_type: application/json
+      schema_id: encounters
+      schema_version: "1"
+      module_key: encounters
+`)...)
+	if err := os.WriteFile(pipelinePath, data, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(filepath.Dir(pipelinePath), "notarius.yml"), []byte("{}\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func writeOperatorExtractionManifest(t *testing.T, workspaceRoot string) string {
+	t.Helper()
+	paths := artifacts.NewLocalStore(workspaceRoot).SessionPathsFor("sample-campaign", "2026-05-03")
+	bundleRoot := filepath.Join(paths.ArtifactsDir, "notarius", "extract-run-1")
+	lanePath := filepath.Join(bundleRoot, "lanes", "encounters.json")
+	indexPath := filepath.Join(bundleRoot, "index.json")
+	mustWriteTestFile(t, lanePath, `{"secret":"DO_NOT_PRINT"}`)
+	mustWriteTestFile(t, indexPath, `{"lanes":[]}`)
+	laneChecksum, err := artifacts.SHA256File(lanePath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	indexChecksum, err := artifacts.SHA256File(indexPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	m := manifest.New("2026-05-03", time.Now().UTC())
+	m.Campaign = "sample-campaign"
+	m.Stages["extract"] = &manifest.StageRecord{
+		Name: "extract", Status: manifest.StatusSucceeded,
+		Metadata: map[string]any{
+			"narratio_run_id": "extract-run-1", "bundle_root": bundleRoot,
+			"receipt": map[string]any{"run_id": "notarius-run-1", "pipeline_id": "campaign.extract"},
+		},
+		Outputs: []manifest.ArtifactRecord{
+			{
+				Kind: "notarius_lane", SourceID: artifacts.ExtractionArtifactSourceID("encounters"), LocalPath: lanePath,
+				ProducerRunID: "extract-run-1", Checksum: laneChecksum,
+				Contract:           &artifactmodel.ContractMetadata{MediaType: "application/json", SchemaID: "encounters", SchemaVersion: "1", ModuleKey: "encounters"},
+				ExternalProvenance: &artifactmodel.ExternalProvenance{System: "notarius", RunID: "notarius-run-1", PipelineID: "campaign.extract", ArtifactID: "encounters"},
+			},
+			{Kind: "notarius_index", LocalPath: indexPath, ProducerRunID: "extract-run-1", Checksum: indexChecksum},
+		},
+	}
+	if err := (&manifest.LocalStore{}).Save(context.Background(), paths.ManifestPath, m); err != nil {
+		t.Fatal(err)
+	}
+	return lanePath
+}
+
 func replaceInFileOrFatal(t *testing.T, path, old, new string) {
 	t.Helper()
 	data, err := os.ReadFile(path)
@@ -1076,7 +1209,7 @@ func writeValidPublishRunConfigFiles(t *testing.T, workspaceRoot string) (string
 	m := manifest.New("2026-05-03", nowUTC())
 	m.Campaign = "sample-campaign"
 	m.RunID = "20260521T160000Z-test"
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze"} {
 		m.MarkStageSucceeded(name, nowUTC(), nil)
 	}
 	path := artifacts.SessionManifestPathForCampaign(cfg.Pipeline.Workspace.Root, cfg.Session.Campaign, cfg.Session.SessionID)

internal/app/operator_locks.go

@@ -67,7 +67,7 @@ func LocksAdd(ctx context.Context, args []string, out io.Writer) error {
 	if err != nil {
 		return fmt.Errorf("locks add: %w", err)
 	}
-	if _, err := config.ValidatePublishLockRules([]config.PublishLockRule{{Source: source}}, cfg.Pipeline.Scriptorium, "locks add"); err != nil {
+	if _, err := config.ValidatePublishLockRules([]config.PublishLockRule{{Source: source}}, cfg.Pipeline.Scriptorium, cfg.Pipeline.Notarius, "locks add"); err != nil {
 		return fmt.Errorf("locks add: %w", err)
 	}
 	if _, ok := lockSourceSet(locks.Static)[source]; ok {
@@ -79,7 +79,7 @@ func LocksAdd(ctx context.Context, args []string, out io.Writer) error {
 	}
 	remoteSet[source] = config.PublishLockRule{Source: source, Reason: strings.TrimSpace(reason)}
 	remoteLocks := lockMapValues(remoteSet)
-	if _, err := config.ValidatePublishLockRules(remoteLocks, cfg.Pipeline.Scriptorium, "locks"); err != nil {
+	if _, err := config.ValidatePublishLockRules(remoteLocks, cfg.Pipeline.Scriptorium, cfg.Pipeline.Notarius, "locks"); err != nil {
 		return fmt.Errorf("locks add: %w", err)
 	}
 	if err := uploadRemoteLockStore(ctx, store, locks.Key, &config.PublishLockStore{Locks: remoteLocks}); err != nil {
@@ -106,7 +106,7 @@ func LocksRemove(ctx context.Context, args []string, out io.Writer) error {
 	if err != nil {
 		return fmt.Errorf("locks remove: %w", err)
 	}
-	if _, err := config.ValidatePublishLockRules([]config.PublishLockRule{{Source: source}}, cfg.Pipeline.Scriptorium, "locks remove"); err != nil {
+	if _, err := config.ValidatePublishLockRules([]config.PublishLockRule{{Source: source}}, cfg.Pipeline.Scriptorium, cfg.Pipeline.Notarius, "locks remove"); err != nil {
 		return fmt.Errorf("locks remove: %w", err)
 	}
 	remoteSet := lockSourceSet(locks.Remote)

internal/app/operator_status.go

@@ -11,6 +11,7 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
 )
 
 // Status reports effective local/remote session state.
@@ -40,11 +41,13 @@ func Status(ctx context.Context, args []string, out io.Writer) error {
 	writeStatusStableInputs(out, inspectStableInputs(cfg))
 	writeStatusLocalAudio(out, inspectLocalAudioPresence(cfg))
 
+	var localManifest *manifest.Manifest
 	if m, err := loadLocalManifest(ctx, paths.ManifestPath); err != nil {
 		fmt.Fprintf(out, "Local manifest: error: %v\n", err)
 	} else if m == nil {
 		fmt.Fprintln(out, "Local manifest: missing")
 	} else {
+		localManifest = m
 		fmt.Fprintf(out, "Local manifest: %s\n", paths.ManifestPath)
 		writeStageStatuses(out, m)
 	}
@@ -72,7 +75,7 @@ func Status(ctx context.Context, args []string, out io.Writer) error {
 	lockChecks := inspectEffectiveLocks(ctx, cfg, store)
 	locks := lockChecks.Locks
 	lockErr := lockChecks.Err
-	if catalog, catalogErr := buildHelperArtifactCatalog(cfg); catalogErr != nil {
+	if catalog, catalogErr := buildHelperArtifactCatalog(cfg, localManifest); catalogErr != nil {
 		fmt.Fprintf(out, "Remote outputs: error: %v\n", catalogErr)
 	} else if storeErr == nil {
 		catalogLocks := locks

internal/app/plan.go

@@ -22,7 +22,7 @@ func Plan(ctx context.Context, args []string, out io.Writer) error {
 	var flags commonConfigFlags
 	var force bool
 	addCommonConfigFlags(fs, &flags)
-	fs.BoolVar(&force, "force", false, "force stage execution (reserved for future behavior)")
+	fs.BoolVar(&force, "force", false, "show all stages as scheduled to rerun")
 
 	if err := parseSessionAwareFlags("plan", fs, args, &flags.sessionID); err != nil {
 		return err

internal/app/plan_test.go

@@ -27,12 +27,12 @@ func TestPlanCreatesAndReusesWorkdir(t *testing.T) {
 	if !strings.Contains(got, "narratio session plan: workdir prepared at") {
 		t.Fatalf("first output = %q, want workdir prepared", got)
 	}
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze", "publish", "notify"} {
 		if !strings.Contains(got, name+": run") {
 			t.Fatalf("first output = %q, missing stage %q", got, name)
 		}
 	}
-	if !strings.Contains(got, "totals: run=10 skip=0") {
+	if !strings.Contains(got, "totals: run=11 skip=0") {
 		t.Fatalf("first output = %q, want totals", got)
 	}
 
@@ -84,8 +84,8 @@ func TestPlanShowsRunAndSkipFromManifest(t *testing.T) {
 	if !strings.Contains(got, "trim: run") {
 		t.Fatalf("output = %q, want trim run", got)
 	}
-	if !strings.Contains(got, "totals: run=8 skip=2") {
-		t.Fatalf("output = %q, want totals run=8 skip=2", got)
+	if !strings.Contains(got, "totals: run=9 skip=2") {
+		t.Fatalf("output = %q, want totals run=9 skip=2", got)
 	}
 }
 

internal/app/planner_test.go

@@ -4,7 +4,7 @@ import "testing"
 
 func TestBuildFullPlanOrder(t *testing.T) {
 	got := BuildFullPlan()
-	want := []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"}
+	want := []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze", "publish", "notify"}
 	if len(got) != len(want) {
 		t.Fatalf("len(plan) = %d, want %d", len(got), len(want))
 	}

internal/app/post_publish_cleanup_test.go

@@ -355,7 +355,7 @@ func publishStageCleanupFixture(t *testing.T) (*config.Config, cleanupSeed, stri
 	if err != nil {
 		t.Fatalf("Load() error = %v", err)
 	}
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze"} {
 		seedManifest.MarkStageSucceeded(name, time.Now().UTC(), nil)
 	}
 	seedManifest.S3SessionPrefix = artifacts.S3SessionPrefix("dnd", cfg.Session.Campaign, cfg.Session.SessionID)

internal/app/remote_locks.go

@@ -55,7 +55,7 @@ func loadRemoteLockStore(ctx context.Context, cfg *config.Config, store storage.
 	if err != nil {
 		return nil, key, fmt.Errorf("read remote locks %q: %w", key, err)
 	}
-	lockStore, err := config.LoadPublishLockStoreBytes("s3://"+s3BucketName(cfg.Pipeline)+"/"+key, data, cfg.Pipeline.Scriptorium)
+	lockStore, err := config.LoadPublishLockStoreBytes("s3://"+s3BucketName(cfg.Pipeline)+"/"+key, data, cfg.Pipeline.Scriptorium, cfg.Pipeline.Notarius)
 	if err != nil {
 		return nil, key, err
 	}

internal/app/restore_execution_test.go

@@ -9,8 +9,10 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
@@ -62,6 +64,55 @@ func TestExecuteRestoreNonDryRunRestoresDurableFiles(t *testing.T) {
 	}
 }
 
+func TestExecuteRestoreRoundTripsPublishedExtractionAndManifestMetadata(t *testing.T) {
+	workspaceRoot := t.TempDir()
+	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
+	fake := &storage.FakeBackend{}
+	cfg, sessionPrefix, manifestKey, runIDKey := seedRestoreCommittedState(t, fake, pipelinePath, campaignPath, sessionPath)
+	seedRestoreObject(fake, sessionPrefix+"artifacts/encounters.json", []byte(`{"encounters":[]}`))
+	seedRestoreObject(fake, runIDKey, []byte("20260519T010203Z-a1b2c3d4\n"))
+
+	remoteManifest := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	remoteManifest.Campaign = cfg.Session.Campaign
+	remoteManifest.RunID = "20260519T010203Z-a1b2c3d4"
+	remoteManifest.Stages["extract"] = &manifest.StageRecord{
+		Name: "extract", Status: manifest.StatusSucceeded,
+		Outputs: []manifest.ArtifactRecord{{
+			Kind: "notarius_lane", SourceID: artifacts.ExtractionArtifactSourceID("encounters"),
+			LocalPath: "/prior/workspace/artifacts/notarius/extract-run-1/lanes/encounters.json",
+			Contract: &artifactmodel.ContractMetadata{
+				MediaType: "application/json", SchemaID: "encounters", SchemaVersion: "1",
+			},
+			ExternalProvenance: &artifactmodel.ExternalProvenance{
+				System: "notarius", RunID: "notarius-run-1", PipelineID: "campaign.extract", ArtifactID: "encounters",
+			},
+		}},
+	}
+	manifestBody, err := json.Marshal(remoteManifest)
+	if err != nil {
+		t.Fatal(err)
+	}
+	seedRestoreObject(fake, manifestKey, manifestBody)
+	restoreWithStoreAndRealPhases(t, fake)
+
+	var stdout bytes.Buffer
+	var stderr bytes.Buffer
+	code := Execute([]string{"session", "restore", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, &stdout, &stderr)
+	if code != 0 {
+		t.Fatalf("exit code = %d, want 0; stderr=%q", code, stderr.String())
+	}
+	sessionRoot := artifacts.SessionWorkDirForCampaign(workspaceRoot, cfg.Session.Campaign, cfg.Session.SessionID)
+	mustReadEquals(t, filepath.Join(sessionRoot, "artifacts", "encounters.json"), `{"encounters":[]}`)
+	restored, err := (&manifest.LocalStore{}).Load(context.Background(), filepath.Join(sessionRoot, "manifest.json"))
+	if err != nil {
+		t.Fatalf("load restored manifest: %v", err)
+	}
+	lane := restored.Stages["extract"].Outputs[0]
+	if lane.Contract == nil || lane.Contract.SchemaID != "encounters" || lane.ExternalProvenance == nil || lane.ExternalProvenance.RunID != "notarius-run-1" {
+		t.Fatalf("restored extraction metadata = %#v", lane)
+	}
+}
+
 func TestExecuteRestoreIncludeAudioRestoresAudio(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)

internal/app/run.go

@@ -18,7 +18,7 @@ func Run(ctx context.Context, args []string, out io.Writer) error {
 	var force bool
 	var selectedArtifacts artifactSelectionFlag
 	addCommonConfigFlags(fs, &flags)
-	fs.BoolVar(&force, "force", false, "force stage execution (reserved for future behavior)")
+	fs.BoolVar(&force, "force", false, "rerun stages even when already succeeded")
 	fs.Var(&selectedArtifacts, "artifacts", "configured artifact names to execute and publish (comma-separated or repeatable)")
 
 	if err := parseSessionAwareFlags("run", fs, args, &flags.sessionID); err != nil {

internal/app/run_control.go

@@ -3,6 +3,7 @@ package app
 import (
 	"context"
 	"fmt"
+	"strings"
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
@@ -23,22 +24,40 @@ type stageDecision struct {
 	Action stageAction
 }
 
+const (
+	staleReasonForcedReplacement = "upstream stage was force-run"
+	staleReasonChangedResult     = "upstream stage result changed"
+	staleReasonFailure           = "upstream stage failed"
+	staleReasonSelfSkip          = "upstream stage self-skipped"
+	staleReasonNotResumable      = "upstream stage result was not resumable"
+)
+
+type priorStageOutcome struct {
+	exists     bool
+	status     manifest.StageStatus
+	skipReason string
+	outputs    int
+}
+
 func decideStageActions(stages []stage.Stage, m *manifest.Manifest, force bool) []stageDecision {
 	out := make([]stageDecision, 0, len(stages))
 	for _, s := range stages {
-		action := stageActionRun
-		// TODO: incorporate stale detection once checksum/input change tracking is implemented.
-		if !force && stageSucceeded(m, s.Name()) {
-			action = stageActionSkip
-		}
 		out = append(out, stageDecision{
 			Stage:  s,
-			Action: action,
+			Action: decideStageAction(s, m, force),
 		})
 	}
 	return out
 }
 
+func decideStageAction(s stage.Stage, m *manifest.Manifest, force bool) stageAction {
+	// TODO: incorporate stale detection once checksum/input change tracking is implemented.
+	if !force && stageSucceeded(m, s.Name()) {
+		return stageActionSkip
+	}
+	return stageActionRun
+}
+
 func stageSucceeded(m *manifest.Manifest, name string) bool {
 	if m == nil || m.Stages == nil {
 		return false
@@ -47,6 +66,29 @@ func stageSucceeded(m *manifest.Manifest, name string) bool {
 	return sr != nil && sr.Status == manifest.StatusSucceeded
 }
 
+func capturePriorStageOutcome(m *manifest.Manifest, name string) priorStageOutcome {
+	if m == nil || m.Stages == nil || m.Stages[name] == nil {
+		return priorStageOutcome{}
+	}
+	record := m.Stages[name]
+	outcome := priorStageOutcome{
+		exists:  true,
+		status:  record.Status,
+		outputs: len(record.Outputs),
+	}
+	if record.Error != nil && record.Error.Code == "skipped" {
+		outcome.skipReason = record.Error.Message
+	}
+	return outcome
+}
+
+func (o priorStageOutcome) isSameSelfSkip(reason string) bool {
+	return o.exists &&
+		o.status == manifest.StatusSkipped &&
+		o.outputs == 0 &&
+		o.skipReason == strings.TrimSpace(reason)
+}
+
 func loadManifestIfPresent(ctx context.Context, cfg *config.Config) (*manifest.Manifest, error) {
 	path := artifacts.SessionManifestPathForCampaign(
 		cfg.Pipeline.Workspace.Root,
@@ -91,7 +133,7 @@ func downstreamStageNames(stageName string) []string {
 	return nil
 }
 
-func invalidateDownstreamSucceededStages(m *manifest.Manifest, upstreamStage string, at time.Time) []string {
+func invalidateDownstreamSucceededStagesWithReason(m *manifest.Manifest, upstreamStage string, at time.Time, reason string) []string {
 	if m == nil || m.Stages == nil {
 		return nil
 	}
@@ -102,7 +144,7 @@ func invalidateDownstreamSucceededStages(m *manifest.Manifest, upstreamStage str
 		if sr == nil || sr.Status != manifest.StatusSucceeded {
 			continue
 		}
-		m.MarkStageStale(downstream, at, "upstream stage rerun with force")
+		m.MarkStageStale(downstream, at, reason)
 		invalidated = append(invalidated, downstream)
 	}
 	return invalidated

internal/app/run_control_test.go

@@ -32,7 +32,7 @@ func TestDecideStageActions(t *testing.T) {
 
 func TestDownstreamStageNames(t *testing.T) {
 	got := downstreamStageNames("polish")
-	want := []string{"normalize", "trim", "render", "analyze", "publish", "notify"}
+	want := []string{"normalize", "trim", "extract", "render", "analyze", "publish", "notify"}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("downstreamStageNames(polish) = %#v, want %#v", got, want)
 	}
@@ -43,7 +43,7 @@ func TestDownstreamStageNames(t *testing.T) {
 	}
 }
 
-func TestInvalidateDownstreamSucceededStages(t *testing.T) {
+func TestInvalidateDownstreamSucceededStagesWithReason(t *testing.T) {
 	now := time.Now().UTC()
 	m := manifest.New("2026-05-03", now)
 	m.MarkStageSucceeded("prepare", now, nil)
@@ -52,15 +52,16 @@ func TestInvalidateDownstreamSucceededStages(t *testing.T) {
 	m.MarkStageSucceeded("polish", now, nil)
 	m.MarkStageSucceeded("normalize", now, nil)
 	m.MarkStageSucceeded("trim", now, nil)
+	m.MarkStageSucceeded("extract", now, nil)
 	m.MarkStageSucceeded("render", now, nil)
 	m.MarkStageFailed("analyze", now, "analysis failed")
 	m.MarkStageSucceeded("publish", now, nil)
 	m.MarkStageSucceeded("notify", now, nil)
 
-	got := invalidateDownstreamSucceededStages(m, "polish", now.Add(1*time.Second))
-	want := []string{"normalize", "trim", "render", "publish", "notify"}
+	got := invalidateDownstreamSucceededStagesWithReason(m, "polish", now.Add(1*time.Second), staleReasonChangedResult)
+	want := []string{"normalize", "trim", "extract", "render", "publish", "notify"}
 	if !reflect.DeepEqual(got, want) {
-		t.Fatalf("invalidateDownstreamSucceededStages() = %#v, want %#v", got, want)
+		t.Fatalf("invalidateDownstreamSucceededStagesWithReason() = %#v, want %#v", got, want)
 	}
 
 	for _, stageName := range want {
@@ -75,3 +76,30 @@ func TestInvalidateDownstreamSucceededStages(t *testing.T) {
 		t.Fatalf("prepare status = %q, want succeeded", m.Stages["prepare"].Status)
 	}
 }
+
+func TestExtractionPositionControlsForceInvalidation(t *testing.T) {
+	now := time.Now().UTC()
+	tests := []struct {
+		upstream string
+		want     []string
+	}{
+		{upstream: "trim", want: []string{"extract", "render", "analyze", "publish", "notify"}},
+		{upstream: "extract", want: []string{"render", "analyze", "publish", "notify"}},
+		{upstream: "render", want: []string{"analyze", "publish", "notify"}},
+	}
+	for _, test := range tests {
+		t.Run(test.upstream, func(t *testing.T) {
+			m := manifest.New("2026-05-03", now)
+			for _, name := range canonicalStageNames() {
+				m.MarkStageSucceeded(name, now, nil)
+			}
+			got := invalidateDownstreamSucceededStagesWithReason(m, test.upstream, now.Add(time.Second), staleReasonForcedReplacement)
+			if !reflect.DeepEqual(got, test.want) {
+				t.Fatalf("invalidated = %#v, want %#v", got, test.want)
+			}
+			if test.upstream == "render" && m.Stages["extract"].Status != manifest.StatusSucceeded {
+				t.Fatalf("forcing render changed extract: %#v", m.Stages["extract"])
+			}
+		})
+	}
+}

internal/app/run_stage.go

@@ -27,7 +27,7 @@ func RunStage(ctx context.Context, args []string, out io.Writer) error {
 	var force bool
 	var selectedArtifacts artifactSelectionFlag
 	addCommonConfigFlags(fs, &flags)
-	fs.BoolVar(&force, "force", false, "force stage execution (reserved for future behavior)")
+	fs.BoolVar(&force, "force", false, "rerun the stage even when already succeeded")
 	fs.Var(&selectedArtifacts, "artifacts", "configured artifact names to execute or publish (comma-separated or repeatable)")
 
 	if err := fs.Parse(args); err != nil {

internal/app/run_stage_test.go

@@ -36,8 +36,8 @@ func TestRunContinuesAfterCompletedStages(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "executed=8 skipped=2") {
-		t.Fatalf("output = %q, want executed=8 skipped=2", out.String())
+	if !strings.Contains(out.String(), "executed=9 skipped=3") {
+		t.Fatalf("output = %q, want executed=9 skipped=3", out.String())
 	}
 
 	loaded, err := store.Load(context.Background(), manifestPath)
@@ -59,6 +59,7 @@ func TestRunNoRemainingStagesRecordsSkippedStages(t *testing.T) {
 	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
 		m.MarkStageSucceeded(name, time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), nil)
 	}
+	m.MarkStageSkipped("extract", time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), "notarius_disabled")
 	if err := store.Save(context.Background(), manifestPath, m); err != nil {
 		t.Fatalf("save manifest: %v", err)
 	}
@@ -68,8 +69,15 @@ func TestRunNoRemainingStagesRecordsSkippedStages(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "executed=0 skipped=10") {
-		t.Fatalf("output = %q, want executed=0 skipped=10", out.String())
+	if !strings.Contains(out.String(), "executed=1 skipped=11") {
+		t.Fatalf("output = %q, want disabled extraction to self-skip", out.String())
+	}
+	loaded, err := store.Load(context.Background(), manifestPath)
+	if err != nil {
+		t.Fatalf("load migrated manifest: %v", err)
+	}
+	if loaded.Stages["extract"] == nil || loaded.Stages["extract"].Status != manifest.StatusSkipped {
+		t.Fatalf("extract record = %#v, want stable disabled skip", loaded.Stages["extract"])
 	}
 }
 
@@ -85,7 +93,7 @@ func TestRunForceRerunsSucceeded(t *testing.T) {
 
 	store := &manifest.LocalStore{}
 	m := manifest.New("2026-05-03", time.Date(2026, 5, 3, 10, 0, 0, 0, time.UTC))
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze", "publish", "notify"} {
 		m.MarkStageSucceeded(name, time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), nil)
 	}
 	if err := store.Save(context.Background(), manifestPath, m); err != nil {
@@ -97,7 +105,7 @@ func TestRunForceRerunsSucceeded(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "executed=10 skipped=0") {
+	if !strings.Contains(out.String(), "executed=11 skipped=1") {
 		t.Fatalf("output = %q, want forced full rerun", out.String())
 	}
 }
@@ -132,6 +140,27 @@ func TestRunStageExecutesOnlySelectedStage(t *testing.T) {
 	}
 }
 
+func TestRunStageExtractIsAcceptedAndSelfSkipsWhenDisabled(t *testing.T) {
+	workspaceRoot := t.TempDir()
+	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
+
+	var out bytes.Buffer
+	err := RunStage(context.Background(), []string{"extract", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, &out)
+	if err != nil {
+		t.Fatalf("RunStage(extract) error = %v", err)
+	}
+	if !strings.Contains(out.String(), "stage=extract executed=1 skipped=1 force=false") {
+		t.Fatalf("output = %q, want disabled extraction self-skip", out.String())
+	}
+	loaded, err := (&manifest.LocalStore{}).Load(context.Background(), filepath.Join(workspaceRoot, "work", "sample-campaign", "2026-05-03", "manifest.json"))
+	if err != nil {
+		t.Fatalf("Load() error = %v", err)
+	}
+	if loaded.Stages["extract"] == nil || loaded.Stages["extract"].Status != manifest.StatusSkipped {
+		t.Fatalf("extract record = %#v, want skipped", loaded.Stages["extract"])
+	}
+}
+
 func TestRunStageSkipAndForce(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspaceRoot)
@@ -176,7 +205,7 @@ func TestRunStageForceMarksDownstreamStaleAndRunContinuesFromStale(t *testing.T)
 
 	store := &manifest.LocalStore{}
 	seed := manifest.New("2026-05-03", time.Date(2026, 5, 3, 10, 0, 0, 0, time.UTC))
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze", "publish", "notify"} {
 		seed.MarkStageSucceeded(name, time.Date(2026, 5, 3, 10, 1, 0, 0, time.UTC), nil)
 	}
 	if err := store.Save(context.Background(), manifestPath, seed); err != nil {
@@ -196,7 +225,7 @@ func TestRunStageForceMarksDownstreamStaleAndRunContinuesFromStale(t *testing.T)
 	if err != nil {
 		t.Fatalf("load manifest after force: %v", err)
 	}
-	for _, name := range []string{"normalize", "trim", "render", "analyze", "publish", "notify"} {
+	for _, name := range []string{"normalize", "trim", "extract", "render", "analyze", "publish", "notify"} {
 		if afterForce.Stages[name] == nil || afterForce.Stages[name].Status != manifest.StatusStale {
 			t.Fatalf("stage %q = %#v, want stale", name, afterForce.Stages[name])
 		}
@@ -207,7 +236,7 @@ func TestRunStageForceMarksDownstreamStaleAndRunContinuesFromStale(t *testing.T)
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
-	if !strings.Contains(out.String(), "executed=6 skipped=4") {
+	if !strings.Contains(out.String(), "executed=7 skipped=5") {
 		t.Fatalf("output = %q, want run to execute stale downstream stages", out.String())
 	}
 }

internal/app/runner.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/audita"
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/notarius"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/notify"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/scriptorium"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/seriatim"
@@ -80,6 +81,9 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 		}
 		env.Audita = runner
 	}
+	if env.Notarius == nil && needsNotariusForRun(env.Config, stages) {
+		env.Notarius = notarius.NewSubprocessRunner()
+	}
 	if env.Scriptorium == nil {
 		env.Scriptorium = scriptorium.NewSubprocessRunner()
 	}
@@ -167,6 +171,29 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 	for _, d := range decisions {
 		s := d.Stage
 		runNames = append(runNames, s.Name())
+		d.Action = decideStageAction(s, m, opts.Force)
+
+		if d.Action == stageActionSkip {
+			if validator, ok := s.(stage.ResumeValidator); ok {
+				validation, err := validator.ValidateResume(ctx, stageEnv, m)
+				if err != nil {
+					return nil, fmt.Errorf("validate resume for stage %q: %w", s.Name(), err)
+				}
+				validation = validation.Normalized()
+				if !validation.Resumable {
+					staleAt := nowUTC()
+					m.MarkStageStale(s.Name(), staleAt, validation.Reason)
+					invalidateDownstreamSucceededStagesWithReason(
+						m, s.Name(), staleAt, staleReasonNotResumable,
+					)
+					if err := env.ManifestStore.Save(ctx, manifestPath, m); err != nil {
+						return nil, fmt.Errorf("save manifest after resume validation for stage %q: %w", s.Name(), err)
+					}
+					env.Logger.Info("stage result is not resumable", "stage", s.Name(), "reason", validation.Reason)
+					d.Action = stageActionRun
+				}
+			}
+		}
 
 		if d.Action == stageActionSkip {
 			skipped = append(skipped, s.Name())
@@ -180,6 +207,7 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 			continue
 		}
 		executed = append(executed, s.Name())
+		priorOutcome := capturePriorStageOutcome(m, s.Name())
 
 		now := nowUTC()
 		runManifest.SetStageAction(s.Name(), manifest.RunStageActionRun, now)
@@ -188,6 +216,9 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 			return nil, fmt.Errorf("save run manifest before stage %q: %w", s.Name(), err)
 		}
 		m.MarkStageRunning(s.Name(), now)
+		if opts.Force {
+			invalidateDownstreamSucceededStagesWithReason(m, s.Name(), now, staleReasonForcedReplacement)
+		}
 		env.Logger.Info("starting stage", "stage", s.Name())
 		if err := env.ManifestStore.Save(ctx, manifestPath, m); err != nil {
 			return nil, fmt.Errorf("save manifest before stage %q: %w", s.Name(), err)
@@ -195,9 +226,13 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 		env.Logger.Debug("manifest saved", "stage", s.Name(), "transition", "running", "path", manifestPath)
 
 		result, err := s.Run(ctx, stageEnv, m)
+		if err == nil {
+			err = validateStageResult(result)
+		}
 		if err != nil {
 			failedAt := nowUTC()
 			m.MarkStageFailed(s.Name(), failedAt, err.Error())
+			invalidateDownstreamSucceededStagesWithReason(m, s.Name(), failedAt, staleReasonFailure)
 			if saveErr := env.ManifestStore.Save(ctx, manifestPath, m); saveErr != nil {
 				return nil, fmt.Errorf("stage %q failed (%v) and manifest save failed (%v)", s.Name(), err, saveErr)
 			}
@@ -209,13 +244,34 @@ func executeStages(ctx context.Context, cfg *config.Config, stages []stage.Stage
 			env.Logger.Info("stage failed", "stage", s.Name(), "error", err)
 			return nil, fmt.Errorf("stage %q failed: %w", s.Name(), err)
 		}
+		if result != nil && result.Disposition == stage.StageDispositionSkipped {
+			skipped = append(skipped, s.Name())
+			skippedAt := nowUTC()
+			m.MarkStageSkipped(s.Name(), skippedAt, result.SkipReason)
+			applyStageResultToManifest(m, s.Name(), result)
+			if !priorOutcome.isSameSelfSkip(result.SkipReason) {
+				invalidateDownstreamSucceededStagesWithReason(m, s.Name(), skippedAt, staleReasonSelfSkip)
+			}
+			if err := env.ManifestStore.Save(ctx, manifestPath, m); err != nil {
+				return nil, fmt.Errorf("save manifest after self-skip %q: %w", s.Name(), err)
+			}
+			runManifest.MarkStageSkipped(s.Name(), skippedAt, result.SkipReason)
+			applyStageResultToRunManifest(runManifest, s.Name(), result)
+			syncRunManifestIdentityFromSession(m, runManifest)
+			if err := runManifestStore.SaveRun(ctx, runManifestPath, runManifest); err != nil {
+				return nil, fmt.Errorf("save run manifest after self-skip %q: %w", s.Name(), err)
+			}
+			env.Logger.Debug("manifest saved", "stage", s.Name(), "transition", "skipped", "path", manifestPath)
+			env.Logger.Info("stage skipped", "stage", s.Name(), "reason", result.SkipReason)
+			continue
+		}
 
 		outputs := mapResultOutputs(s.Name(), result, runID)
 		succeededAt := nowUTC()
 		m.MarkStageSucceeded(s.Name(), succeededAt, outputs)
 		applyStageResultToManifest(m, s.Name(), result)
-		if opts.Force {
-			invalidateDownstreamSucceededStages(m, s.Name(), succeededAt)
+		if !priorOutcome.exists || priorOutcome.status != manifest.StatusSucceeded {
+			invalidateDownstreamSucceededStagesWithReason(m, s.Name(), succeededAt, staleReasonChangedResult)
 		}
 
 		if err := env.ManifestStore.Save(ctx, manifestPath, m); err != nil {
@@ -407,26 +463,69 @@ func mapResultOutputs(stageName string, result *stage.StageResult, runID string)
 			localPath = ref.RelativePath
 		}
 		kind := ref.Kind
-		sourceID := ""
-		if stageName == "analyze" {
-			sourceID = artifacts.ConfiguredArtifactSourceID(ref.Kind)
-			kind = "scriptorium_artifact"
-		} else {
-			sourceID = sourceIDForOutputKind(kind)
+		sourceID := strings.TrimSpace(ref.SourceID)
+		if sourceID == "" {
+			if stageName == "analyze" {
+				sourceID = artifacts.ConfiguredArtifactSourceID(ref.Kind)
+				kind = "scriptorium_artifact"
+			} else {
+				sourceID = sourceIDForOutputKind(kind)
+			}
 		}
 		out = append(out, manifest.ArtifactRecord{
-			Kind:          kind,
-			SourceID:      sourceID,
-			LocalPath:     localPath,
-			ProducerRunID: runID,
-			RemoteKey:     ref.RemoteKey,
-			Checksum:      ref.Checksum,
+			Kind:               kind,
+			SourceID:           sourceID,
+			LocalPath:          localPath,
+			Contract:           cloneContractMetadata(ref.Contract),
+			ExternalProvenance: cloneExternalProvenance(ref.ExternalProvenance),
+			ProducerRunID:      runID,
+			RemoteKey:          ref.RemoteKey,
+			Checksum:           ref.Checksum,
 		})
 	}
 
 	return out
 }
 
+func validateStageResult(result *stage.StageResult) error {
+	if result == nil {
+		return nil
+	}
+	switch result.Disposition {
+	case stage.StageDispositionSucceeded:
+		if strings.TrimSpace(result.SkipReason) != "" {
+			return fmt.Errorf("successful result contains a skip reason")
+		}
+		return nil
+	case stage.StageDispositionSkipped:
+		if strings.TrimSpace(result.SkipReason) == "" {
+			return fmt.Errorf("skipped result requires a skip reason")
+		}
+		if len(result.Outputs) != 0 {
+			return fmt.Errorf("skipped result contains %d output(s)", len(result.Outputs))
+		}
+		return nil
+	default:
+		return fmt.Errorf("unsupported stage result disposition %q", result.Disposition)
+	}
+}
+
+func cloneContractMetadata(value *artifactmodel.ContractMetadata) *artifactmodel.ContractMetadata {
+	if value == nil {
+		return nil
+	}
+	cloned := *value
+	return &cloned
+}
+
+func cloneExternalProvenance(value *artifactmodel.ExternalProvenance) *artifactmodel.ExternalProvenance {
+	if value == nil {
+		return nil
+	}
+	cloned := *value
+	return &cloned
+}
+
 func sourceIDForOutputKind(kind string) string {
 	trimmed := strings.TrimSpace(kind)
 	if trimmed == "" {
@@ -598,6 +697,18 @@ func needsObjectStoreForRun(cfg *config.Config, stages []stage.Stage) bool {
 	return true
 }
 
+func needsNotariusForRun(cfg *config.Config, stages []stage.Stage) bool {
+	if cfg == nil || cfg.Pipeline == nil || cfg.Pipeline.Notarius == nil || !cfg.Pipeline.Notarius.Enabled {
+		return false
+	}
+	for _, candidate := range stages {
+		if candidate != nil && candidate.Name() == "extract" {
+			return true
+		}
+	}
+	return false
+}
+
 func needsRemoteLocksForRun(cfg *config.Config, stages []stage.Stage) bool {
 	if cfg == nil || cfg.Pipeline == nil || cfg.Session == nil {
 		return false

internal/app/runner_test.go

@@ -2,6 +2,7 @@ package app
 
 import (
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"os"
@@ -16,6 +17,7 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/seriatim"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/whisperx"
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
@@ -38,6 +40,25 @@ type countingStage struct {
 	runs *int
 }
 
+type resultStage struct {
+	name   string
+	result *stage.StageResult
+	runs   *int
+	order  *[]string
+}
+
+func (s resultStage) Name() string           { return s.name }
+func (s resultStage) Declares() stage.IODecl { return stage.IODecl{} }
+func (s resultStage) Run(_ context.Context, _ *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
+	if s.runs != nil {
+		*s.runs = *s.runs + 1
+	}
+	if s.order != nil {
+		*s.order = append(*s.order, s.name)
+	}
+	return s.result, nil
+}
+
 func (s countingStage) Name() string           { return s.name }
 func (s countingStage) Declares() stage.IODecl { return stage.IODecl{} }
 func (s countingStage) Run(_ context.Context, _ *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
@@ -50,6 +71,34 @@ type captureSelectedArtifactsStage struct {
 	captured *[]string
 }
 
+type captureNotariusStage struct {
+	captured *bool
+}
+
+type resumeCheckingStage struct {
+	name        string
+	validation  stage.ResumeValidation
+	validateErr error
+	runs        *int
+}
+
+func (s resumeCheckingStage) Name() string           { return s.name }
+func (s resumeCheckingStage) Declares() stage.IODecl { return stage.IODecl{} }
+func (s resumeCheckingStage) Run(_ context.Context, _ *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
+	*s.runs++
+	return &stage.StageResult{}, nil
+}
+func (s resumeCheckingStage) ValidateResume(_ context.Context, _ *stage.Env, _ *manifest.Manifest) (stage.ResumeValidation, error) {
+	return s.validation, s.validateErr
+}
+
+func (s captureNotariusStage) Name() string           { return "extract" }
+func (s captureNotariusStage) Declares() stage.IODecl { return stage.IODecl{} }
+func (s captureNotariusStage) Run(_ context.Context, env *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
+	*s.captured = env.Notarius != nil
+	return &stage.StageResult{}, nil
+}
+
 func (s captureSelectedArtifactsStage) Name() string           { return s.name }
 func (s captureSelectedArtifactsStage) Declares() stage.IODecl { return stage.IODecl{} }
 func (s captureSelectedArtifactsStage) Run(_ context.Context, env *stage.Env, _ *manifest.Manifest) (*stage.StageResult, error) {
@@ -134,6 +183,28 @@ func TestExecuteStagesPropagatesSelectedArtifactsToEnv(t *testing.T) {
 	}
 }
 
+func TestExecuteStagesComposesNotariusOnlyForEnabledExtraction(t *testing.T) {
+	cfg := testConfig(t)
+	cfg.Pipeline.Notarius = &config.NotariusConfig{Enabled: true}
+	captured := false
+
+	_, err := executeStages(context.Background(), cfg, []stage.Stage{captureNotariusStage{captured: &captured}}, RunOptions{})
+	if err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	if !captured {
+		t.Fatal("extract stage did not receive the default Notarius runner")
+	}
+
+	if needsNotariusForRun(cfg, []stage.Stage{countingStage{name: "analyze", runs: new(int)}}) {
+		t.Fatal("Notarius runner requested without extract in the selected plan")
+	}
+	cfg.Pipeline.Notarius.Enabled = false
+	if needsNotariusForRun(cfg, []stage.Stage{captureNotariusStage{captured: new(bool)}}) {
+		t.Fatal("Notarius runner requested while extraction is disabled")
+	}
+}
+
 func TestExecuteStagesAnalyzeOutputsPersistAsScriptoriumArtifacts(t *testing.T) {
 	cfg := testConfig(t)
 	storeForPaths := artifacts.NewLocalStore(cfg.Pipeline.Workspace.Root)
@@ -199,6 +270,61 @@ func TestExecuteStagesAnalyzeOutputsPersistAsScriptoriumArtifacts(t *testing.T)
 	}
 }
 
+func TestMapResultOutputsPrefersExplicitSourceAndCopiesMetadata(t *testing.T) {
+	contract := &artifactmodel.ContractMetadata{
+		MediaType:     "application/json",
+		SchemaID:      "notarius.dnd.npc_registry",
+		SchemaVersion: "v1",
+	}
+	provenance := &artifactmodel.ExternalProvenance{
+		System:     "notarius",
+		RunID:      "external-run",
+		PipelineID: "dnd-session",
+		ArtifactID: "npc-registry",
+	}
+	result := &stage.StageResult{Outputs: []artifacts.Ref{{
+		Kind:               "structured_data",
+		SourceID:           "narratio.example.npcs",
+		RelativePath:       "artifacts/npcs.json",
+		Contract:           contract,
+		ExternalProvenance: provenance,
+	}}}
+
+	got := mapResultOutputs("analyze", result, "narratio-run")
+	if len(got) != 1 {
+		t.Fatalf("outputs len = %d, want 1", len(got))
+	}
+	if got[0].SourceID != "narratio.example.npcs" {
+		t.Fatalf("source_id = %q, want explicit source", got[0].SourceID)
+	}
+	if got[0].Kind != "structured_data" {
+		t.Fatalf("kind = %q, want explicit output kind preserved", got[0].Kind)
+	}
+	if got[0].Contract == nil || *got[0].Contract != *contract {
+		t.Fatalf("contract = %#v, want %#v", got[0].Contract, contract)
+	}
+	if got[0].ExternalProvenance == nil || *got[0].ExternalProvenance != *provenance {
+		t.Fatalf("external provenance = %#v, want %#v", got[0].ExternalProvenance, provenance)
+	}
+	if got[0].Contract == contract || got[0].ExternalProvenance == provenance {
+		t.Fatal("mapped metadata should not alias the stage result")
+	}
+}
+
+func TestMapResultOutputsRetainsFallbackInference(t *testing.T) {
+	transcript := mapResultOutputs("trim", &stage.StageResult{Outputs: []artifacts.Ref{{
+		Kind: artifactmodel.TranscriptOutputKindFinalTrimmed,
+	}}}, "run-id")
+	if len(transcript) != 1 || transcript[0].SourceID != artifacts.ArtifactTranscriptFinalTrimmed {
+		t.Fatalf("transcript fallback = %#v, want final-trimmed source", transcript)
+	}
+
+	analyze := mapResultOutputs("analyze", &stage.StageResult{Outputs: []artifacts.Ref{{Kind: "session_recap"}}}, "run-id")
+	if len(analyze) != 1 || analyze[0].SourceID != "narratio.artifact.session_recap" || analyze[0].Kind != "scriptorium_artifact" {
+		t.Fatalf("analyze fallback = %#v, want configured artifact inference", analyze)
+	}
+}
+
 func TestNeedsObjectStoreForRunPrepareWithPreviousRequirements(t *testing.T) {
 	tests := []struct {
 		name              string
@@ -273,7 +399,7 @@ func TestExecuteStagesPublishSkipsRequiredUnselectedConfiguredOutput(t *testing.
 	manifestPath := manifestPathFor(cfg)
 	seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
 	seed.Campaign = cfg.Session.Campaign
-	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render"} {
+	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render"} {
 		seed.MarkStageSucceeded(stageName, time.Now().UTC(), nil)
 	}
 	if err := os.MkdirAll(filepath.Dir(manifestPath), 0o755); err != nil {
@@ -332,8 +458,8 @@ func TestExecuteStagesPlaceholderSuccessUpdatesManifest(t *testing.T) {
 	if err != nil {
 		t.Fatalf("executeStages() error = %v", err)
 	}
-	if len(summary.StageNames) != 10 || len(summary.Executed) != 10 || len(summary.Skipped) != 0 {
-		t.Fatalf("summary = %#v, want all 10 executed", summary)
+	if len(summary.StageNames) != 11 || len(summary.Executed) != 11 || len(summary.Skipped) != 1 || summary.Skipped[0] != "extract" {
+		t.Fatalf("summary = %#v, want full plan with disabled extraction self-skip", summary)
 	}
 
 	store := &manifest.LocalStore{}
@@ -342,11 +468,17 @@ func TestExecuteStagesPlaceholderSuccessUpdatesManifest(t *testing.T) {
 		t.Fatalf("Load manifest error = %v", err)
 	}
 
-	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze", "publish", "notify"} {
+	for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze", "publish", "notify"} {
 		sr := m.Stages[name]
 		if sr == nil {
 			t.Fatalf("missing stage record %q", name)
 		}
+		if name == "extract" {
+			if sr.Status != manifest.StatusSkipped || sr.Error == nil || sr.Error.Message != "notarius_disabled" {
+				t.Fatalf("extract stage = %#v, want disabled skip", sr)
+			}
+			continue
+		}
 		if sr.Status != manifest.StatusSucceeded {
 			t.Fatalf("stage %q status = %q, want %q", name, sr.Status, manifest.StatusSucceeded)
 		}
@@ -497,6 +629,91 @@ func TestExecuteStagesSkipSucceededWhenNotForced(t *testing.T) {
 	}
 }
 
+func TestExecuteStagesUsesOptionalResumeValidation(t *testing.T) {
+	for _, test := range []struct {
+		name        string
+		validation  stage.ResumeValidation
+		wantRuns    int
+		wantSkipped int
+	}{
+		{name: "resumable", validation: stage.Resumable(), wantSkipped: 1},
+		{name: "rerun", validation: stage.NonResumable("durable output changed"), wantRuns: 1},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			cfg := testConfig(t)
+			store := &manifest.LocalStore{}
+			seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+			seed.MarkStageSucceeded("checked", time.Now().UTC(), nil)
+			if err := store.Save(context.Background(), manifestPathFor(cfg), seed); err != nil {
+				t.Fatalf("Save() error = %v", err)
+			}
+			runs := 0
+			candidate := resumeCheckingStage{name: "checked", validation: test.validation, runs: &runs}
+			summary, err := executeStages(context.Background(), cfg, []stage.Stage{candidate}, RunOptions{})
+			if err != nil {
+				t.Fatalf("executeStages() error = %v", err)
+			}
+			if runs != test.wantRuns || len(summary.Skipped) != test.wantSkipped {
+				t.Fatalf("runs = %d summary = %#v", runs, summary)
+			}
+		})
+	}
+}
+
+func TestExecuteStagesNonResumableResultRerunsSucceededDownstream(t *testing.T) {
+	cfg := testConfig(t)
+	store := &manifest.LocalStore{}
+	seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	seed.MarkStageSucceeded("extract", time.Now().UTC(), nil)
+	seed.MarkStageSucceeded("render", time.Now().UTC(), nil)
+	if err := store.Save(context.Background(), manifestPathFor(cfg), seed); err != nil {
+		t.Fatalf("Save() error = %v", err)
+	}
+	extractRuns, renderRuns := 0, 0
+	stages := []stage.Stage{
+		resumeCheckingStage{name: "extract", validation: stage.NonResumable("checksum changed"), runs: &extractRuns},
+		countingStage{name: "render", runs: &renderRuns},
+	}
+	summary, err := executeStages(context.Background(), cfg, stages, RunOptions{})
+	if err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	if extractRuns != 1 || renderRuns != 1 || len(summary.Executed) != 2 || len(summary.Skipped) != 0 {
+		t.Fatalf("extract runs=%d render runs=%d summary=%#v", extractRuns, renderRuns, summary)
+	}
+}
+
+func TestExecuteStagesResumeValidationErrorPreservesSucceededRecord(t *testing.T) {
+	cfg := testConfig(t)
+	store := &manifest.LocalStore{}
+	seed := manifest.New(cfg.Session.SessionID, time.Now().UTC())
+	seed.MarkStageSucceeded("checked", time.Now().UTC(), []manifest.ArtifactRecord{{Kind: "kept", LocalPath: "kept.json"}})
+	seed.Stages["checked"].Metadata = map[string]any{"kept": true}
+	if err := store.Save(context.Background(), manifestPathFor(cfg), seed); err != nil {
+		t.Fatalf("Save() error = %v", err)
+	}
+	before, err := json.Marshal(seed.Stages["checked"])
+	if err != nil {
+		t.Fatalf("Marshal(before) error = %v", err)
+	}
+	runs := 0
+	candidate := resumeCheckingStage{name: "checked", validateErr: errors.New("inspection unavailable"), runs: &runs}
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{candidate}, RunOptions{}); err == nil || !strings.Contains(err.Error(), "inspection unavailable") {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	loaded, err := store.Load(context.Background(), manifestPathFor(cfg))
+	if err != nil {
+		t.Fatalf("Load() error = %v", err)
+	}
+	after, err := json.Marshal(loaded.Stages["checked"])
+	if err != nil {
+		t.Fatalf("Marshal(after) error = %v", err)
+	}
+	if string(after) != string(before) || runs != 0 {
+		t.Fatalf("succeeded record changed: before=%s after=%s runs=%d", before, after, runs)
+	}
+}
+
 func TestExecuteStagesForceRerunsSucceeded(t *testing.T) {
 	cfg := testConfig(t)
 	manifestPath := manifestPathFor(cfg)
@@ -525,13 +742,48 @@ func TestExecuteStagesForceRerunsSucceeded(t *testing.T) {
 	}
 }
 
+func TestExecuteStagesSuccessfulReplacementDoesNotInheritResultDetails(t *testing.T) {
+	cfg := testConfig(t)
+	manifestPath := manifestPathFor(cfg)
+	store := &manifest.LocalStore{}
+
+	existing := manifest.New(cfg.Session.SessionID, time.Date(2026, 5, 3, 1, 0, 0, 0, time.UTC))
+	existing.MarkStageSucceeded("transcribe", time.Date(2026, 5, 3, 1, 1, 0, 0, time.UTC), []manifest.ArtifactRecord{{
+		Kind: "transcript_raw", LocalPath: "transcripts/old.json",
+	}})
+	existingStage := existing.Stages["transcribe"]
+	existingStage.Logs = []string{"logs/old.log"}
+	existingStage.GeneratedConfigs = []string{"generated/old.yaml"}
+	existingStage.Metadata = map[string]any{"old_result": true}
+	if err := store.Save(context.Background(), manifestPath, existing); err != nil {
+		t.Fatalf("Save manifest error = %v", err)
+	}
+
+	runs := 0
+	replacement := resultStage{name: "transcribe", result: &stage.StageResult{}, runs: &runs}
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{replacement}, RunOptions{Force: true}); err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	loaded, err := store.Load(context.Background(), manifestPath)
+	if err != nil {
+		t.Fatalf("Load manifest error = %v", err)
+	}
+	record := loaded.Stages["transcribe"]
+	if runs != 1 || record == nil || record.Status != manifest.StatusSucceeded {
+		t.Fatalf("runs = %d, record = %#v, want one successful replacement", runs, record)
+	}
+	if len(record.Outputs) != 0 || len(record.Logs) != 0 || len(record.GeneratedConfigs) != 0 || len(record.Metadata) != 0 {
+		t.Fatalf("replacement inherited result details: %#v", record)
+	}
+}
+
 func TestExecuteStagesForceSuccessInvalidatesDownstreamSucceededStages(t *testing.T) {
 	cfg := testConfig(t)
 	manifestPath := manifestPathFor(cfg)
 	store := &manifest.LocalStore{}
 
 	existing := manifest.New(cfg.Session.SessionID, time.Date(2026, 5, 3, 1, 0, 0, 0, time.UTC))
-	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "publish", "notify"} {
+	for _, stageName := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "publish", "notify"} {
 		existing.MarkStageSucceeded(stageName, time.Date(2026, 5, 3, 1, 1, 0, 0, time.UTC), nil)
 	}
 	existing.MarkStageFailed("analyze", time.Date(2026, 5, 3, 1, 1, 0, 0, time.UTC), "previous analyze failure")
@@ -562,7 +814,7 @@ func TestExecuteStagesForceSuccessInvalidatesDownstreamSucceededStages(t *testin
 	if loaded.Stages["polish"] == nil || loaded.Stages["polish"].Status != manifest.StatusSucceeded {
 		t.Fatalf("polish status = %#v, want succeeded", loaded.Stages["polish"])
 	}
-	for _, stageName := range []string{"normalize", "trim", "render", "publish", "notify"} {
+	for _, stageName := range []string{"normalize", "trim", "extract", "render", "publish", "notify"} {
 		if loaded.Stages[stageName] == nil || loaded.Stages[stageName].Status != manifest.StatusStale {
 			t.Fatalf("%s status = %#v, want stale", stageName, loaded.Stages[stageName])
 		}
@@ -738,6 +990,126 @@ func TestExecuteStagesRunManifestRecordsSkippedStage(t *testing.T) {
 	}
 }
 
+func TestExecuteStagesPersistsSelfSkipAndContinues(t *testing.T) {
+	cfg := testConfig(t)
+	store := &manifest.LocalStore{}
+	manifestPath := manifestPathFor(cfg)
+	seed := manifest.New(cfg.Session.SessionID, time.Date(2026, 5, 3, 1, 0, 0, 0, time.UTC))
+	seed.MarkStageSucceeded("optional", time.Date(2026, 5, 3, 1, 1, 0, 0, time.UTC), []manifest.ArtifactRecord{{
+		Kind:      "old_output",
+		SourceID:  "narratio.example.old",
+		LocalPath: "artifacts/old.json",
+	}})
+	seed.Stages["optional"].Logs = []string{"old.log"}
+	seed.Stages["optional"].GeneratedConfigs = []string{"old.yml"}
+	seed.Stages["optional"].Metadata = map[string]any{"old": true}
+	if err := os.MkdirAll(filepath.Dir(manifestPath), 0o755); err != nil {
+		t.Fatalf("MkdirAll() error = %v", err)
+	}
+	if err := store.Save(context.Background(), manifestPath, seed); err != nil {
+		t.Fatalf("Save() seed manifest error = %v", err)
+	}
+
+	order := []string{}
+	optionalRuns := 0
+	stages := []stage.Stage{
+		resultStage{
+			name:  "optional",
+			runs:  &optionalRuns,
+			order: &order,
+			result: &stage.StageResult{
+				Disposition:      stage.StageDispositionSkipped,
+				SkipReason:       "integration_disabled",
+				Logs:             []string{"runs/current/optional.log"},
+				GeneratedConfigs: []string{"runs/current/optional.yml"},
+				Metadata:         map[string]any{"enabled": false},
+			},
+		},
+		resultStage{name: "later", order: &order, result: &stage.StageResult{}},
+	}
+	summary, err := executeStages(context.Background(), cfg, stages, RunOptions{Force: true})
+	if err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	if strings.Join(order, ",") != "optional,later" {
+		t.Fatalf("execution order = %v, want optional then later", order)
+	}
+	if len(summary.Executed) != 2 || len(summary.Skipped) != 1 || summary.Skipped[0] != "optional" {
+		t.Fatalf("summary = %#v, want optional executed and self-skipped before later", summary)
+	}
+
+	sessionManifest, err := store.Load(context.Background(), summary.ManifestPath)
+	if err != nil {
+		t.Fatalf("Load() session manifest error = %v", err)
+	}
+	selfSkipped := sessionManifest.Stages["optional"]
+	if selfSkipped == nil || selfSkipped.Status != manifest.StatusSkipped {
+		t.Fatalf("optional stage = %#v, want skipped", selfSkipped)
+	}
+	if len(selfSkipped.Outputs) != 0 {
+		t.Fatalf("optional outputs = %#v, want old outputs cleared", selfSkipped.Outputs)
+	}
+	if selfSkipped.Error == nil || selfSkipped.Error.Message != "integration_disabled" {
+		t.Fatalf("optional skip reason = %#v, want integration_disabled", selfSkipped.Error)
+	}
+	if len(selfSkipped.Logs) != 1 || selfSkipped.Logs[0] != "runs/current/optional.log" ||
+		len(selfSkipped.GeneratedConfigs) != 1 || selfSkipped.GeneratedConfigs[0] != "runs/current/optional.yml" ||
+		selfSkipped.Metadata["enabled"] != false || selfSkipped.Metadata["old"] != nil {
+		t.Fatalf("optional result details = %#v, want current bounded diagnostics and metadata", selfSkipped)
+	}
+	if later := sessionManifest.Stages["later"]; later == nil || later.Status != manifest.StatusSucceeded {
+		t.Fatalf("later stage = %#v, want succeeded", later)
+	}
+
+	runManifest, err := store.LoadRun(context.Background(), summary.RunManifestPath)
+	if err != nil {
+		t.Fatalf("LoadRun() error = %v", err)
+	}
+	runStage := runManifest.Stages["optional"]
+	if runStage == nil || runStage.Action != manifest.RunStageActionRun || runStage.Status != manifest.StatusSkipped {
+		t.Fatalf("run optional stage = %#v, want run action with skipped status", runStage)
+	}
+	if len(runStage.Logs) != 1 || runStage.Metadata["enabled"] != false {
+		t.Fatalf("run optional stage details = %#v, want result diagnostics and metadata", runStage)
+	}
+
+	_, err = executeStages(context.Background(), cfg, []stage.Stage{stages[0]}, RunOptions{})
+	if err != nil {
+		t.Fatalf("second executeStages() error = %v", err)
+	}
+	if optionalRuns != 2 {
+		t.Fatalf("optional runs = %d, want self-skipped stage reconsidered", optionalRuns)
+	}
+}
+
+func TestExecuteStagesRejectsSkippedResultWithOutputs(t *testing.T) {
+	cfg := testConfig(t)
+	invalid := resultStage{name: "optional", result: &stage.StageResult{
+		Disposition: stage.StageDispositionSkipped,
+		SkipReason:  "integration_disabled",
+		Outputs:     []artifacts.Ref{{Kind: "unexpected"}},
+	}}
+
+	summary, err := executeStages(context.Background(), cfg, []stage.Stage{invalid}, RunOptions{})
+	if err == nil {
+		t.Fatal("executeStages() error = nil, want invalid skipped result failure")
+	}
+	if summary != nil {
+		t.Fatalf("summary = %#v, want nil", summary)
+	}
+	if !strings.Contains(err.Error(), "skipped result contains 1 output") {
+		t.Fatalf("error = %q, want skipped-output validation", err)
+	}
+
+	loaded, loadErr := (&manifest.LocalStore{}).Load(context.Background(), manifestPathFor(cfg))
+	if loadErr != nil {
+		t.Fatalf("Load() session manifest error = %v", loadErr)
+	}
+	if got := loaded.Stages["optional"]; got == nil || got.Status != manifest.StatusFailed {
+		t.Fatalf("optional stage = %#v, want failed", got)
+	}
+}
+
 func TestExecuteStagesRunLocalArtifactsAndCanonicalSync(t *testing.T) {
 	cfg := testConfig(t)
 	stages := []stage.Stage{
@@ -986,7 +1358,7 @@ func TestAdapterBackedStageFailureMarksManifestFailed(t *testing.T) {
 				seed.S3Bucket = "my-dnd-archive"
 				seed.S3SessionPrefix = "dnd/campaigns/" + cfg.Session.Campaign + "/sessions/" + cfg.Session.SessionID + "/"
 				seed.S3RunPrefix = seed.S3SessionPrefix + "runs/" + runID + "/"
-				for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "render", "analyze"} {
+				for _, name := range []string{"prepare", "transcribe", "merge", "polish", "normalize", "trim", "extract", "render", "analyze"} {
 					seed.MarkStageSucceeded(name, time.Now().UTC(), nil)
 				}
 				if err := tc.env.ManifestStore.Save(context.Background(), manifestPathFor(cfg), seed); err != nil {

internal/app/session_oriented_cli_test.go

@@ -346,7 +346,7 @@ func TestExecuteSessionLocksMutationAcceptsPositionalSessionID(t *testing.T) {
 	if code != 0 {
 		t.Fatalf("locks remove exit code = %d, want 0; stderr=%q", code, stderr.String())
 	}
-	store, err := config.LoadPublishLockStoreBytes("locks.yml", fake.Objects[key].Data, nil)
+	store, err := config.LoadPublishLockStoreBytes("locks.yml", fake.Objects[key].Data, nil, nil)
 	if err != nil {
 		t.Fatalf("LoadPublishLockStoreBytes() error = %v", err)
 	}

internal/artifactmodel/metadata.go

@@ -0,0 +1,17 @@
+package artifactmodel
+
+// ContractMetadata identifies the data contract implemented by an artifact.
+type ContractMetadata struct {
+	MediaType     string `json:"media_type"`
+	SchemaID      string `json:"schema_id"`
+	SchemaVersion string `json:"schema_version"`
+	ModuleKey     string `json:"module_key,omitempty"`
+}
+
+// ExternalProvenance identifies an artifact produced by an external system.
+type ExternalProvenance struct {
+	System     string `json:"system"`
+	RunID      string `json:"run_id"`
+	PipelineID string `json:"pipeline_id"`
+	ArtifactID string `json:"artifact_id"`
+}

internal/artifactmodel/metadata_test.go

@@ -0,0 +1,56 @@
+package artifactmodel
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func TestArtifactMetadataJSON(t *testing.T) {
+	type envelope struct {
+		Contract           *ContractMetadata   `json:"contract,omitempty"`
+		ExternalProvenance *ExternalProvenance `json:"external_provenance,omitempty"`
+	}
+
+	complete, err := json.Marshal(envelope{
+		Contract: &ContractMetadata{
+			MediaType:     "application/json",
+			SchemaID:      "notarius.dnd.npc_registry",
+			SchemaVersion: "v1",
+			ModuleKey:     "dnd/npc-registry",
+		},
+		ExternalProvenance: &ExternalProvenance{
+			System:     "notarius",
+			RunID:      "run-123",
+			PipelineID: "dnd-session",
+			ArtifactID: "npc-registry",
+		},
+	})
+	if err != nil {
+		t.Fatalf("Marshal() complete metadata error = %v", err)
+	}
+	wantComplete := `{"contract":{"media_type":"application/json","schema_id":"notarius.dnd.npc_registry","schema_version":"v1","module_key":"dnd/npc-registry"},"external_provenance":{"system":"notarius","run_id":"run-123","pipeline_id":"dnd-session","artifact_id":"npc-registry"}}`
+	if string(complete) != wantComplete {
+		t.Fatalf("complete metadata JSON = %s, want %s", complete, wantComplete)
+	}
+
+	omitted, err := json.Marshal(envelope{})
+	if err != nil {
+		t.Fatalf("Marshal() omitted metadata error = %v", err)
+	}
+	if string(omitted) != `{}` {
+		t.Fatalf("omitted metadata JSON = %s, want {}", omitted)
+	}
+
+	withoutModule, err := json.Marshal(envelope{Contract: &ContractMetadata{
+		MediaType:     "application/json",
+		SchemaID:      "notarius.dnd.npc_registry",
+		SchemaVersion: "v1",
+	}})
+	if err != nil {
+		t.Fatalf("Marshal() contract without module key error = %v", err)
+	}
+	if strings.Contains(string(withoutModule), "module_key") {
+		t.Fatalf("contract JSON unexpectedly contains omitted module_key: %s", withoutModule)
+	}
+}

internal/artifactpolicy/policy.go

@@ -18,11 +18,14 @@ const (
 	SourceInputGlossary = "narratio.input.glossary"
 
 	configuredSourcePrefix      = "narratio.artifact."
+	extractionSourcePrefix      = "narratio.extraction."
 	previousConfiguredSrcPrefix = "narratio.previous_session.artifact."
 )
 
 var configuredSourceRE = regexp.MustCompile(`^narratio\.artifact\.([a-z][a-z0-9_]*)$`)
+var extractionSourceRE = regexp.MustCompile(`^narratio\.extraction\.([a-z][a-z0-9_]*)$`)
 var previousSourceRE = regexp.MustCompile(`^narratio\.previous_session\.artifact\.([a-z][a-z0-9_]*)$`)
+var configuredKeyRE = regexp.MustCompile(`^[a-z][a-z0-9_]*$`)
 
 var (
 	ErrUnsupportedScriptoriumInputSource = errors.New("unsupported scriptorium input source")
@@ -34,6 +37,7 @@ type SourceKind string
 const (
 	SourceKindBuiltIn            SourceKind = "built_in"
 	SourceKindConfiguredArtifact SourceKind = "configured_artifact"
+	SourceKindExtraction         SourceKind = "extraction"
 	SourceKindPreviousArtifact   SourceKind = "previous_session_configured_artifact"
 	SourceKindStableInput        SourceKind = "stable_input"
 )
@@ -67,11 +71,30 @@ func (e *UnknownConfiguredArtifactError) Error() string {
 	return fmt.Sprintf("references unknown artifact %q", e.ConfiguredKey)
 }
 
+// UnknownExtractionArtifactError reports a source that references an undefined extraction key.
+type UnknownExtractionArtifactError struct {
+	ConfiguredKey string
+}
+
+func (e *UnknownExtractionArtifactError) Error() string {
+	return fmt.Sprintf("references unknown extraction output %q", e.ConfiguredKey)
+}
+
+// IsConfiguredKey reports whether a key follows the configured-artifact key grammar.
+func IsConfiguredKey(key string) bool {
+	return configuredKeyRE.MatchString(strings.TrimSpace(key))
+}
+
 // ConfiguredSourceID converts a configured artifact key into source id form.
 func ConfiguredSourceID(key string) string {
 	return configuredSourcePrefix + strings.TrimSpace(key)
 }
 
+// ExtractionSourceID converts an extraction output key into source id form.
+func ExtractionSourceID(key string) string {
+	return extractionSourcePrefix + strings.TrimSpace(key)
+}
+
 // PreviousSessionSourceID converts a configured artifact key into previous-session source id form.
 func PreviousSessionSourceID(key string) string {
 	return previousConfiguredSrcPrefix + strings.TrimSpace(key)
@@ -86,6 +109,15 @@ func ParseConfiguredSource(source string) (string, bool) {
 	return matches[1], true
 }
 
+// ParseExtractionSource extracts configured key from narratio.extraction.<key>.
+func ParseExtractionSource(source string) (string, bool) {
+	matches := extractionSourceRE.FindStringSubmatch(strings.TrimSpace(source))
+	if len(matches) != 2 {
+		return "", false
+	}
+	return matches[1], true
+}
+
 // ParsePreviousSessionSource extracts configured key from narratio.previous_session.artifact.<key>.
 func ParsePreviousSessionSource(source string) (string, bool) {
 	matches := previousSourceRE.FindStringSubmatch(strings.TrimSpace(source))
@@ -110,6 +142,9 @@ func ClassifySource(source string) (Source, error) {
 	if key, ok := ParseConfiguredSource(trimmed); ok {
 		return Source{ID: trimmed, Kind: SourceKindConfiguredArtifact, ConfiguredKey: key}, nil
 	}
+	if key, ok := ParseExtractionSource(trimmed); ok {
+		return Source{ID: trimmed, Kind: SourceKindExtraction, ConfiguredKey: key}, nil
+	}
 	if key, ok := ParsePreviousSessionSource(trimmed); ok {
 		return Source{ID: trimmed, Kind: SourceKindPreviousArtifact, ConfiguredKey: key}, nil
 	}
@@ -186,24 +221,48 @@ func PreviousSessionSourceDescriptorForConfiguredKey(configuredKey string) (Prev
 func ValidateInputConfiguredReference(
 	descriptor ScriptoriumInputSourceDescriptor,
 	configured map[string]struct{},
+) error {
+	return ValidateInputReference(descriptor, configured, nil)
+}
+
+// ValidateInputReference checks that configured and extraction sources are declared
+// by the effective pipeline configuration.
+func ValidateInputReference(
+	descriptor ScriptoriumInputSourceDescriptor,
+	configured map[string]struct{},
+	extractions map[string]struct{},
 ) error {
 	switch descriptor.Source.Kind {
 	case SourceKindConfiguredArtifact, SourceKindPreviousArtifact:
 		if _, ok := configured[descriptor.Source.ConfiguredKey]; !ok {
 			return &UnknownConfiguredArtifactError{ConfiguredKey: descriptor.Source.ConfiguredKey}
 		}
+	case SourceKindExtraction:
+		if _, ok := extractions[descriptor.Source.ConfiguredKey]; !ok {
+			return &UnknownExtractionArtifactError{ConfiguredKey: descriptor.Source.ConfiguredKey}
+		}
 	}
 	return nil
 }
 
 // ValidatePublishSource validates that a source is publish-compatible and references a known configured artifact.
 func ValidatePublishSource(source string, configured map[string]string) (Source, error) {
+	return ValidatePublishSourceWithExtractions(source, configured, nil)
+}
+
+// ValidatePublishSourceWithExtractions validates publish sources against the
+// configured Scriptorium artifacts and extraction outputs.
+func ValidatePublishSourceWithExtractions(
+	source string,
+	configured map[string]string,
+	extractions map[string]struct{},
+) (Source, error) {
 	classified, err := ClassifySource(source)
 	if err != nil {
-		return Source{}, fmt.Errorf("must be a built-in source id or narratio.artifact.<name>")
+		return Source{}, fmt.Errorf("must be a built-in source id, narratio.artifact.<name>, or configured narratio.extraction.<name>")
 	}
 	if classified.Kind == SourceKindPreviousArtifact {
-		return Source{}, fmt.Errorf("must be a built-in source id or narratio.artifact.<name>")
+		return Source{}, fmt.Errorf("must be a built-in source id, narratio.artifact.<name>, or configured narratio.extraction.<name>")
 	}
 	if classified.Kind == SourceKindConfiguredArtifact {
 		if configured == nil {
@@ -213,6 +272,11 @@ func ValidatePublishSource(source string, configured map[string]string) (Source,
 			return Source{}, fmt.Errorf("configured artifact %q is not defined in pipeline.scriptorium.artifacts", classified.ConfiguredKey)
 		}
 	}
+	if classified.Kind == SourceKindExtraction {
+		if _, ok := extractions[classified.ConfiguredKey]; !ok {
+			return Source{}, fmt.Errorf("extraction output %q is not defined in pipeline.notarius.outputs", classified.ConfiguredKey)
+		}
+	}
 	return classified, nil
 }
 
@@ -247,7 +311,17 @@ func DeriveDefaultPublishedDestination(source Source, configured map[string]stri
 // ResolvePublishedDestination validates and normalizes an explicit destination,
 // or derives one when omitted.
 func ResolvePublishedDestination(sourceID, explicitDest string, configured map[string]string) (string, error) {
-	source, err := ValidatePublishSource(sourceID, configured)
+	return ResolvePublishedDestinationWithExtractions(sourceID, explicitDest, configured, nil)
+}
+
+// ResolvePublishedDestinationWithExtractions validates and normalizes a destination
+// while accepting extraction sources declared by the effective Notarius configuration.
+func ResolvePublishedDestinationWithExtractions(
+	sourceID, explicitDest string,
+	configured map[string]string,
+	extractions map[string]struct{},
+) (string, error) {
+	source, err := ValidatePublishSourceWithExtractions(sourceID, configured, extractions)
 	if err != nil {
 		return "", err
 	}

internal/artifactpolicy/policy_test.go

@@ -17,6 +17,7 @@ func TestClassifySource(t *testing.T) {
 		{name: "built in transcript", source: "narratio.transcript.final_trimmed", wantKind: SourceKindBuiltIn},
 		{name: "built in bounds", source: "narratio.bounds.session", wantKind: SourceKindBuiltIn},
 		{name: "configured artifact", source: "narratio.artifact.session_recap", wantKind: SourceKindConfiguredArtifact, wantKey: "session_recap"},
+		{name: "extraction", source: "narratio.extraction.npc_registry", wantKind: SourceKindExtraction, wantKey: "npc_registry"},
 		{name: "previous session configured", source: "narratio.previous_session.artifact.session_recap", wantKind: SourceKindPreviousArtifact, wantKey: "session_recap"},
 		{name: "unsupported", source: "narratio.unknown", wantErrLike: "unsupported artifact source"},
 	}
@@ -56,6 +57,60 @@ func TestValidatePublishSource(t *testing.T) {
 	}
 }
 
+func TestExtractionSourcePolicy(t *testing.T) {
+	if got := ExtractionSourceID(" npc_registry "); got != "narratio.extraction.npc_registry" {
+		t.Fatalf("ExtractionSourceID() = %q, want narratio.extraction.npc_registry", got)
+	}
+	if key, ok := ParseExtractionSource(" narratio.extraction.npc_registry "); !ok || key != "npc_registry" {
+		t.Fatalf("ParseExtractionSource() = %q, %t; want npc_registry, true", key, ok)
+	}
+	for _, source := range []string{
+		"narratio.extraction.",
+		"narratio.extraction.NPC",
+		"narratio.extraction.npc-registry",
+		"narratio.extraction.npc_registry.extra",
+	} {
+		if _, ok := ParseExtractionSource(source); ok {
+			t.Fatalf("ParseExtractionSource(%q) unexpectedly matched", source)
+		}
+	}
+
+	descriptor, err := DescribeScriptoriumInputSource("narratio.extraction.npc_registry")
+	if err != nil {
+		t.Fatalf("DescribeScriptoriumInputSource(extraction) error = %v", err)
+	}
+	if descriptor.Source.Kind != SourceKindExtraction || descriptor.Source.ConfiguredKey != "npc_registry" {
+		t.Fatalf("extraction descriptor = %#v", descriptor)
+	}
+	declared := map[string]struct{}{"npc_registry": {}}
+	if err := ValidateInputReference(descriptor, nil, declared); err != nil {
+		t.Fatalf("ValidateInputReference(declared extraction) error = %v", err)
+	}
+	if err := ValidateInputReference(descriptor, nil, nil); err == nil {
+		t.Fatal("ValidateInputReference(unknown extraction) error = nil, want error")
+	}
+
+	if _, err := ValidatePublishSourceWithExtractions("narratio.extraction.npc_registry", nil, declared); err != nil {
+		t.Fatalf("ValidatePublishSourceWithExtractions(declared) error = %v", err)
+	}
+	if _, err := ValidatePublishSourceWithExtractions("narratio.extraction.unknown", nil, declared); err == nil {
+		t.Fatal("ValidatePublishSourceWithExtractions(unknown) error = nil, want error")
+	}
+
+	identities := map[string]struct{}{}
+	for _, sourceID := range []string{
+		ExtractionSourceID("npc_registry"),
+		ConfiguredSourceID("npc_registry"),
+		PreviousSessionSourceID("npc_registry"),
+		SourceBoundsSession,
+	} {
+		if _, exists := identities[sourceID]; exists {
+			t.Fatalf("source identity collision at %q", sourceID)
+		}
+		identities[sourceID] = struct{}{}
+	}
+}
+
 func TestResolvePublishedDestination(t *testing.T) {
 	configured := map[string]string{"session_recap": "artifacts/session_recap.md"}
 

internal/artifacts/artifact_resolver.go

@@ -144,6 +144,22 @@ func ConfiguredArtifactName(source string) (string, bool) {
 	return artifactpolicy.ParseConfiguredSource(source)
 }
 
+// ExtractionArtifactSourceID returns narratio.extraction.<name> for a configured key.
+func ExtractionArtifactSourceID(key string) string {
+	return artifactpolicy.ExtractionSourceID(key)
+}
+
+// IsExtractionArtifactSource reports whether source is narratio.extraction.<name>.
+func IsExtractionArtifactSource(source string) bool {
+	_, ok := artifactpolicy.ParseExtractionSource(source)
+	return ok
+}
+
+// ExtractionArtifactName extracts <name> from narratio.extraction.<name>.
+func ExtractionArtifactName(source string) (string, bool) {
+	return artifactpolicy.ParseExtractionSource(source)
+}
+
 // IsPreviousSessionArtifactSource returns true when source is narratio.previous_session.artifact.<name>.
 func IsPreviousSessionArtifactSource(source string) bool {
 	_, ok := artifactpolicy.ParsePreviousSessionSource(source)
@@ -204,17 +220,19 @@ func ResolveSessionArtifact(paths SessionPaths, m *manifest.Manifest, source str
 }
 
 // ResolveSessionArtifactWithCatalog resolves built-in sources using existing rules and resolves
-// configured narratio.artifact.<name> sources through runtime catalog availability.
+// configured artifact and extraction sources through runtime catalog availability.
 func ResolveSessionArtifactWithCatalog(paths SessionPaths, m *manifest.Manifest, source string, catalog *ArtifactCatalog) (ResolvedSessionArtifact, error) {
 	normalized := strings.TrimSpace(source)
 	if IsPreviousSessionArtifactSource(normalized) {
 		return ResolvePreviousSessionArtifactWithCatalog(paths, m, normalized, catalog)
 	}
-	if !IsConfiguredArtifactSource(normalized) {
+	configuredSource := IsConfiguredArtifactSource(normalized)
+	extractionSource := IsExtractionArtifactSource(normalized)
+	if !configuredSource && !extractionSource {
 		return ResolveSessionArtifact(paths, m, normalized)
 	}
 	if catalog == nil {
-		return ResolvedSessionArtifact{}, fmt.Errorf("configured artifact source %q requires runtime artifact catalog", source)
+		return ResolvedSessionArtifact{}, fmt.Errorf("catalog-backed artifact source %q requires runtime artifact catalog", source)
 	}
 	entry, ok := catalog.Lookup(normalized)
 	if !ok {
@@ -223,7 +241,11 @@ func ResolveSessionArtifactWithCatalog(paths SessionPaths, m *manifest.Manifest,
 	if !entry.Available {
 		return ResolvedSessionArtifact{}, &SessionArtifactNotFoundError{ArtifactID: normalized}
 	}
-	if err := validateResolvedContent(entry.Path, contentText); err != nil {
+	contentKind := contentText
+	if extractionSource {
+		contentKind = contentJSON
+	}
+	if err := validateResolvedContent(entry.Path, contentKind); err != nil {
 		return ResolvedSessionArtifact{}, fmt.Errorf("validate %q: %w", normalized, err)
 	}
 	return ResolvedSessionArtifact{
@@ -231,6 +253,7 @@ func ResolveSessionArtifactWithCatalog(paths SessionPaths, m *manifest.Manifest,
 		Path:          filepath.Clean(entry.Path),
 		ProducerStage: entry.ProducerStage,
 		OutputKind:    entry.OutputKind,
+		ProducerRunID: entry.ProducerRunID,
 		Provenance:    entry.Provenance,
 	}, nil
 }

internal/artifacts/artifact_resolver_test.go

@@ -150,6 +150,23 @@ func TestPreviousSessionArtifactSourceHelpers(t *testing.T) {
 	}
 }
 
+func TestExtractionArtifactSourceHelpers(t *testing.T) {
+	if got := ExtractionArtifactSourceID("npc_registry"); got != "narratio.extraction.npc_registry" {
+		t.Fatalf("ExtractionArtifactSourceID() = %q", got)
+	}
+	if !IsExtractionArtifactSource(" narratio.extraction.npc_registry ") {
+		t.Fatal("IsExtractionArtifactSource(valid) = false")
+	}
+	if key, ok := ExtractionArtifactName("narratio.extraction.npc_registry"); !ok || key != "npc_registry" {
+		t.Fatalf("ExtractionArtifactName() = %q, %t", key, ok)
+	}
+	for _, source := range []string{"narratio.extraction.", "narratio.extraction.npc-registry", "narratio.artifact.npc_registry"} {
+		if IsExtractionArtifactSource(source) {
+			t.Fatalf("IsExtractionArtifactSource(%q) = true, want false", source)
+		}
+	}
+}
+
 func TestResolveSessionArtifactPrefersManifestOutput(t *testing.T) {
 	workspace := t.TempDir()
 	paths := buildSessionPaths(workspace, "campaign", "session")

internal/artifacts/catalog.go

@@ -6,11 +6,13 @@ import (
 	"strings"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactpolicy"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
 const (
 	ArtifactProvenanceGeneratedCurrentAnalyzeRun = "generated.current_analyze_run"
 	ArtifactProvenanceDisabledFromDisk           = "filesystem.disabled_artifact_output"
+	ArtifactProvenanceCurrentExtractManifest     = "manifest.current_extract_run"
 )
 
 // ConfiguredArtifactDefinition describes one configured analyze artifact.
@@ -19,10 +21,40 @@ type ConfiguredArtifactDefinition struct {
 	OutputPath string
 }
 
+// ExtractionArtifactDefinition describes one configured Notarius output lane.
+type ExtractionArtifactDefinition struct {
+	LaneID        string
+	PipelineID    string
+	MediaType     string
+	SchemaID      string
+	SchemaVersion string
+	ModuleKey     string
+}
+
+// ExtractionDefinitionsFromConfig converts the effective Notarius output map into catalog definitions.
+func ExtractionDefinitionsFromConfig(cfg *config.NotariusConfig) map[string]ExtractionArtifactDefinition {
+	if cfg == nil || len(cfg.Outputs) == 0 {
+		return nil
+	}
+	definitions := make(map[string]ExtractionArtifactDefinition, len(cfg.Outputs))
+	for key, output := range cfg.Outputs {
+		definitions[key] = ExtractionArtifactDefinition{
+			LaneID:        output.LaneID,
+			PipelineID:    cfg.PipelineID,
+			MediaType:     output.MediaType,
+			SchemaID:      output.SchemaID,
+			SchemaVersion: output.SchemaVersion,
+			ModuleKey:     output.ModuleKey,
+		}
+	}
+	return definitions
+}
+
 // CatalogEntry is one runtime catalog entry resolved by source ID.
 type CatalogEntry struct {
 	SourceID         string
 	ConfiguredKey    string
+	ExtractionKey    string
 	CanonicalRelPath string
 	ProducerStage    string
 	OutputKind       string
@@ -31,12 +63,14 @@ type CatalogEntry struct {
 	Available        bool
 	Path             string
 	Provenance       string
+	ProducerRunID    string
 }
 
-// ArtifactCatalog tracks built-in and configured artifact definitions and runtime state.
+// ArtifactCatalog tracks built-in, configured, and extraction artifact definitions and runtime state.
 type ArtifactCatalog struct {
 	entries         map[string]CatalogEntry
 	configuredIndex map[string]string
+	extractionIndex map[string]string
 }
 
 // NewArtifactCatalog returns an empty runtime artifact catalog.
@@ -44,9 +78,41 @@ func NewArtifactCatalog() *ArtifactCatalog {
 	return &ArtifactCatalog{
 		entries:         map[string]CatalogEntry{},
 		configuredIndex: map[string]string{},
+		extractionIndex: map[string]string{},
 	}
 }
 
+// RegisterExtractionArtifacts registers the configured Notarius output lanes.
+func (c *ArtifactCatalog) RegisterExtractionArtifacts(configured map[string]ExtractionArtifactDefinition) error {
+	keys := make([]string, 0, len(configured))
+	for key := range configured {
+		keys = append(keys, key)
+	}
+	sort.Strings(keys)
+
+	for _, key := range keys {
+		trimmed := strings.TrimSpace(key)
+		if trimmed == "" {
+			return fmt.Errorf("extraction artifact keys must be non-empty")
+		}
+		if _, exists := c.extractionIndex[trimmed]; exists {
+			return fmt.Errorf("duplicate extraction artifact key %q", trimmed)
+		}
+		sourceID := ExtractionArtifactSourceID(trimmed)
+		if err := c.addEntry(CatalogEntry{
+			SourceID:      sourceID,
+			ExtractionKey: trimmed,
+			ProducerStage: "extract",
+			OutputKind:    "notarius_lane",
+			Planned:       true,
+		}); err != nil {
+			return fmt.Errorf("register extraction artifact %q: %w", trimmed, err)
+		}
+		c.extractionIndex[trimmed] = sourceID
+	}
+	return nil
+}
+
 // ConfiguredArtifactSourceID converts a configured artifact key into canonical source ID.
 func ConfiguredArtifactSourceID(key string) string {
 	return artifactpolicy.ConfiguredSourceID(key)
@@ -151,6 +217,15 @@ func (c *ArtifactCatalog) SourceIDForConfiguredKey(key string) (string, bool) {
 	return sourceID, ok
 }
 
+// SourceIDForExtractionKey returns the canonical source ID for one Notarius output key.
+func (c *ArtifactCatalog) SourceIDForExtractionKey(key string) (string, bool) {
+	if c == nil {
+		return "", false
+	}
+	sourceID, ok := c.extractionIndex[strings.TrimSpace(key)]
+	return sourceID, ok
+}
+
 // ListConfigured returns configured entries sorted by configured key.
 func (c *ArtifactCatalog) ListConfigured() []CatalogEntry {
 	if c == nil || len(c.configuredIndex) == 0 {
@@ -169,6 +244,23 @@ func (c *ArtifactCatalog) ListConfigured() []CatalogEntry {
 	return out
 }
 
+// ListExtraction returns extraction entries sorted by configured output key.
+func (c *ArtifactCatalog) ListExtraction() []CatalogEntry {
+	if c == nil || len(c.extractionIndex) == 0 {
+		return nil
+	}
+	keys := make([]string, 0, len(c.extractionIndex))
+	for key := range c.extractionIndex {
+		keys = append(keys, key)
+	}
+	sort.Strings(keys)
+	out := make([]CatalogEntry, 0, len(keys))
+	for _, key := range keys {
+		out = append(out, c.entries[c.extractionIndex[key]])
+	}
+	return out
+}
+
 // MarkAvailableGenerated marks one source as available in current analyze execution.
 func (c *ArtifactCatalog) MarkAvailableGenerated(sourceID, path string) error {
 	return c.markAvailable(sourceID, path, ArtifactProvenanceGeneratedCurrentAnalyzeRun)
@@ -179,6 +271,16 @@ func (c *ArtifactCatalog) MarkAvailableFromDisk(sourceID, path string) error {
 	return c.markAvailable(sourceID, path, ArtifactProvenanceDisabledFromDisk)
 }
 
+func (c *ArtifactCatalog) markAvailableFromExtractManifest(sourceID, path, producerRunID string) error {
+	if err := c.markAvailable(sourceID, path, ArtifactProvenanceCurrentExtractManifest); err != nil {
+		return err
+	}
+	entry := c.entries[strings.TrimSpace(sourceID)]
+	entry.ProducerRunID = strings.TrimSpace(producerRunID)
+	c.entries[entry.SourceID] = entry
+	return nil
+}
+
 func (c *ArtifactCatalog) markAvailable(sourceID, path, provenance string) error {
 	if c == nil {
 		return fmt.Errorf("artifact catalog is nil")

internal/artifacts/extraction_catalog.go

@@ -0,0 +1,180 @@
+package artifacts
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+)
+
+const (
+	extractStageName       = "extract"
+	extractionLaneKind     = "notarius_lane"
+	extractionIndexKind    = "notarius_index"
+	extractionMetadataRun  = "narratio_run_id"
+	extractionMetadataRoot = "bundle_root"
+)
+
+type hydratedExtraction struct {
+	sourceID string
+	path     string
+}
+
+// HydrateExtractionArtifacts marks extraction sources available only when the current
+// manifest contains one complete, internally consistent, succeeded extraction bundle.
+// Invalid, stale, incomplete, or unsafe records leave every extraction source unavailable.
+func (c *ArtifactCatalog) HydrateExtractionArtifacts(
+	paths SessionPaths,
+	m *manifest.Manifest,
+	configured map[string]ExtractionArtifactDefinition,
+) {
+	if c == nil || m == nil || len(configured) == 0 {
+		return
+	}
+	record := m.Stages[extractStageName]
+	if record == nil || record.Name != extractStageName || record.Status != manifest.StatusSucceeded {
+		return
+	}
+	producerRunID := extractionMetadataString(record.Metadata, extractionMetadataRun)
+	if !safeExtractionPathSegment(producerRunID) {
+		return
+	}
+	bundleRoot := filepath.Clean(filepath.Join(paths.ArtifactsDir, "notarius", producerRunID))
+	if !filepath.IsAbs(bundleRoot) || extractionMetadataString(record.Metadata, extractionMetadataRoot) != bundleRoot {
+		return
+	}
+	if !safeExistingExtractionDirectory(paths.Root, bundleRoot) {
+		return
+	}
+	receiptRunID, receiptPipelineID := extractionReceiptIdentity(record.Metadata)
+	if receiptRunID == "" || receiptPipelineID == "" {
+		return
+	}
+
+	expected := make(map[string]ExtractionArtifactDefinition, len(configured))
+	for key, definition := range configured {
+		sourceID, ok := c.SourceIDForExtractionKey(key)
+		if !ok {
+			return
+		}
+		expected[sourceID] = definition
+	}
+
+	seen := make(map[string]struct{}, len(expected))
+	hydrated := make([]hydratedExtraction, 0, len(expected))
+	indexSeen := false
+	for _, output := range record.Outputs {
+		if strings.TrimSpace(output.ProducerRunID) != producerRunID {
+			return
+		}
+		if output.SourceID == "" {
+			if indexSeen || output.Kind != extractionIndexKind || filepath.Clean(output.LocalPath) != filepath.Join(bundleRoot, "index.json") ||
+				!validExtractionPayload(bundleRoot, output.LocalPath, output.Checksum) {
+				return
+			}
+			indexSeen = true
+			continue
+		}
+
+		definition, ok := expected[output.SourceID]
+		if !ok || output.Kind != extractionLaneKind {
+			return
+		}
+		if _, duplicate := seen[output.SourceID]; duplicate {
+			return
+		}
+		if !compatibleCatalogExtractionContract(output.Contract, definition) ||
+			!compatibleCatalogExtractionProvenance(output.ExternalProvenance, receiptRunID, receiptPipelineID, definition) ||
+			!validExtractionPayload(bundleRoot, output.LocalPath, output.Checksum) {
+			return
+		}
+		seen[output.SourceID] = struct{}{}
+		hydrated = append(hydrated, hydratedExtraction{sourceID: output.SourceID, path: output.LocalPath})
+	}
+	if !indexSeen || len(seen) != len(expected) || len(record.Outputs) != len(expected)+1 {
+		return
+	}
+	for _, item := range hydrated {
+		_ = c.markAvailableFromExtractManifest(item.sourceID, item.path, producerRunID)
+	}
+}
+
+func compatibleCatalogExtractionContract(got *artifactmodel.ContractMetadata, want ExtractionArtifactDefinition) bool {
+	return got != nil && got.MediaType == want.MediaType && got.SchemaID == want.SchemaID &&
+		got.SchemaVersion == want.SchemaVersion && (want.ModuleKey == "" || got.ModuleKey == want.ModuleKey)
+}
+
+func compatibleCatalogExtractionProvenance(
+	got *artifactmodel.ExternalProvenance,
+	runID, pipelineID string,
+	want ExtractionArtifactDefinition,
+) bool {
+	return got != nil && got.System == "notarius" && got.RunID == runID && got.PipelineID == pipelineID &&
+		pipelineID == strings.TrimSpace(want.PipelineID) && got.ArtifactID == want.LaneID
+}
+
+func validExtractionPayload(bundleRoot, path, checksum string) bool {
+	if !filepath.IsAbs(path) || !pathWithinExtractionRoot(bundleRoot, path) || strings.TrimSpace(checksum) == "" {
+		return false
+	}
+	info, err := os.Lstat(path)
+	if err != nil || !info.Mode().IsRegular() || info.Mode()&os.ModeSymlink != 0 {
+		return false
+	}
+	if !safeExtractionComponents(bundleRoot, path) {
+		return false
+	}
+	actual, err := SHA256File(path)
+	if err != nil || actual != checksum {
+		return false
+	}
+	body, err := os.ReadFile(path)
+	return err == nil && json.Valid(body)
+}
+
+func safeExistingExtractionDirectory(sessionRoot, bundleRoot string) bool {
+	if !pathWithinExtractionRoot(sessionRoot, bundleRoot) || !safeExtractionComponents(sessionRoot, bundleRoot) {
+		return false
+	}
+	info, err := os.Lstat(bundleRoot)
+	return err == nil && info.IsDir() && info.Mode()&os.ModeSymlink == 0
+}
+
+func safeExtractionComponents(root, target string) bool {
+	relative, err := filepath.Rel(filepath.Clean(root), filepath.Clean(target))
+	if err != nil || relative == "." || relative == ".." || strings.HasPrefix(relative, ".."+string(filepath.Separator)) {
+		return false
+	}
+	current := filepath.Clean(root)
+	for _, part := range strings.Split(relative, string(filepath.Separator)) {
+		current = filepath.Join(current, part)
+		info, err := os.Lstat(current)
+		if err != nil || info.Mode()&os.ModeSymlink != 0 {
+			return false
+		}
+	}
+	return true
+}
+
+func pathWithinExtractionRoot(root, target string) bool {
+	relative, err := filepath.Rel(filepath.Clean(root), filepath.Clean(target))
+	return err == nil && relative != "." && relative != ".." && !strings.HasPrefix(relative, ".."+string(filepath.Separator))
+}
+
+func safeExtractionPathSegment(value string) bool {
+	return value != "" && value != "." && value != ".." && filepath.Base(value) == value &&
+		!strings.ContainsAny(value, `/\\`)
+}
+
+func extractionMetadataString(metadata map[string]any, key string) string {
+	value, _ := metadata[key].(string)
+	return strings.TrimSpace(value)
+}
+
+func extractionReceiptIdentity(metadata map[string]any) (string, string) {
+	receipt, _ := metadata["receipt"].(map[string]any)
+	return extractionMetadataString(receipt, "run_id"), extractionMetadataString(receipt, "pipeline_id")
+}

internal/artifacts/extraction_catalog_test.go

@@ -0,0 +1,222 @@
+package artifacts
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
+	"gitea.maximumdirect.net/eric/narratio/internal/manifest"
+)
+
+func TestArtifactCatalogRegistersExtractionArtifactsDeterministically(t *testing.T) {
+	catalog := NewArtifactCatalog()
+	if err := catalog.RegisterConfiguredArtifacts(map[string]ConfiguredArtifactDefinition{
+		"summary": {Enabled: true},
+	}, nil); err != nil {
+		t.Fatal(err)
+	}
+	if err := catalog.RegisterExtractionArtifacts(map[string]ExtractionArtifactDefinition{
+		"zeta":    {LaneID: "zeta"},
+		"summary": {LaneID: "summary"},
+		"alpha":   {LaneID: "alpha"},
+	}); err != nil {
+		t.Fatalf("RegisterExtractionArtifacts() error = %v", err)
+	}
+
+	entries := catalog.ListExtraction()
+	if len(entries) != 3 || entries[0].ExtractionKey != "alpha" || entries[1].ExtractionKey != "summary" || entries[2].ExtractionKey != "zeta" {
+		t.Fatalf("ListExtraction() = %#v, want alpha, summary, zeta", entries)
+	}
+	extractionID, ok := catalog.SourceIDForExtractionKey("summary")
+	if !ok || extractionID != "narratio.extraction.summary" {
+		t.Fatalf("SourceIDForExtractionKey(summary) = %q, %v", extractionID, ok)
+	}
+	configuredID, _ := catalog.SourceIDForConfiguredKey("summary")
+	if configuredID == extractionID {
+		t.Fatalf("configured and extraction source families collided at %q", extractionID)
+	}
+	entry, ok := catalog.Lookup(extractionID)
+	if !ok || entry.ProducerStage != "extract" || entry.OutputKind != "notarius_lane" || !entry.Planned {
+		t.Fatalf("Lookup(%q) = %#v, %v", extractionID, entry, ok)
+	}
+}
+
+func TestArtifactCatalogRejectsDuplicateExtractionRegistration(t *testing.T) {
+	catalog := NewArtifactCatalog()
+	definitions := map[string]ExtractionArtifactDefinition{"summary": {LaneID: "summary"}}
+	if err := catalog.RegisterExtractionArtifacts(definitions); err != nil {
+		t.Fatal(err)
+	}
+	if err := catalog.RegisterExtractionArtifacts(definitions); err == nil {
+		t.Fatal("second RegisterExtractionArtifacts() error = nil, want collision error")
+	}
+}
+
+func TestHydrateExtractionArtifactsAcceptsOnlyCompleteCurrentBundle(t *testing.T) {
+	paths, currentManifest, definitions := validExtractionCatalogFixture(t)
+	catalog := registeredExtractionCatalog(t, definitions)
+	catalog.HydrateExtractionArtifacts(paths, currentManifest, definitions)
+
+	entry, ok := catalog.Lookup(ExtractionArtifactSourceID("encounters"))
+	if !ok || !entry.Available {
+		t.Fatalf("hydrated entry = %#v, %v; want available", entry, ok)
+	}
+	if entry.Provenance != ArtifactProvenanceCurrentExtractManifest || entry.ProducerRunID != "extract-run-1" {
+		t.Fatalf("hydrated provenance = %#v", entry)
+	}
+	resolved, err := ResolveSessionArtifactWithCatalog(paths, currentManifest, entry.SourceID, catalog)
+	if err != nil {
+		t.Fatalf("ResolveSessionArtifactWithCatalog() error = %v", err)
+	}
+	if resolved.Path != entry.Path || resolved.ProducerRunID != "extract-run-1" {
+		t.Fatalf("resolved = %#v", resolved)
+	}
+}
+
+func TestHydrateExtractionArtifactsRejectsUntrustedManifestState(t *testing.T) {
+	tests := []struct {
+		name   string
+		mutate func(t *testing.T, paths SessionPaths, m *manifest.Manifest)
+	}{
+		{name: "missing stage", mutate: func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) { delete(m.Stages, "extract") }},
+		{name: "skipped", mutate: setExtractionStatus(manifest.StatusSkipped)},
+		{name: "failed", mutate: setExtractionStatus(manifest.StatusFailed)},
+		{name: "stale", mutate: setExtractionStatus(manifest.StatusStale)},
+		{name: "interrupted", mutate: setExtractionStatus(manifest.StatusInterrupted)},
+		{name: "missing source", mutate: func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			m.Stages["extract"].Outputs = m.Stages["extract"].Outputs[1:]
+		}},
+		{name: "inconsistent producer identity", mutate: func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			m.Stages["extract"].Outputs[0].ProducerRunID = "another-run"
+		}},
+		{name: "incompatible contract", mutate: func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			m.Stages["extract"].Outputs[0].Contract.SchemaVersion = "99"
+		}},
+		{name: "incompatible provenance", mutate: func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			m.Stages["extract"].Outputs[0].ExternalProvenance.RunID = "another-notarius-run"
+		}},
+		{name: "missing file", mutate: func(t *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			if err := os.Remove(m.Stages["extract"].Outputs[0].LocalPath); err != nil {
+				t.Fatal(err)
+			}
+		}},
+		{name: "tampered checksum", mutate: func(t *testing.T, _ SessionPaths, m *manifest.Manifest) {
+			if err := os.WriteFile(m.Stages["extract"].Outputs[0].LocalPath, []byte(`{"tampered":true}`), 0o644); err != nil {
+				t.Fatal(err)
+			}
+		}},
+		{name: "unsafe path", mutate: func(t *testing.T, paths SessionPaths, m *manifest.Manifest) {
+			outside := filepath.Join(paths.Root, "incidental.json")
+			writeExtractionFixtureFile(t, outside, `{"incidental":true}`)
+			m.Stages["extract"].Outputs[0].LocalPath = outside
+			m.Stages["extract"].Outputs[0].Checksum = extractionFixtureChecksum(t, outside)
+		}},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			paths, currentManifest, definitions := validExtractionCatalogFixture(t)
+			test.mutate(t, paths, currentManifest)
+			catalog := registeredExtractionCatalog(t, definitions)
+			catalog.HydrateExtractionArtifacts(paths, currentManifest, definitions)
+			entry, _ := catalog.Lookup(ExtractionArtifactSourceID("encounters"))
+			if entry.Available {
+				t.Fatalf("entry became available from %s manifest", test.name)
+			}
+		})
+	}
+}
+
+func TestResolveExtractionArtifactNeverDiscoversIncidentalBundleFile(t *testing.T) {
+	root := t.TempDir()
+	paths := SessionPaths{Root: root, ArtifactsDir: filepath.Join(root, "artifacts")}
+	incidental := filepath.Join(paths.ArtifactsDir, "notarius", "incidental", "lanes", "encounters.json")
+	writeExtractionFixtureFile(t, incidental, `{"encounters":[]}`)
+	definitions := extractionFixtureDefinitions()
+	catalog := registeredExtractionCatalog(t, definitions)
+
+	_, err := ResolveSessionArtifactWithCatalog(paths, manifest.New("session", fixtureTime), ExtractionArtifactSourceID("encounters"), catalog)
+	if err == nil || !errors.Is(err, ErrSessionArtifactNotFound) {
+		t.Fatalf("ResolveSessionArtifactWithCatalog() error = %v, want not found", err)
+	}
+}
+
+var fixtureTime = mustFixtureTime()
+
+func mustFixtureTime() (value time.Time) {
+	return time.Date(2026, 8, 10, 0, 0, 0, 0, time.UTC)
+}
+
+func setExtractionStatus(status manifest.StageStatus) func(*testing.T, SessionPaths, *manifest.Manifest) {
+	return func(_ *testing.T, _ SessionPaths, m *manifest.Manifest) { m.Stages["extract"].Status = status }
+}
+
+func extractionFixtureDefinitions() map[string]ExtractionArtifactDefinition {
+	return map[string]ExtractionArtifactDefinition{
+		"encounters": {
+			LaneID: "encounters", PipelineID: "campaign.extract", MediaType: "application/json",
+			SchemaID: "encounters", SchemaVersion: "1", ModuleKey: "encounters",
+		},
+	}
+}
+
+func registeredExtractionCatalog(t *testing.T, definitions map[string]ExtractionArtifactDefinition) *ArtifactCatalog {
+	t.Helper()
+	catalog := NewArtifactCatalog()
+	if err := catalog.RegisterExtractionArtifacts(definitions); err != nil {
+		t.Fatal(err)
+	}
+	return catalog
+}
+
+func validExtractionCatalogFixture(t *testing.T) (SessionPaths, *manifest.Manifest, map[string]ExtractionArtifactDefinition) {
+	t.Helper()
+	root := t.TempDir()
+	paths := SessionPaths{Root: root, ArtifactsDir: filepath.Join(root, "artifacts")}
+	bundleRoot := filepath.Join(paths.ArtifactsDir, "notarius", "extract-run-1")
+	lanePath := filepath.Join(bundleRoot, "lanes", "encounters.json")
+	indexPath := filepath.Join(bundleRoot, "index.json")
+	writeExtractionFixtureFile(t, lanePath, `{"encounters":[]}`)
+	writeExtractionFixtureFile(t, indexPath, `{"lanes":[]}`)
+	definitions := extractionFixtureDefinitions()
+	m := manifest.New("session", fixtureTime)
+	m.Stages["extract"] = &manifest.StageRecord{
+		Name: "extract", Status: manifest.StatusSucceeded,
+		Metadata: map[string]any{
+			"narratio_run_id": "extract-run-1", "bundle_root": bundleRoot,
+			"receipt": map[string]any{"run_id": "notarius-run-1", "pipeline_id": "campaign.extract"},
+		},
+		Outputs: []manifest.ArtifactRecord{
+			{
+				Kind: "notarius_lane", SourceID: ExtractionArtifactSourceID("encounters"), LocalPath: lanePath,
+				ProducerRunID: "extract-run-1", Checksum: extractionFixtureChecksum(t, lanePath),
+				Contract:           &artifactmodel.ContractMetadata{MediaType: "application/json", SchemaID: "encounters", SchemaVersion: "1", ModuleKey: "encounters"},
+				ExternalProvenance: &artifactmodel.ExternalProvenance{System: "notarius", RunID: "notarius-run-1", PipelineID: "campaign.extract", ArtifactID: "encounters"},
+			},
+			{Kind: "notarius_index", LocalPath: indexPath, ProducerRunID: "extract-run-1", Checksum: extractionFixtureChecksum(t, indexPath)},
+		},
+	}
+	return paths, m, definitions
+}
+
+func writeExtractionFixtureFile(t *testing.T, path, body string) {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(body), 0o644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func extractionFixtureChecksum(t *testing.T, path string) string {
+	t.Helper()
+	checksum, err := SHA256File(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return checksum
+}

internal/artifacts/paths.go

@@ -86,6 +86,36 @@ func SessionRunStageDirForCampaign(rootDir, campaign, sessionID, runID, stageNam
 	return filepath.Join(SessionRunRootForCampaign(rootDir, campaign, sessionID, runID), stageName)
 }
 
+// SessionRunExtractDirForCampaign returns the invocation-local extraction directory.
+func SessionRunExtractDirForCampaign(rootDir, campaign, sessionID, runID string) string {
+	return SessionRunStageDirForCampaign(rootDir, campaign, sessionID, runID, "extract")
+}
+
+// SessionRunNotariusReceiptPathForCampaign returns the invocation-local receipt path.
+func SessionRunNotariusReceiptPathForCampaign(rootDir, campaign, sessionID, runID string) string {
+	return filepath.Join(SessionRunExtractDirForCampaign(rootDir, campaign, sessionID, runID), "notarius.receipt.json")
+}
+
+// SessionRunNotariusLogPathForCampaign returns the invocation-local stderr log path.
+func SessionRunNotariusLogPathForCampaign(rootDir, campaign, sessionID, runID string) string {
+	return filepath.Join(SessionRunExtractDirForCampaign(rootDir, campaign, sessionID, runID), "notarius.stderr.log")
+}
+
+// SessionRunNotariusOutputRootForCampaign returns the invocation-local Notarius output root.
+func SessionRunNotariusOutputRootForCampaign(rootDir, campaign, sessionID, runID string) string {
+	return filepath.Join(SessionRunExtractDirForCampaign(rootDir, campaign, sessionID, runID), "notarius-output")
+}
+
+// SessionNotariusBundleDirForCampaign returns one immutable durable bundle destination.
+func SessionNotariusBundleDirForCampaign(rootDir, campaign, sessionID, runID string) string {
+	return filepath.Join(
+		SessionWorkDirForCampaign(rootDir, campaign, sessionID),
+		config.PathArtifactsDirSegment,
+		"notarius",
+		runID,
+	)
+}
+
 // SessionSpoolAudioDir returns the campaign/session/run scoped local spool audio path.
 func SessionSpoolAudioDir(spoolRoot, campaign, sessionID, runID string) string {
 	return filepath.Join(spoolRoot, campaign, sessionID, runID, config.PathAudioDirSegment)

internal/artifacts/paths_model_test.go

@@ -49,6 +49,33 @@ func TestSessionRunManifestPathForCampaign(t *testing.T) {
 	}
 }
 
+func TestSessionNotariusPathsForCampaign(t *testing.T) {
+	root := "/tmp/workspace"
+	campaign := "forsaken"
+	sessionID := "2026-04-19"
+	runID := "20260515T031522Z-a1b2c3d4"
+	extractDir := filepath.Join(root, "work", campaign, sessionID, "runs", runID, "extract")
+
+	tests := []struct {
+		name string
+		got  string
+		want string
+	}{
+		{name: "extract directory", got: SessionRunExtractDirForCampaign(root, campaign, sessionID, runID), want: extractDir},
+		{name: "receipt", got: SessionRunNotariusReceiptPathForCampaign(root, campaign, sessionID, runID), want: filepath.Join(extractDir, "notarius.receipt.json")},
+		{name: "stderr", got: SessionRunNotariusLogPathForCampaign(root, campaign, sessionID, runID), want: filepath.Join(extractDir, "notarius.stderr.log")},
+		{name: "output root", got: SessionRunNotariusOutputRootForCampaign(root, campaign, sessionID, runID), want: filepath.Join(extractDir, "notarius-output")},
+		{name: "durable bundle", got: SessionNotariusBundleDirForCampaign(root, campaign, sessionID, runID), want: filepath.Join(root, "work", campaign, sessionID, "artifacts", "notarius", runID)},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			if test.got != test.want {
+				t.Fatalf("path = %q, want %q", test.got, test.want)
+			}
+		})
+	}
+}
+
 func TestSessionPreviousPathsForCampaign(t *testing.T) {
 	root := "/tmp/workspace"
 	previousDir := SessionPreviousDirForCampaign(root, "forsaken", "2026-04-19")

internal/artifacts/s3.go

@@ -1,57 +0,0 @@
-package artifacts
-
-import (
-	"fmt"
-	"os"
-)
-
-// S3Store is a placeholder for future remote artifact persistence support.
-type S3Store struct {
-	Bucket string
-	Prefix string
-}
-
-// SessionPathsFor is not implemented for S3-backed storage.
-func (s *S3Store) SessionPathsFor(_, _ string) SessionPaths {
-	return SessionPaths{}
-}
-
-// EnsureLayoutFor returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) EnsureLayoutFor(_, _ string) (SessionPaths, error) {
-	return SessionPaths{}, fmt.Errorf("artifacts s3 ensure layout for campaign/session: not yet implemented")
-}
-
-// CopyInputFor returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) CopyInputFor(_, _, _, _ string) (Ref, error) {
-	return Ref{}, fmt.Errorf("artifacts s3 copy input for campaign/session: not yet implemented")
-}
-
-// Exists returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) Exists(_ string) (bool, error) {
-	return false, fmt.Errorf("artifacts s3 exists: not yet implemented")
-}
-
-// ExistsRef returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) ExistsRef(_ Ref) (bool, error) {
-	return false, fmt.Errorf("artifacts s3 exists ref: not yet implemented")
-}
-
-// WriteFileAtomic returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) WriteFileAtomic(_ string, _ []byte, _ os.FileMode) error {
-	return fmt.Errorf("artifacts s3 write file atomic: not yet implemented")
-}
-
-// Checksum returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) Checksum(_ string) (string, error) {
-	return "", fmt.Errorf("artifacts s3 checksum: not yet implemented")
-}
-
-// AcquireSessionLockFor returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) AcquireSessionLockFor(_, _ string) (*LockHandle, error) {
-	return nil, fmt.Errorf("artifacts s3 acquire lock for campaign/session: not yet implemented")
-}
-
-// ReleaseSessionLock returns a not-yet-implemented error in the scaffold.
-func (s *S3Store) ReleaseSessionLock(_ *LockHandle) error {
-	return fmt.Errorf("artifacts s3 release lock: not yet implemented")
-}

internal/artifacts/store.go

@@ -1,16 +1,23 @@
 package artifacts
 
-import "os"
+import (
+	"os"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
+)
 
 // Ref identifies a pipeline artifact and its local/remote coordinates.
 type Ref struct {
-	Kind         string
-	Category     string
-	SessionID    string
-	RelativePath string
-	AbsolutePath string
-	RemoteKey    string
-	Checksum     string
+	Kind               string
+	SourceID           string
+	Category           string
+	SessionID          string
+	RelativePath       string
+	AbsolutePath       string
+	RemoteKey          string
+	Checksum           string
+	Contract           *artifactmodel.ContractMetadata
+	ExternalProvenance *artifactmodel.ExternalProvenance
 }
 
 // Store is the local artifact/workdir abstraction used by orchestration code.

internal/config/config.go

@@ -30,6 +30,7 @@ type PipelineConfig struct {
 	Trim         *TrimConfig        `yaml:"trim"`
 	Render       *RenderConfig      `yaml:"render"`
 	Scriptorium  *ScriptoriumConfig `yaml:"scriptorium"`
+	Notarius     *NotariusConfig    `yaml:"notarius"`
 	Notification NotificationConfig `yaml:"notification"`
 }
 
@@ -252,6 +253,26 @@ type ScriptoriumInputConfig struct {
 	Required bool   `yaml:"required"`
 }
 
+// NotariusConfig configures structured artifact extraction by Notarius.
+type NotariusConfig struct {
+	Enabled          bool                            `yaml:"enabled"`
+	Binary           string                          `yaml:"binary"`
+	ConfigPath       string                          `yaml:"config_path"`
+	PipelineID       string                          `yaml:"pipeline_id"`
+	Timeout          string                          `yaml:"timeout"`
+	WorkingDirectory string                          `yaml:"working_directory"`
+	Outputs          map[string]NotariusOutputConfig `yaml:"outputs"`
+}
+
+// NotariusOutputConfig declares one required extraction lane contract.
+type NotariusOutputConfig struct {
+	LaneID        string `yaml:"lane_id"`
+	MediaType     string `yaml:"media_type"`
+	SchemaID      string `yaml:"schema_id"`
+	SchemaVersion string `yaml:"schema_version"`
+	ModuleKey     string `yaml:"module_key"`
+}
+
 // NotificationConfig configures notification backend settings.
 type NotificationConfig struct {
 	Backend   string `yaml:"backend"`

internal/config/defaults.go

@@ -30,13 +30,17 @@ const (
 	DefaultSeriatimCoalesceGap  = 3.0
 	DefaultSeriatimReport       = true
 
-	DefaultAuditaBinary  = "audita"
-	DefaultAuditaTimeout = "3h"
-	DefaultAuditaReport  = true
+	DefaultAuditaBinary    = "audita"
+	DefaultAuditaTimeout   = "3h"
+	DefaultAuditaReport    = true
+	DefaultNotariusBinary  = "notarius"
+	DefaultNotariusTimeout = "3h"
 
-	DefaultScriptoriumBinary             = "scriptorium"
-	DefaultScriptoriumTimeout            = "10m"
-	DefaultScriptoriumArtifactOutputRoot = "artifacts"
+	DefaultScriptoriumBinary                 = "scriptorium"
+	DefaultScriptoriumTimeout                = "10m"
+	DefaultScriptoriumArtifactOutputRoot     = "artifacts"
+	DefaultScriptoriumStickySessionVarName   = "session_id"
+	DefaultScriptoriumStickySessionVarPrefix = "narratio-session-"
 
 	DefaultTrimEnabled                   = true
 	DefaultTrimOutputPath                = artifactmodel.TranscriptPathFinalTrimmed

internal/config/load.go

@@ -19,6 +19,9 @@ func LoadPipeline(path string) (*PipelineConfig, error) {
 		return nil, fmt.Errorf("load pipeline config: %w", err)
 	}
 	applyPipelineDefaults(&cfg)
+	if err := resolveNotariusPaths(&cfg, path); err != nil {
+		return nil, fmt.Errorf("load pipeline config: %w", err)
+	}
 	return &cfg, nil
 }
 
@@ -86,12 +89,17 @@ func LoadSessionBytesWithOptions(label string, data []byte, opts SessionLoadOpti
 
 // LoadPublishLockStoreBytes loads a mutable session lock store with strict
 // field checking and source validation.
-func LoadPublishLockStoreBytes(label string, data []byte, scriptorium *ScriptoriumConfig) (*PublishLockStore, error) {
+func LoadPublishLockStoreBytes(
+	label string,
+	data []byte,
+	scriptorium *ScriptoriumConfig,
+	notarius *NotariusConfig,
+) (*PublishLockStore, error) {
 	var store PublishLockStore
 	if err := decodeStrictYAMLFromReader("publish lock store", label, strings.NewReader(string(data)), &store); err != nil {
 		return nil, fmt.Errorf("load publish lock store: %w", err)
 	}
-	locks, err := ValidatePublishLockRules(store.Locks, scriptorium, "locks")
+	locks, err := ValidatePublishLockRules(store.Locks, scriptorium, notarius, "locks")
 	if err != nil {
 		return nil, fmt.Errorf("load publish lock store: %w", err)
 	}
@@ -359,6 +367,7 @@ func applyPipelineDefaults(cfg *PipelineConfig) {
 	}
 	applyRenderDefaults(&cfg.Render)
 	applyScriptoriumDefaults(cfg.Scriptorium)
+	applyNotariusDefaults(cfg.Notarius)
 }
 
 func applyCampaignsDefaults(cfg *CampaignsConfig) {
@@ -516,6 +525,57 @@ func applyScriptoriumDefaults(cfg *ScriptoriumConfig) {
 	}
 }
 
+func applyNotariusDefaults(cfg *NotariusConfig) {
+	if cfg == nil {
+		return
+	}
+	if cfg.Binary == "" {
+		cfg.Binary = DefaultNotariusBinary
+	}
+	if cfg.Timeout == "" {
+		cfg.Timeout = DefaultNotariusTimeout
+	}
+}
+
+func resolveNotariusPaths(cfg *PipelineConfig, pipelinePath string) error {
+	if cfg == nil || cfg.Notarius == nil || !cfg.Notarius.Enabled {
+		return nil
+	}
+	pipelineAbs, err := filepath.Abs(pipelinePath)
+	if err != nil {
+		return fmt.Errorf("resolve pipeline config path %q: %w", pipelinePath, err)
+	}
+	baseDir := filepath.Dir(pipelineAbs)
+	resolve := func(value string) (string, error) {
+		trimmed := strings.TrimSpace(value)
+		if trimmed == "" {
+			return "", nil
+		}
+		if !filepath.IsAbs(trimmed) {
+			trimmed = filepath.Join(baseDir, trimmed)
+		}
+		return filepath.Abs(trimmed)
+	}
+
+	resolvedConfigPath, err := resolve(cfg.Notarius.ConfigPath)
+	if err != nil {
+		return fmt.Errorf("resolve pipeline.notarius.config_path: %w", err)
+	}
+	cfg.Notarius.ConfigPath = resolvedConfigPath
+	if strings.TrimSpace(cfg.Notarius.WorkingDirectory) == "" {
+		if resolvedConfigPath != "" {
+			cfg.Notarius.WorkingDirectory = filepath.Dir(resolvedConfigPath)
+		}
+		return nil
+	}
+	workingDirectory, err := resolve(cfg.Notarius.WorkingDirectory)
+	if err != nil {
+		return fmt.Errorf("resolve pipeline.notarius.working_directory: %w", err)
+	}
+	cfg.Notarius.WorkingDirectory = workingDirectory
+	return nil
+}
+
 func applyTrimDefaults(cfg *TrimConfig) {
 	if cfg == nil {
 		return

internal/config/load_validate_test.go

@@ -3,6 +3,7 @@ package config
 import (
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"testing"
 )
@@ -1049,6 +1050,11 @@ func TestExamplesLoadAndValidate(t *testing.T) {
 			pipelineFile: "pipeline.full.annotated.yml",
 			sessionFile:  "session.local-audio.yml",
 		},
+		{
+			name:         "extraction subset pipeline with local audio session",
+			pipelineFile: "pipeline.extraction-subset.yml",
+			sessionFile:  "session.local-audio.yml",
+		},
 	}
 
 	for _, tt := range tests {
@@ -1068,6 +1074,47 @@ func TestExamplesLoadAndValidate(t *testing.T) {
 	}
 }
 
+func TestMaintainedExtractionExamplesPreservePublishedContracts(t *testing.T) {
+	examplesDir := filepath.Join("..", "..", "examples")
+	full, err := LoadPipeline(filepath.Join(examplesDir, "pipeline.full.annotated.yml"))
+	if err != nil {
+		t.Fatalf("load full example error = %v", err)
+	}
+	want := map[string]NotariusOutputConfig{
+		"item_registry":        {LaneID: "item-registry", MediaType: "application/json", SchemaID: "notarius.dnd.item_registry", SchemaVersion: "v1", ModuleKey: "dnd/item-registry"},
+		"npc_registry":         {LaneID: "npc-registry", MediaType: "application/json", SchemaID: "notarius.dnd.npc_registry", SchemaVersion: "v1", ModuleKey: "dnd/npc-registry"},
+		"location_registry":    {LaneID: "location-registry", MediaType: "application/json", SchemaID: "notarius.dnd.location_registry", SchemaVersion: "v1", ModuleKey: "dnd/location-registry"},
+		"scene_descriptions":   {LaneID: "scene-descriptions", MediaType: "application/json", SchemaID: "notarius.dnd.scene_descriptions", SchemaVersion: "v1", ModuleKey: "dnd/scene-descriptions"},
+		"item_occurrences":     {LaneID: "item-occurrences", MediaType: "application/json", SchemaID: "notarius.dnd.item_occurrences", SchemaVersion: "v1", ModuleKey: "dnd/item-occurrences"},
+		"spells":               {LaneID: "spells", MediaType: "application/json", SchemaID: "notarius.dnd.spells", SchemaVersion: "v1", ModuleKey: "dnd/spells"},
+		"combat_turns":         {LaneID: "combat-turns", MediaType: "application/json", SchemaID: "notarius.dnd.combat_turns", SchemaVersion: "v1", ModuleKey: "dnd/combat-turns"},
+		"npc_occurrences":      {LaneID: "npc-occurrences", MediaType: "application/json", SchemaID: "notarius.dnd.npc_occurrences", SchemaVersion: "v1", ModuleKey: "dnd/npc-occurrences"},
+		"location_occurrences": {LaneID: "location-occurrences", MediaType: "application/json", SchemaID: "notarius.dnd.location_occurrences", SchemaVersion: "v1", ModuleKey: "dnd/location-occurrences"},
+		"enemy_events":         {LaneID: "enemy-events", MediaType: "application/json", SchemaID: "notarius.dnd.enemy_events", SchemaVersion: "v1", ModuleKey: "dnd/enemy-events"},
+	}
+	if full.Notarius == nil || !reflect.DeepEqual(full.Notarius.Outputs, want) {
+		t.Fatalf("full example outputs = %#v, want %#v", full.Notarius, want)
+	}
+
+	subset, err := LoadPipeline(filepath.Join(examplesDir, "pipeline.extraction-subset.yml"))
+	if err != nil {
+		t.Fatalf("load subset example error = %v", err)
+	}
+	brief := subset.Scriptorium.Artifacts["session_brief"]
+	wantSources := map[string]string{
+		"npcs":      "narratio.extraction.npc_registry",
+		"locations": "narratio.extraction.location_registry",
+		"scenes":    "narratio.extraction.scene_descriptions",
+	}
+	gotSources := make(map[string]string, len(brief.Inputs))
+	for name, input := range brief.Inputs {
+		gotSources[name] = input.Source
+	}
+	if !reflect.DeepEqual(gotSources, wantSources) {
+		t.Fatalf("subset example sources = %#v, want %#v", gotSources, wantSources)
+	}
+}
+
 func writeConfigFiles(t *testing.T, pipelineYAML, sessionYAML string) (string, string) {
 	t.Helper()
 	if !strings.Contains(pipelineYAML, "\naudita:") && !strings.HasPrefix(pipelineYAML, "audita:") {

internal/config/notarius_test.go

@@ -0,0 +1,283 @@
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestNotariusOmittedAndDisabledBehavior(t *testing.T) {
+	dir := t.TempDir()
+	omittedPath := filepath.Join(dir, "omitted.yml")
+	if err := os.WriteFile(omittedPath, []byte(testPipelineBaseYAML), 0o644); err != nil {
+		t.Fatalf("write omitted pipeline: %v", err)
+	}
+	omitted, err := LoadPipeline(omittedPath)
+	if err != nil {
+		t.Fatalf("LoadPipeline(omitted) error = %v", err)
+	}
+	if omitted.Notarius != nil {
+		t.Fatalf("Notarius = %#v, want nil when omitted", omitted.Notarius)
+	}
+
+	disabledPath := filepath.Join(dir, "disabled.yml")
+	disabledYAML := testPipelineBaseYAML + `
+notarius:
+  enabled: false
+  config_path: relative/notarius.yml
+`
+	if err := os.WriteFile(disabledPath, []byte(disabledYAML), 0o644); err != nil {
+		t.Fatalf("write disabled pipeline: %v", err)
+	}
+	disabled, err := LoadPipeline(disabledPath)
+	if err != nil {
+		t.Fatalf("LoadPipeline(disabled) error = %v", err)
+	}
+	if disabled.Notarius == nil {
+		t.Fatal("Notarius = nil, want configured disabled section")
+	}
+	if disabled.Notarius.Enabled {
+		t.Fatal("Notarius.Enabled = true, want false")
+	}
+	if disabled.Notarius.Binary != DefaultNotariusBinary || disabled.Notarius.Timeout != DefaultNotariusTimeout {
+		t.Fatalf("disabled defaults = %#v", disabled.Notarius)
+	}
+	if disabled.Notarius.ConfigPath != "relative/notarius.yml" || disabled.Notarius.WorkingDirectory != "" {
+		t.Fatalf("disabled paths were resolved unexpectedly: %#v", disabled.Notarius)
+	}
+}
+
+func TestNotariusEnabledDefaultsAndPathResolution(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "deployment", "pipeline.yml")
+	if err := os.MkdirAll(filepath.Dir(pipelinePath), 0o755); err != nil {
+		t.Fatalf("MkdirAll() error = %v", err)
+	}
+	pipelineYAML := testPipelineBaseYAML + `
+notarius:
+  enabled: true
+  config_path: notarius/config.yml
+  pipeline_id: dnd-session
+  outputs:
+    npc_registry:
+      lane_id: npc-registry
+      media_type: application/json
+      schema_id: notarius.dnd.npc_registry
+      schema_version: v1
+`
+	if err := os.WriteFile(pipelinePath, []byte(pipelineYAML), 0o644); err != nil {
+		t.Fatalf("write pipeline: %v", err)
+	}
+
+	cfg, err := LoadPipeline(pipelinePath)
+	if err != nil {
+		t.Fatalf("LoadPipeline() error = %v", err)
+	}
+	wantConfigPath := filepath.Join(filepath.Dir(pipelinePath), "notarius", "config.yml")
+	if cfg.Notarius.Binary != "notarius" || cfg.Notarius.Timeout != "3h" {
+		t.Fatalf("defaults = %#v", cfg.Notarius)
+	}
+	if cfg.Notarius.ConfigPath != wantConfigPath {
+		t.Fatalf("config_path = %q, want %q", cfg.Notarius.ConfigPath, wantConfigPath)
+	}
+	if cfg.Notarius.WorkingDirectory != filepath.Dir(wantConfigPath) {
+		t.Fatalf("working_directory = %q, want %q", cfg.Notarius.WorkingDirectory, filepath.Dir(wantConfigPath))
+	}
+
+	explicitPath := filepath.Join(dir, "explicit.yml")
+	explicitYAML := strings.Replace(pipelineYAML, "  pipeline_id: dnd-session\n", "  pipeline_id: dnd-session\n  working_directory: runtime\n", 1)
+	if err := os.WriteFile(explicitPath, []byte(explicitYAML), 0o644); err != nil {
+		t.Fatalf("write explicit pipeline: %v", err)
+	}
+	explicit, err := LoadPipeline(explicitPath)
+	if err != nil {
+		t.Fatalf("LoadPipeline(explicit working directory) error = %v", err)
+	}
+	if explicit.Notarius.WorkingDirectory != filepath.Join(dir, "runtime") {
+		t.Fatalf("explicit working_directory = %q, want %q", explicit.Notarius.WorkingDirectory, filepath.Join(dir, "runtime"))
+	}
+}
+
+func TestNotariusStrictYAML(t *testing.T) {
+	tests := []struct {
+		name string
+		yaml string
+	}{
+		{name: "unknown section field", yaml: "notarius:\n  unknown: true\n"},
+		{name: "unknown output field", yaml: "notarius:\n  outputs:\n    npc_registry:\n      lane_id: npc-registry\n      unknown: true\n"},
+		{name: "unsupported session id", yaml: "notarius:\n  session_id: forbidden\n"},
+		{name: "unsupported model", yaml: "notarius:\n  model: forbidden\n"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			path := filepath.Join(t.TempDir(), "pipeline.yml")
+			if err := os.WriteFile(path, []byte(testPipelineBaseYAML+"\n"+tt.yaml), 0o644); err != nil {
+				t.Fatalf("write pipeline: %v", err)
+			}
+			if _, err := LoadPipeline(path); err == nil || !strings.Contains(err.Error(), "strict decode failed") {
+				t.Fatalf("LoadPipeline() error = %v, want strict decode failure", err)
+			}
+		})
+	}
+}
+
+func TestNotariusEnabledValidation(t *testing.T) {
+	valid := `notarius:
+  enabled: true
+  config_path: ./notarius.yml
+  pipeline_id: dnd-session
+  timeout: 45m
+  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
+`
+	tests := []struct {
+		name    string
+		section string
+		wantErr string
+	}{
+		{name: "valid", section: valid},
+		{name: "blank binary", section: strings.Replace(valid, "  enabled: true\n", "  enabled: true\n  binary: \"   \"\n", 1), wantErr: "pipeline.notarius.binary is required"},
+		{name: "missing config path", section: strings.Replace(valid, "  config_path: ./notarius.yml\n", "", 1), wantErr: "pipeline.notarius.config_path is required"},
+		{name: "missing pipeline id", section: strings.Replace(valid, "  pipeline_id: dnd-session\n", "", 1), wantErr: "pipeline.notarius.pipeline_id is required"},
+		{name: "missing outputs", section: strings.Split(valid, "  outputs:\n")[0], wantErr: "pipeline.notarius.outputs must contain at least one output"},
+		{name: "zero timeout", section: strings.Replace(valid, "  timeout: 45m", "  timeout: 0s", 1), wantErr: "pipeline.notarius.timeout must be positive"},
+		{name: "negative timeout", section: strings.Replace(valid, "  timeout: 45m", "  timeout: -1s", 1), wantErr: "pipeline.notarius.timeout must be positive"},
+		{name: "invalid timeout", section: strings.Replace(valid, "  timeout: 45m", "  timeout: later", 1), wantErr: "pipeline.notarius.timeout must be a valid duration"},
+		{name: "invalid output key", section: strings.Replace(valid, "    npc_registry:", "    npc-registry:", 1), wantErr: "outputs keys must match"},
+		{name: "missing lane", section: strings.Replace(valid, "      lane_id: npc-registry\n", "", 1), wantErr: "lane_id is required"},
+		{name: "missing media type", section: strings.Replace(valid, "      media_type: application/json\n", "", 1), wantErr: "media_type is required"},
+		{name: "missing schema id", section: strings.Replace(valid, "      schema_id: notarius.dnd.npc_registry\n", "", 1), wantErr: "schema_id is required"},
+		{name: "missing schema version", section: strings.Replace(valid, "      schema_version: v1\n", "", 1), wantErr: "schema_version is required"},
+		{
+			name: "normalized key collision",
+			section: strings.Replace(valid, "      module_key: dnd/npc-registry\n", `      module_key: dnd/npc-registry
+    " npc_registry ":
+      lane_id: npc-registry-two
+      media_type: application/json
+      schema_id: two
+      schema_version: v1
+`, 1),
+			wantErr: "normalize to duplicate source",
+		},
+		{
+			name: "duplicate normalized lane",
+			section: strings.Replace(valid, "      module_key: dnd/npc-registry\n", `      module_key: dnd/npc-registry
+    spells:
+      lane_id: " npc-registry "
+      media_type: application/json
+      schema_id: two
+      schema_version: v1
+`, 1),
+			wantErr: "duplicates pipeline.notarius.outputs.npc_registry.lane_id",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			pipelinePath, sessionPath := writeConfigFiles(t, testPipelineBaseYAML+"\n"+tt.section, testSessionBaseYAML)
+			cfg, err := Load(pipelinePath, sessionPath)
+			if err != nil {
+				t.Fatalf("Load() error = %v", err)
+			}
+			err = Validate(cfg)
+			if tt.wantErr == "" {
+				if err != nil {
+					t.Fatalf("Validate() error = %v", err)
+				}
+				output := cfg.Pipeline.Notarius.Outputs["npc_registry"]
+				if output.LaneID != "npc-registry" || output.ModuleKey != "dnd/npc-registry" {
+					t.Fatalf("normalized output = %#v", output)
+				}
+				return
+			}
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want containing %q", err, tt.wantErr)
+			}
+		})
+	}
+}
+
+func TestExtractionReferencesRequireDeclaredOutput(t *testing.T) {
+	declared := `notarius:
+  enabled: false
+  outputs:
+    npc_registry: {}
+`
+	tests := []struct {
+		name    string
+		body    string
+		wantErr string
+	}{
+		{
+			name: "scriptorium declared extraction",
+			body: declared + `scriptorium:
+  artifacts:
+    recap:
+      inputs:
+        npcs:
+          source: narratio.extraction.npc_registry
+`,
+		},
+		{
+			name: "scriptorium unknown extraction",
+			body: declared + `scriptorium:
+  artifacts:
+    recap:
+      inputs:
+        npcs:
+          source: narratio.extraction.unknown
+`,
+			wantErr: `references unknown extraction output "unknown"`,
+		},
+		{
+			name: "publish declared extraction",
+			body: declared + `publish:
+  outputs:
+    - source: narratio.extraction.npc_registry
+      dest: artifacts/npc-registry.json
+`,
+		},
+		{
+			name: "publish unknown extraction",
+			body: declared + `publish:
+  outputs:
+    - source: narratio.extraction.unknown
+      dest: artifacts/unknown.json
+`,
+			wantErr: `extraction output "unknown" is not defined`,
+		},
+		{
+			name: "publish lock declared extraction",
+			body: declared + `publish:
+  locks:
+    - source: narratio.extraction.npc_registry
+`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			pipelinePath, sessionPath := writeConfigFiles(t, testPipelineBaseYAML+"\n"+tt.body, testSessionBaseYAML)
+			cfg, err := Load(pipelinePath, sessionPath)
+			if err != nil {
+				t.Fatalf("Load() error = %v", err)
+			}
+			err = Validate(cfg)
+			if tt.wantErr == "" {
+				if err != nil {
+					t.Fatalf("Validate() error = %v", err)
+				}
+				return
+			}
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("Validate() error = %v, want containing %q", err, tt.wantErr)
+			}
+		})
+	}
+}

internal/config/storage_publish_test.go

@@ -537,7 +537,7 @@ func TestPublishLockStoreBytesStrictDecodeAndValidation(t *testing.T) {
 	store, err := LoadPublishLockStoreBytes("locks.yml", []byte(`locks:
   - source: narratio.transcript.final_trimmed
     reason: reviewed
-`), nil)
+`), nil, nil)
 	if err != nil {
 		t.Fatalf("LoadPublishLockStoreBytes() error = %v", err)
 	}
@@ -548,7 +548,7 @@ func TestPublishLockStoreBytesStrictDecodeAndValidation(t *testing.T) {
 	_, err = LoadPublishLockStoreBytes("locks.yml", []byte(`locks:
   - source: narratio.transcript.final_trimmed
     dest: transcripts/final.trimmed.json
-`), nil)
+`), nil, nil)
 	if err == nil || !strings.Contains(err.Error(), "strict decode failed") {
 		t.Fatalf("unknown field error = %v, want strict decode failed", err)
 	}
@@ -556,7 +556,7 @@ func TestPublishLockStoreBytesStrictDecodeAndValidation(t *testing.T) {
 	_, err = LoadPublishLockStoreBytes("locks.yml", []byte(`locks:
   - source: narratio.transcript.final_trimmed
   - source: narratio.transcript.final_trimmed
-`), nil)
+`), nil, nil)
 	if err == nil || !strings.Contains(err.Error(), "duplicates another publish lock source") {
 		t.Fatalf("duplicate error = %v", err)
 	}

internal/config/validate.go

@@ -6,9 +6,11 @@ import (
 	"net/url"
 	"path/filepath"
 	"regexp"
+	"sort"
 	"strings"
 	"time"
 
+	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactpolicy"
 	"gitea.maximumdirect.net/eric/narratio/internal/pathsafe"
 )
@@ -85,7 +87,10 @@ func validatePipeline(cfg *PipelineConfig) error {
 	if err := validateCache(cfg.Cache); err != nil {
 		return err
 	}
-	if err := validatePublish(cfg.Publish, cfg.Scriptorium); err != nil {
+	if err := validateNotarius(cfg.Notarius, cfg.Scriptorium); err != nil {
+		return err
+	}
+	if err := validatePublish(cfg.Publish, cfg.Scriptorium, cfg.Notarius); err != nil {
 		return err
 	}
 	if err := validateWhisperX(cfg.WhisperX); err != nil {
@@ -106,7 +111,7 @@ func validatePipeline(cfg *PipelineConfig) error {
 	if err := validateRender(cfg.Render); err != nil {
 		return err
 	}
-	if err := validateScriptorium(cfg.Scriptorium); err != nil {
+	if err := validateScriptorium(cfg.Scriptorium, cfg.Notarius); err != nil {
 		return err
 	}
 	if err := validateDuration("pipeline.notification.timeout", cfg.Notification.Timeout); err != nil {
@@ -149,11 +154,12 @@ func validateCache(cfg CacheConfig) error {
 	return nil
 }
 
-func validatePublish(cfg *PublishConfig, scriptorium *ScriptoriumConfig) error {
+func validatePublish(cfg *PublishConfig, scriptorium *ScriptoriumConfig, notarius *NotariusConfig) error {
 	if cfg == nil {
 		return nil
 	}
 	configuredOutputs := scriptoriumOutputPathMap(scriptorium)
+	extractionOutputs := notariusOutputKeySet(notarius)
 	seenDest := map[string]struct{}{}
 	for i, item := range cfg.Outputs {
 		prefix := fmt.Sprintf("pipeline.publish.outputs[%d]", i)
@@ -161,12 +167,12 @@ func validatePublish(cfg *PublishConfig, scriptorium *ScriptoriumConfig) error {
 		if source == "" {
 			return fmt.Errorf("%s.source is required", prefix)
 		}
-		if _, err := artifactpolicy.ValidatePublishSource(source, configuredOutputs); err != nil {
+		if _, err := artifactpolicy.ValidatePublishSourceWithExtractions(source, configuredOutputs, extractionOutputs); err != nil {
 			return fmt.Errorf("%s.source %q is unsupported: %w", prefix, item.Source, err)
 		}
 		dest := strings.TrimSpace(item.Dest)
 		if dest == "" {
-			derivedDest, err := artifactpolicy.ResolvePublishedDestination(source, "", configuredOutputs)
+			derivedDest, err := artifactpolicy.ResolvePublishedDestinationWithExtractions(source, "", configuredOutputs, extractionOutputs)
 			if err != nil {
 				return fmt.Errorf("%s.dest is required when destination cannot be derived from %q: %w", prefix, source, err)
 			}
@@ -189,7 +195,7 @@ func validatePublish(cfg *PublishConfig, scriptorium *ScriptoriumConfig) error {
 		}
 		seenDest[normalizedDest] = struct{}{}
 	}
-	locks, err := ValidatePublishLockRules(cfg.Locks, scriptorium, "pipeline.publish.locks")
+	locks, err := ValidatePublishLockRules(cfg.Locks, scriptorium, notarius, "pipeline.publish.locks")
 	if err != nil {
 		return err
 	}
@@ -198,10 +204,11 @@ func validatePublish(cfg *PublishConfig, scriptorium *ScriptoriumConfig) error {
 }
 
 // ValidatePublishLockRules validates and normalizes source-based publish locks.
-func ValidatePublishLockRules(locks []PublishLockRule, scriptorium *ScriptoriumConfig, label string) ([]PublishLockRule, error) {
+func ValidatePublishLockRules(locks []PublishLockRule, scriptorium *ScriptoriumConfig, notarius *NotariusConfig, label string) ([]PublishLockRule, error) {
 	seenLocks := map[string]struct{}{}
 	out := make([]PublishLockRule, 0, len(locks))
 	configuredOutputs := scriptoriumOutputPathMap(scriptorium)
+	extractionOutputs := notariusOutputKeySet(notarius)
 	if strings.TrimSpace(label) == "" {
 		label = "publish.locks"
 	}
@@ -211,7 +218,7 @@ func ValidatePublishLockRules(locks []PublishLockRule, scriptorium *ScriptoriumC
 		if source == "" {
 			return nil, fmt.Errorf("%s.source is required", prefix)
 		}
-		if _, err := artifactpolicy.ValidatePublishSource(source, configuredOutputs); err != nil {
+		if _, err := artifactpolicy.ValidatePublishSourceWithExtractions(source, configuredOutputs, extractionOutputs); err != nil {
 			return nil, fmt.Errorf("%s.source %q is unsupported: %w", prefix, item.Source, err)
 		}
 		if _, ok := seenLocks[source]; ok {
@@ -264,6 +271,20 @@ func scriptoriumOutputPathMap(scriptorium *ScriptoriumConfig) map[string]string
 	return out
 }
 
+func notariusOutputKeySet(notarius *NotariusConfig) map[string]struct{} {
+	out := map[string]struct{}{}
+	if notarius == nil {
+		return out
+	}
+	for key := range notarius.Outputs {
+		trimmed := strings.TrimSpace(key)
+		if artifactpolicy.IsConfiguredKey(trimmed) {
+			out[trimmed] = struct{}{}
+		}
+	}
+	return out
+}
+
 func validateSecrets(cfg *SecretsConfig) error {
 	if cfg == nil {
 		return nil
@@ -474,7 +495,103 @@ func validateAudita(cfg AuditaConfig) error {
 	return nil
 }
 
-func validateScriptorium(cfg *ScriptoriumConfig) error {
+func validateNotarius(cfg *NotariusConfig, scriptorium *ScriptoriumConfig) error {
+	if cfg == nil || !cfg.Enabled {
+		return nil
+	}
+	if strings.TrimSpace(cfg.Binary) == "" {
+		return fmt.Errorf("pipeline.notarius.binary is required when pipeline.notarius.enabled is true")
+	}
+	if strings.TrimSpace(cfg.ConfigPath) == "" {
+		return fmt.Errorf("pipeline.notarius.config_path is required when pipeline.notarius.enabled is true")
+	}
+	if strings.TrimSpace(cfg.PipelineID) == "" {
+		return fmt.Errorf("pipeline.notarius.pipeline_id is required when pipeline.notarius.enabled is true")
+	}
+	if len(cfg.Outputs) == 0 {
+		return fmt.Errorf("pipeline.notarius.outputs must contain at least one output when pipeline.notarius.enabled is true")
+	}
+	duration, err := time.ParseDuration(strings.TrimSpace(cfg.Timeout))
+	if err != nil {
+		return fmt.Errorf("pipeline.notarius.timeout must be a valid duration: %w", err)
+	}
+	if duration <= 0 {
+		return fmt.Errorf("pipeline.notarius.timeout must be positive")
+	}
+	if strings.TrimSpace(cfg.WorkingDirectory) == "" {
+		return fmt.Errorf("pipeline.notarius.working_directory is required when pipeline.notarius.enabled is true")
+	}
+
+	reservedSources := map[string]string{}
+	for _, spec := range artifactmodel.RuntimeTranscriptArtifacts() {
+		reservedSources[spec.SourceID] = "built-in source"
+	}
+	reservedSources[artifactpolicy.SourceBoundsSession] = "built-in source"
+	if scriptorium != nil {
+		for key := range scriptorium.Artifacts {
+			normalizedKey := strings.TrimSpace(key)
+			reservedSources[artifactpolicy.ConfiguredSourceID(normalizedKey)] = "configured Scriptorium source"
+			reservedSources[artifactpolicy.PreviousSessionSourceID(normalizedKey)] = "previous-session source"
+		}
+	}
+
+	rawKeys := make([]string, 0, len(cfg.Outputs))
+	for key := range cfg.Outputs {
+		rawKeys = append(rawKeys, key)
+	}
+	sort.Strings(rawKeys)
+	normalizedOutputs := make(map[string]NotariusOutputConfig, len(cfg.Outputs))
+	sourceOwners := map[string]string{}
+	laneOwners := map[string]string{}
+	for _, rawKey := range rawKeys {
+		output := cfg.Outputs[rawKey]
+		key := strings.TrimSpace(rawKey)
+		if !artifactpolicy.IsConfiguredKey(key) {
+			return fmt.Errorf("pipeline.notarius.outputs keys must match ^[a-z][a-z0-9_]*$")
+		}
+		sourceID := artifactpolicy.ExtractionSourceID(key)
+		if previousKey, ok := sourceOwners[sourceID]; ok {
+			return fmt.Errorf("pipeline.notarius.outputs keys %q and %q normalize to duplicate source %q", previousKey, rawKey, sourceID)
+		}
+		if owner, ok := reservedSources[sourceID]; ok {
+			return fmt.Errorf("pipeline.notarius.outputs.%s source %q collides with %s", key, sourceID, owner)
+		}
+		sourceOwners[sourceID] = rawKey
+
+		output.LaneID = strings.TrimSpace(output.LaneID)
+		output.MediaType = strings.TrimSpace(output.MediaType)
+		output.SchemaID = strings.TrimSpace(output.SchemaID)
+		output.SchemaVersion = strings.TrimSpace(output.SchemaVersion)
+		output.ModuleKey = strings.TrimSpace(output.ModuleKey)
+		prefix := "pipeline.notarius.outputs." + key
+		if output.LaneID == "" {
+			return fmt.Errorf("%s.lane_id is required", prefix)
+		}
+		if previousKey, ok := laneOwners[output.LaneID]; ok {
+			return fmt.Errorf("%s.lane_id %q duplicates pipeline.notarius.outputs.%s.lane_id", prefix, output.LaneID, previousKey)
+		}
+		laneOwners[output.LaneID] = key
+		if output.MediaType == "" {
+			return fmt.Errorf("%s.media_type is required", prefix)
+		}
+		if output.SchemaID == "" {
+			return fmt.Errorf("%s.schema_id is required", prefix)
+		}
+		if output.SchemaVersion == "" {
+			return fmt.Errorf("%s.schema_version is required", prefix)
+		}
+		normalizedOutputs[key] = output
+	}
+	cfg.Binary = strings.TrimSpace(cfg.Binary)
+	cfg.ConfigPath = filepath.Clean(cfg.ConfigPath)
+	cfg.PipelineID = strings.TrimSpace(cfg.PipelineID)
+	cfg.Timeout = strings.TrimSpace(cfg.Timeout)
+	cfg.WorkingDirectory = filepath.Clean(cfg.WorkingDirectory)
+	cfg.Outputs = normalizedOutputs
+	return nil
+}
+
+func validateScriptorium(cfg *ScriptoriumConfig, notarius *NotariusConfig) error {
 	if cfg == nil {
 		return nil
 	}
@@ -491,7 +608,7 @@ func validateScriptorium(cfg *ScriptoriumConfig) error {
 	configuredArtifacts := make(map[string]struct{}, len(cfg.Artifacts))
 	referencedArtifacts := make(map[string]struct{})
 	for artifactName := range cfg.Artifacts {
-		if !scriptoriumArtifactKeyRE.MatchString(strings.TrimSpace(artifactName)) {
+		if !artifactpolicy.IsConfiguredKey(artifactName) {
 			return fmt.Errorf("pipeline.scriptorium.artifacts keys must match ^[a-z][a-z0-9_]*$")
 		}
 		configuredArtifacts[artifactName] = struct{}{}
@@ -544,7 +661,7 @@ func validateScriptorium(cfg *ScriptoriumConfig) error {
 				return fmt.Errorf("pipeline.scriptorium.artifacts.%s.inputs.%s.source is required", artifactName, inputName)
 			}
 
-			referencedArtifact, err := validateScriptoriumInputSource(artifactName, inputName, source, configuredArtifacts)
+			referencedArtifact, err := validateScriptoriumInputSource(artifactName, inputName, source, configuredArtifacts, notariusOutputKeySet(notarius))
 			if err != nil {
 				return err
 			}
@@ -684,9 +801,12 @@ func publishUploadConfiguredForS3(pipeline *PipelineConfig) bool {
 
 var windowsAbsPathRE = regexp.MustCompile(`^[A-Za-z]:[\\/].*`)
 var envVarNameRE = regexp.MustCompile(`^[A-Za-z_][A-Za-z0-9_]*$`)
-var scriptoriumArtifactKeyRE = regexp.MustCompile(`^[a-z][a-z0-9_]*$`)
 
-func validateScriptoriumInputSource(artifactName, inputName, source string, configuredArtifacts map[string]struct{}) (string, error) {
+func validateScriptoriumInputSource(
+	artifactName, inputName, source string,
+	configuredArtifacts map[string]struct{},
+	extractionOutputs map[string]struct{},
+) (string, error) {
 	trimmedSource := strings.TrimSpace(source)
 	descriptor, err := artifactpolicy.DescribeScriptoriumInputSource(trimmedSource)
 	if err != nil {
@@ -705,7 +825,7 @@ func validateScriptoriumInputSource(artifactName, inputName, source string, conf
 			source,
 		)
 	}
-	if err := artifactpolicy.ValidateInputConfiguredReference(descriptor, configuredArtifacts); err != nil {
+	if err := artifactpolicy.ValidateInputReference(descriptor, configuredArtifacts, extractionOutputs); err != nil {
 		var unknownConfigured *artifactpolicy.UnknownConfiguredArtifactError
 		if errors.As(err, &unknownConfigured) {
 			return "", fmt.Errorf(
@@ -716,6 +836,16 @@ func validateScriptoriumInputSource(artifactName, inputName, source string, conf
 				unknownConfigured.ConfiguredKey,
 			)
 		}
+		var unknownExtraction *artifactpolicy.UnknownExtractionArtifactError
+		if errors.As(err, &unknownExtraction) {
+			return "", fmt.Errorf(
+				"pipeline.scriptorium.artifacts.%s.inputs.%s.source %q references unknown extraction output %q",
+				artifactName,
+				inputName,
+				source,
+				unknownExtraction.ConfiguredKey,
+			)
+		}
 		return "", fmt.Errorf(
 			"pipeline.scriptorium.artifacts.%s.inputs.%s.source %q is unsupported",
 			artifactName,

internal/contracts/artifact.go

@@ -1,7 +0,0 @@
-package contracts
-
-// ArtifactResult is a placeholder generated artifact contract.
-type ArtifactResult struct {
-	Schema string `json:"schema"`
-	Path   string `json:"path"`
-}

internal/contracts/session.go

@@ -1,7 +0,0 @@
-package contracts
-
-// SessionManifest is a placeholder durable session-run contract.
-type SessionManifest struct {
-	Schema    string `json:"schema"`
-	SessionID string `json:"session_id"`
-}

internal/contracts/transcript.go

@@ -1,19 +0,0 @@
-package contracts
-
-// SpeakerTranscript is a placeholder transcription artifact contract.
-type SpeakerTranscript struct {
-	Schema    string `json:"schema"`
-	SessionID string `json:"session_id"`
-}
-
-// CanonicalTranscript is a placeholder merged transcript contract.
-type CanonicalTranscript struct {
-	Schema    string `json:"schema"`
-	SessionID string `json:"session_id"`
-}
-
-// ProcessedTranscript is a placeholder polished transcript contract.
-type ProcessedTranscript struct {
-	Schema    string `json:"schema"`
-	SessionID string `json:"session_id"`
-}

internal/fileops/directory.go

@@ -0,0 +1,308 @@
+package fileops
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+const (
+	promotedDirectoryMode = 0o755
+	promotedFileMode      = 0o644
+)
+
+// ErrAtomicDirectoryPromotionUnsupported indicates that the current operating
+// system lacks the atomic no-replace primitive required by PromoteDirectory.
+var ErrAtomicDirectoryPromotionUnsupported = errors.New("atomic no-replace directory promotion is unsupported")
+
+// PromoteDirectory copies an existing regular-file tree into a new directory
+// and installs the complete copy atomically. It never removes the source or
+// replaces an existing destination.
+func PromoteDirectory(src, dst string) error {
+	if err := checkAtomicDirectoryPromotionSupport(); err != nil {
+		return err
+	}
+	return promoteDirectory(src, dst, renameDirectoryNoReplace)
+}
+
+func promoteDirectory(src, dst string, install func(string, string) error) error {
+	return promoteDirectoryWithHooks(src, dst, install, sourceTraversalHooks{})
+}
+
+type sourceTraversalHooks struct {
+	afterRootInspect  func()
+	afterEntryInspect func(string)
+}
+
+func promoteDirectoryWithHooks(
+	src, dst string,
+	install func(string, string) error,
+	hooks sourceTraversalHooks,
+) error {
+	if strings.TrimSpace(src) == "" || strings.TrimSpace(dst) == "" {
+		return fmt.Errorf("source and destination directory paths are required")
+	}
+
+	sourceInfo, err := os.Lstat(src)
+	if err != nil {
+		return fmt.Errorf("inspect source directory: %w", err)
+	}
+	if !sourceInfo.IsDir() {
+		return fmt.Errorf("source path %q is not a directory", src)
+	}
+
+	if _, err := os.Lstat(dst); err == nil {
+		return fmt.Errorf("destination path %q already exists", dst)
+	} else if !errors.Is(err, os.ErrNotExist) {
+		return fmt.Errorf("inspect destination path: %w", err)
+	}
+
+	insideSource, err := pathWithin(src, dst)
+	if err != nil {
+		return err
+	}
+	if insideSource {
+		return fmt.Errorf("destination path %q must not be inside source directory %q", dst, src)
+	}
+
+	destinationParent := filepath.Dir(dst)
+	parentInfo, err := os.Lstat(destinationParent)
+	if err != nil {
+		return fmt.Errorf("inspect destination parent: %w", err)
+	}
+	if !parentInfo.IsDir() {
+		return fmt.Errorf("destination parent %q is not a directory", destinationParent)
+	}
+
+	temporary, err := os.MkdirTemp(destinationParent, "."+filepath.Base(dst)+".tmp-*")
+	if err != nil {
+		return fmt.Errorf("create temporary destination directory: %w", err)
+	}
+	removeTemporary := true
+	defer func() {
+		if removeTemporary {
+			_ = os.RemoveAll(temporary)
+		}
+	}()
+
+	sourceRoot, err := openVerifiedSourceRoot(src, sourceInfo, hooks)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = sourceRoot.Close() }()
+
+	if err := copyRegularTree(sourceRoot, src, temporary, hooks); err != nil {
+		return err
+	}
+	if err := os.Chmod(temporary, promotedDirectoryMode); err != nil {
+		return fmt.Errorf("set temporary root permissions: %w", err)
+	}
+	if err := syncDirectory(temporary); err != nil {
+		return fmt.Errorf("sync temporary root: %w", err)
+	}
+	if err := install(temporary, dst); err != nil {
+		return fmt.Errorf("install promoted directory: %w", err)
+	}
+
+	removeTemporary = false
+	if err := syncDirectory(destinationParent); err != nil {
+		return fmt.Errorf("sync destination parent: %w", err)
+	}
+	return nil
+}
+
+func openVerifiedSourceRoot(path string, inspected os.FileInfo, hooks sourceTraversalHooks) (*os.Root, error) {
+	if hooks.afterRootInspect != nil {
+		hooks.afterRootInspect()
+	}
+
+	root, err := os.OpenRoot(path)
+	if err != nil {
+		return nil, fmt.Errorf("open source directory %q: %w", path, err)
+	}
+	verified := false
+	defer func() {
+		if !verified {
+			_ = root.Close()
+		}
+	}()
+
+	opened, err := root.Stat(".")
+	if err != nil {
+		return nil, fmt.Errorf("inspect opened source directory %q: %w", path, err)
+	}
+	if !opened.IsDir() || !os.SameFile(inspected, opened) {
+		return nil, fmt.Errorf("source directory %q changed while being opened", path)
+	}
+	current, err := os.Lstat(path)
+	if err != nil {
+		return nil, fmt.Errorf("reinspect source directory %q: %w", path, err)
+	}
+	if current.Mode()&os.ModeSymlink != 0 || !current.IsDir() || !os.SameFile(opened, current) {
+		return nil, fmt.Errorf("source directory %q changed while being opened", path)
+	}
+
+	verified = true
+	return root, nil
+}
+
+func copyRegularTree(src *os.Root, sourcePath, dst string, hooks sourceTraversalHooks) error {
+	directory, err := src.Open(".")
+	if err != nil {
+		return fmt.Errorf("open source directory %q for traversal: %w", sourcePath, err)
+	}
+	defer func() { _ = directory.Close() }()
+
+	entries, err := directory.ReadDir(-1)
+	if err != nil {
+		return fmt.Errorf("read source directory %q: %w", sourcePath, err)
+	}
+	sort.Slice(entries, func(i, j int) bool {
+		return entries[i].Name() < entries[j].Name()
+	})
+
+	for _, entry := range entries {
+		entryPath := filepath.Join(sourcePath, entry.Name())
+		destinationPath := filepath.Join(dst, entry.Name())
+		info, err := src.Lstat(entry.Name())
+		if err != nil {
+			return fmt.Errorf("inspect source entry %q: %w", entryPath, err)
+		}
+
+		switch {
+		case info.Mode().IsRegular():
+			if hooks.afterEntryInspect != nil {
+				hooks.afterEntryInspect(entryPath)
+			}
+			if err := copyRegularFile(src, entry.Name(), entryPath, destinationPath, info); err != nil {
+				return err
+			}
+		case info.IsDir():
+			if hooks.afterEntryInspect != nil {
+				hooks.afterEntryInspect(entryPath)
+			}
+			if err := copyRegularDirectory(src, entry.Name(), entryPath, destinationPath, info, hooks); err != nil {
+				return err
+			}
+		default:
+			return fmt.Errorf("source entry %q has unsupported file type %s", entryPath, info.Mode().Type())
+		}
+	}
+
+	return nil
+}
+
+func copyRegularDirectory(
+	parent *os.Root,
+	name, sourcePath, dst string,
+	inspected os.FileInfo,
+	hooks sourceTraversalHooks,
+) error {
+	child, err := parent.OpenRoot(name)
+	if err != nil {
+		return fmt.Errorf("open source directory %q: %w", sourcePath, err)
+	}
+	defer func() { _ = child.Close() }()
+
+	opened, err := child.Stat(".")
+	if err != nil {
+		return fmt.Errorf("inspect opened source directory %q: %w", sourcePath, err)
+	}
+	if !opened.IsDir() || !os.SameFile(inspected, opened) {
+		return fmt.Errorf("source directory %q changed while being copied", sourcePath)
+	}
+	current, err := parent.Lstat(name)
+	if err != nil {
+		return fmt.Errorf("reinspect source directory %q: %w", sourcePath, err)
+	}
+	if current.Mode()&os.ModeSymlink != 0 || !current.IsDir() || !os.SameFile(opened, current) {
+		return fmt.Errorf("source directory %q changed while being copied", sourcePath)
+	}
+
+	if err := os.Mkdir(dst, promotedDirectoryMode); err != nil {
+		return fmt.Errorf("create destination directory %q: %w", dst, err)
+	}
+	if err := copyRegularTree(child, sourcePath, dst, hooks); err != nil {
+		return err
+	}
+	if err := os.Chmod(dst, promotedDirectoryMode); err != nil {
+		return fmt.Errorf("set destination directory permissions %q: %w", dst, err)
+	}
+	if err := syncDirectory(dst); err != nil {
+		return fmt.Errorf("sync destination directory %q: %w", dst, err)
+	}
+	return nil
+}
+
+func copyRegularFile(
+	root *os.Root,
+	name, sourcePath, dst string,
+	inspected os.FileInfo,
+) error {
+	in, err := root.Open(name)
+	if err != nil {
+		return fmt.Errorf("open source file %q: %w", sourcePath, err)
+	}
+	defer func() { _ = in.Close() }()
+
+	opened, err := in.Stat()
+	if err != nil {
+		return fmt.Errorf("inspect opened source file %q: %w", sourcePath, err)
+	}
+	if !opened.Mode().IsRegular() || !os.SameFile(inspected, opened) {
+		return fmt.Errorf("source file %q changed while being copied", sourcePath)
+	}
+	current, err := root.Lstat(name)
+	if err != nil {
+		return fmt.Errorf("reinspect source file %q: %w", sourcePath, err)
+	}
+	if current.Mode()&os.ModeSymlink != 0 || !current.Mode().IsRegular() || !os.SameFile(opened, current) {
+		return fmt.Errorf("source file %q changed while being copied", sourcePath)
+	}
+
+	out, err := os.OpenFile(dst, os.O_WRONLY|os.O_CREATE|os.O_EXCL, promotedFileMode)
+	if err != nil {
+		return fmt.Errorf("create destination file %q: %w", dst, err)
+	}
+	closed := false
+	defer func() {
+		if !closed {
+			_ = out.Close()
+		}
+	}()
+
+	if _, err := io.Copy(out, in); err != nil {
+		return fmt.Errorf("copy source file %q: %w", sourcePath, err)
+	}
+	if err := out.Chmod(promotedFileMode); err != nil {
+		return fmt.Errorf("set destination file permissions %q: %w", dst, err)
+	}
+	if err := out.Sync(); err != nil {
+		return fmt.Errorf("sync destination file %q: %w", dst, err)
+	}
+	if err := out.Close(); err != nil {
+		return fmt.Errorf("close destination file %q: %w", dst, err)
+	}
+	closed = true
+	return nil
+}
+
+func pathWithin(parent, candidate string) (bool, error) {
+	absoluteParent, err := filepath.Abs(parent)
+	if err != nil {
+		return false, fmt.Errorf("resolve source directory: %w", err)
+	}
+	absoluteCandidate, err := filepath.Abs(candidate)
+	if err != nil {
+		return false, fmt.Errorf("resolve destination directory: %w", err)
+	}
+	relative, err := filepath.Rel(absoluteParent, absoluteCandidate)
+	if err != nil {
+		return false, fmt.Errorf("compare source and destination directories: %w", err)
+	}
+	return relative != "." && relative != ".." && !strings.HasPrefix(relative, ".."+string(filepath.Separator)), nil
+}

internal/fileops/directory_special_unix_test.go

@@ -0,0 +1,35 @@
+//go:build linux || darwin
+
+package fileops
+
+import (
+	"os"
+	"path/filepath"
+	"syscall"
+	"testing"
+)
+
+func TestPromoteDirectoryRejectsNamedPipe(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	if err := os.Mkdir(src, 0o755); err != nil {
+		t.Fatalf("Mkdir(source) error = %v", err)
+	}
+	pipe := filepath.Join(src, "events.pipe")
+	if err := syscall.Mkfifo(pipe, 0o600); err != nil {
+		t.Skipf("Mkfifo() unavailable: %v", err)
+	}
+
+	if err := PromoteDirectory(src, dst); err == nil {
+		t.Fatal("PromoteDirectory() error = nil, want named pipe rejection")
+	}
+	info, err := os.Lstat(pipe)
+	if err != nil || info.Mode()&os.ModeNamedPipe == 0 {
+		t.Fatalf("source named pipe was not preserved: info=%v err=%v", info, err)
+	}
+	if _, err := os.Lstat(dst); !os.IsNotExist(err) {
+		t.Fatalf("Lstat(destination) error = %v, want not exist", err)
+	}
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}

internal/fileops/directory_test.go

@@ -0,0 +1,501 @@
+//go:build linux || darwin || windows
+
+package fileops
+
+import (
+	"bytes"
+	"os"
+	"path/filepath"
+	"reflect"
+	"runtime"
+	"strings"
+	"testing"
+)
+
+func TestPromoteDirectoryCopiesNestedRegularTree(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "promoted")
+	mustWriteFile(t, filepath.Join(src, "z-last.txt"), []byte("last"), 0o777)
+	mustWriteFile(t, filepath.Join(src, "nested", "binary.dat"), []byte{0, 1, 2, 0xff}, 0o600)
+	mustWriteFile(t, filepath.Join(src, "a-first.txt"), []byte("first"), 0o400)
+	if err := os.Mkdir(filepath.Join(src, "empty"), 0o700); err != nil {
+		t.Fatalf("Mkdir(empty) error = %v", err)
+	}
+
+	if err := PromoteDirectory(src, dst); err != nil {
+		t.Fatalf("PromoteDirectory() error = %v", err)
+	}
+
+	wantLayout := []string{".", "a-first.txt", "empty", "nested", "nested/binary.dat", "z-last.txt"}
+	if got := treeLayout(t, dst); !reflect.DeepEqual(got, wantLayout) {
+		t.Fatalf("promoted layout = %#v, want %#v", got, wantLayout)
+	}
+	assertFileBytes(t, filepath.Join(dst, "a-first.txt"), []byte("first"))
+	assertFileBytes(t, filepath.Join(dst, "nested", "binary.dat"), []byte{0, 1, 2, 0xff})
+	assertFileBytes(t, filepath.Join(dst, "z-last.txt"), []byte("last"))
+
+	if runtime.GOOS != "windows" {
+		for _, path := range []string{dst, filepath.Join(dst, "nested"), filepath.Join(dst, "empty")} {
+			info, err := os.Stat(path)
+			if err != nil {
+				t.Fatalf("Stat(%q) error = %v", path, err)
+			}
+			if got := info.Mode().Perm(); got != promotedDirectoryMode {
+				t.Fatalf("directory mode for %q = %o, want %o", path, got, promotedDirectoryMode)
+			}
+		}
+		for _, path := range []string{filepath.Join(dst, "a-first.txt"), filepath.Join(dst, "nested", "binary.dat"), filepath.Join(dst, "z-last.txt")} {
+			info, err := os.Stat(path)
+			if err != nil {
+				t.Fatalf("Stat(%q) error = %v", path, err)
+			}
+			if got := info.Mode().Perm(); got != promotedFileMode {
+				t.Fatalf("file mode for %q = %o, want %o", path, got, promotedFileMode)
+			}
+		}
+	}
+
+	assertFileBytes(t, filepath.Join(src, "nested", "binary.dat"), []byte{0, 1, 2, 0xff})
+	assertNoMatchingTempDirectories(t, root, ".promoted.tmp-")
+}
+
+func TestPromoteDirectoryRejectsInvalidPaths(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	if err := os.Mkdir(src, 0o755); err != nil {
+		t.Fatalf("Mkdir(source) error = %v", err)
+	}
+
+	tests := []struct {
+		name string
+		src  string
+		dst  string
+	}{
+		{name: "empty source", src: "", dst: filepath.Join(root, "out-a")},
+		{name: "empty destination", src: src, dst: " "},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			if err := PromoteDirectory(test.src, test.dst); err == nil {
+				t.Fatal("PromoteDirectory() error = nil, want path validation failure")
+			}
+		})
+	}
+}
+
+func TestPromoteDirectoryRejectsExistingDestination(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(src, "value.txt"), []byte("source"), 0o644)
+	mustWriteFile(t, filepath.Join(dst, "value.txt"), []byte("existing"), 0o644)
+
+	err := PromoteDirectory(src, dst)
+	if err == nil || !strings.Contains(err.Error(), "already exists") {
+		t.Fatalf("PromoteDirectory() error = %v, want existing destination error", err)
+	}
+	assertFileBytes(t, filepath.Join(dst, "value.txt"), []byte("existing"))
+	assertFileBytes(t, filepath.Join(src, "value.txt"), []byte("source"))
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsNonDirectorySource(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source.txt")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, src, []byte("source"), 0o644)
+
+	if err := PromoteDirectory(src, dst); err == nil {
+		t.Fatal("PromoteDirectory() error = nil, want non-directory source error")
+	}
+	assertFileBytes(t, src, []byte("source"))
+	if _, err := os.Lstat(dst); !os.IsNotExist(err) {
+		t.Fatalf("Lstat(destination) error = %v, want not exist", err)
+	}
+}
+
+func TestPromoteDirectoryRejectsSymlinksWithoutFollowingThem(t *testing.T) {
+	root := t.TempDir()
+	externalFile := filepath.Join(root, "external.txt")
+	externalDirectory := filepath.Join(root, "external-directory")
+	mustWriteFile(t, externalFile, []byte("outside"), 0o644)
+	mustWriteFile(t, filepath.Join(externalDirectory, "secret.txt"), []byte("secret"), 0o644)
+
+	tests := []struct {
+		name   string
+		target string
+		link   string
+	}{
+		{name: "file", target: externalFile, link: "file-link"},
+		{name: "directory", target: externalDirectory, link: "directory-link"},
+		{name: "escaping", target: filepath.Join("..", "external.txt"), link: "escaping-link"},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			src := filepath.Join(root, "source-"+test.name)
+			dst := filepath.Join(root, "destination-"+test.name)
+			if err := os.Mkdir(src, 0o755); err != nil {
+				t.Fatalf("Mkdir(source) error = %v", err)
+			}
+			if err := os.Symlink(test.target, filepath.Join(src, test.link)); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+
+			if err := PromoteDirectory(src, dst); err == nil {
+				t.Fatal("PromoteDirectory() error = nil, want symlink rejection")
+			}
+			linkInfo, err := os.Lstat(filepath.Join(src, test.link))
+			if err != nil || linkInfo.Mode()&os.ModeSymlink == 0 {
+				t.Fatalf("source symlink was not preserved: info=%v err=%v", linkInfo, err)
+			}
+			if _, err := os.Lstat(dst); !os.IsNotExist(err) {
+				t.Fatalf("Lstat(destination) error = %v, want not exist", err)
+			}
+			assertNoMatchingTempDirectories(t, root, ".destination-"+test.name+".tmp-")
+		})
+	}
+	assertFileBytes(t, externalFile, []byte("outside"))
+	assertFileBytes(t, filepath.Join(externalDirectory, "secret.txt"), []byte("secret"))
+}
+
+func TestPromoteDirectoryRejectsSymlinkSourceRoot(t *testing.T) {
+	root := t.TempDir()
+	target := filepath.Join(root, "target")
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(target, "value.txt"), []byte("outside"), 0o644)
+	if err := os.Symlink(target, src); err != nil {
+		t.Skipf("Symlink() unavailable: %v", err)
+	}
+
+	if err := PromoteDirectory(src, dst); err == nil {
+		t.Fatal("PromoteDirectory() error = nil, want source-root symlink rejection")
+	}
+	assertFileBytes(t, filepath.Join(target, "value.txt"), []byte("outside"))
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsSourceRootReplacementBeforeOpen(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	preserved := filepath.Join(root, "source-preserved")
+	replacement := filepath.Join(root, "replacement")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(src, "value.txt"), []byte("original"), 0o644)
+	mustWriteFile(t, filepath.Join(replacement, "value.txt"), []byte("replacement"), 0o644)
+
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterRootInspect: func() {
+			if err := os.Rename(src, preserved); err != nil {
+				t.Fatalf("Rename(original source) error = %v", err)
+			}
+			if err := os.Rename(replacement, src); err != nil {
+				t.Fatalf("Rename(replacement source) error = %v", err)
+			}
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want source identity failure")
+	}
+	assertFileBytes(t, filepath.Join(preserved, "value.txt"), []byte("original"))
+	assertFileBytes(t, filepath.Join(src, "value.txt"), []byte("replacement"))
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsIdentityPreservingSourceRootSymlinkReplacement(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	preserved := filepath.Join(root, "source-preserved")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(src, "value.txt"), []byte("original"), 0o644)
+
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterRootInspect: func() {
+			if err := os.Rename(src, preserved); err != nil {
+				t.Fatalf("Rename(inspected source) error = %v", err)
+			}
+			if err := os.Symlink(filepath.Base(preserved), src); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want source-root symlink replacement failure")
+	}
+	assertFileBytes(t, filepath.Join(preserved, "value.txt"), []byte("original"))
+	assertPathIsSymlink(t, src)
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsInspectedDirectorySymlinkReplacement(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	child := filepath.Join(src, "child")
+	preserved := filepath.Join(src, "child-preserved")
+	outside := filepath.Join(root, "outside")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(child, "value.txt"), []byte("original"), 0o644)
+	mustWriteFile(t, filepath.Join(outside, "sentinel.txt"), []byte("outside"), 0o644)
+
+	replaced := false
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterEntryInspect: func(path string) {
+			if replaced || path != child {
+				return
+			}
+			replaced = true
+			if err := os.Rename(child, preserved); err != nil {
+				t.Fatalf("Rename(inspected child) error = %v", err)
+			}
+			if err := os.Symlink(filepath.Join("..", "outside"), child); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+			if err := os.Mkdir(dst, 0o755); err != nil {
+				t.Fatalf("Mkdir(concurrent destination) error = %v", err)
+			}
+			mustWriteFile(t, filepath.Join(dst, "value.txt"), []byte("concurrent"), 0o644)
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want symlink replacement failure")
+	}
+	assertFileBytes(t, filepath.Join(preserved, "value.txt"), []byte("original"))
+	assertFileBytes(t, filepath.Join(outside, "sentinel.txt"), []byte("outside"))
+	assertFileBytes(t, filepath.Join(dst, "value.txt"), []byte("concurrent"))
+	assertPathMissing(t, filepath.Join(dst, "sentinel.txt"))
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsInspectedFileIdentityMismatch(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	file := filepath.Join(src, "value.txt")
+	preserved := filepath.Join(src, "value-preserved.txt")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, file, []byte("original"), 0o644)
+
+	replaced := false
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterEntryInspect: func(path string) {
+			if replaced || path != file {
+				return
+			}
+			replaced = true
+			if err := os.Rename(file, preserved); err != nil {
+				t.Fatalf("Rename(inspected file) error = %v", err)
+			}
+			mustWriteFile(t, file, []byte("replacement"), 0o644)
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want file identity failure")
+	}
+	assertFileBytes(t, preserved, []byte("original"))
+	assertFileBytes(t, file, []byte("replacement"))
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsIdentityPreservingFileSymlinkReplacement(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	file := filepath.Join(src, "value.txt")
+	preserved := filepath.Join(src, "value-preserved.txt")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, file, []byte("original"), 0o644)
+
+	replaced := false
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterEntryInspect: func(path string) {
+			if replaced || path != file {
+				return
+			}
+			replaced = true
+			if err := os.Rename(file, preserved); err != nil {
+				t.Fatalf("Rename(inspected file) error = %v", err)
+			}
+			if err := os.Symlink(filepath.Base(preserved), file); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want file symlink replacement failure")
+	}
+	assertFileBytes(t, preserved, []byte("original"))
+	assertPathIsSymlink(t, file)
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsInspectedDirectoryIdentityMismatch(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	child := filepath.Join(src, "child")
+	preserved := filepath.Join(src, "child-preserved")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(child, "value.txt"), []byte("original"), 0o644)
+
+	replaced := false
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterEntryInspect: func(path string) {
+			if replaced || path != child {
+				return
+			}
+			replaced = true
+			if err := os.Rename(child, preserved); err != nil {
+				t.Fatalf("Rename(inspected directory) error = %v", err)
+			}
+			mustWriteFile(t, filepath.Join(child, "value.txt"), []byte("replacement"), 0o644)
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want directory identity failure")
+	}
+	assertFileBytes(t, filepath.Join(preserved, "value.txt"), []byte("original"))
+	assertFileBytes(t, filepath.Join(child, "value.txt"), []byte("replacement"))
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsIdentityPreservingDirectorySymlinkReplacement(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	child := filepath.Join(src, "child")
+	preserved := filepath.Join(src, "child-preserved")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(child, "value.txt"), []byte("original"), 0o644)
+
+	replaced := false
+	err := promoteDirectoryWithHooks(src, dst, renameDirectoryNoReplace, sourceTraversalHooks{
+		afterEntryInspect: func(path string) {
+			if replaced || path != child {
+				return
+			}
+			replaced = true
+			if err := os.Rename(child, preserved); err != nil {
+				t.Fatalf("Rename(inspected directory) error = %v", err)
+			}
+			if err := os.Symlink(filepath.Base(preserved), child); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		},
+	})
+	if err == nil {
+		t.Fatal("promoteDirectoryWithHooks() error = nil, want directory symlink replacement failure")
+	}
+	assertFileBytes(t, filepath.Join(preserved, "value.txt"), []byte("original"))
+	assertPathIsSymlink(t, child)
+	assertPathMissing(t, dst)
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryDoesNotReplaceDestinationCreatedBeforeInstall(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(src, "value.txt"), []byte("source"), 0o644)
+
+	err := promoteDirectory(src, dst, func(temporary, destination string) error {
+		if err := os.Mkdir(destination, 0o755); err != nil {
+			t.Fatalf("Mkdir(concurrent destination) error = %v", err)
+		}
+		mustWriteFile(t, filepath.Join(destination, "value.txt"), []byte("concurrent"), 0o644)
+		return renameDirectoryNoReplace(temporary, destination)
+	})
+	if err == nil {
+		t.Fatal("promoteDirectory() error = nil, want no-replace install failure")
+	}
+	assertFileBytes(t, filepath.Join(dst, "value.txt"), []byte("concurrent"))
+	assertFileBytes(t, filepath.Join(src, "value.txt"), []byte("source"))
+	assertNoMatchingTempDirectories(t, root, ".destination.tmp-")
+}
+
+func TestPromoteDirectoryRejectsDestinationInsideSource(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(src, "nested", "destination")
+	if err := os.MkdirAll(filepath.Dir(dst), 0o755); err != nil {
+		t.Fatalf("MkdirAll(destination parent) error = %v", err)
+	}
+
+	if err := PromoteDirectory(src, dst); err == nil {
+		t.Fatal("PromoteDirectory() error = nil, want nested destination rejection")
+	}
+	if _, err := os.Lstat(dst); !os.IsNotExist(err) {
+		t.Fatalf("Lstat(destination) error = %v, want not exist", err)
+	}
+}
+
+func mustWriteFile(t *testing.T, path string, data []byte, mode os.FileMode) {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatalf("MkdirAll(%q) error = %v", filepath.Dir(path), err)
+	}
+	if err := os.WriteFile(path, data, mode); err != nil {
+		t.Fatalf("WriteFile(%q) error = %v", path, err)
+	}
+}
+
+func assertFileBytes(t *testing.T, path string, want []byte) {
+	t.Helper()
+	got, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("ReadFile(%q) error = %v", path, err)
+	}
+	if !bytes.Equal(got, want) {
+		t.Fatalf("ReadFile(%q) = %v, want %v", path, got, want)
+	}
+}
+
+func assertPathMissing(t *testing.T, path string) {
+	t.Helper()
+	if _, err := os.Lstat(path); !os.IsNotExist(err) {
+		t.Fatalf("Lstat(%q) error = %v, want not exist", path, err)
+	}
+}
+
+func assertPathIsSymlink(t *testing.T, path string) {
+	t.Helper()
+	info, err := os.Lstat(path)
+	if err != nil || info.Mode()&os.ModeSymlink == 0 {
+		t.Fatalf("Lstat(%q) info = %v, error = %v, want symlink", path, info, err)
+	}
+}
+
+func treeLayout(t *testing.T, root string) []string {
+	t.Helper()
+	var layout []string
+	err := filepath.WalkDir(root, func(path string, _ os.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+		relative, err := filepath.Rel(root, path)
+		if err != nil {
+			return err
+		}
+		layout = append(layout, filepath.ToSlash(relative))
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("WalkDir(%q) error = %v", root, err)
+	}
+	return layout
+}
+
+func assertNoMatchingTempDirectories(t *testing.T, parent, prefix string) {
+	t.Helper()
+	entries, err := os.ReadDir(parent)
+	if err != nil {
+		t.Fatalf("ReadDir(%q) error = %v", parent, err)
+	}
+	for _, entry := range entries {
+		if strings.HasPrefix(entry.Name(), prefix) {
+			t.Fatalf("unexpected temporary directory residue: %s", filepath.Join(parent, entry.Name()))
+		}
+	}
+}

internal/fileops/directory_unsupported_test.go

@@ -0,0 +1,37 @@
+//go:build !linux && !darwin && !windows
+
+package fileops
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestPromoteDirectoryFailsBeforeCreatingTemporaryTree(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	if err := os.Mkdir(src, 0o755); err != nil {
+		t.Fatalf("Mkdir(source) error = %v", err)
+	}
+	if err := os.WriteFile(filepath.Join(src, "value.txt"), []byte("source"), 0o644); err != nil {
+		t.Fatalf("WriteFile(source) error = %v", err)
+	}
+
+	err := PromoteDirectory(src, dst)
+	if !errors.Is(err, ErrAtomicDirectoryPromotionUnsupported) {
+		t.Fatalf("PromoteDirectory() error = %v, want unsupported capability", err)
+	}
+	entries, readErr := os.ReadDir(root)
+	if readErr != nil {
+		t.Fatalf("ReadDir(root) error = %v", readErr)
+	}
+	for _, entry := range entries {
+		if entry.Name() == filepath.Base(dst) || strings.HasPrefix(entry.Name(), ".destination.tmp-") {
+			t.Fatalf("unsupported promotion created %q", entry.Name())
+		}
+	}
+}

internal/fileops/rename_noreplace_darwin.go

@@ -0,0 +1,11 @@
+//go:build darwin
+
+package fileops
+
+import "golang.org/x/sys/unix"
+
+func checkAtomicDirectoryPromotionSupport() error { return nil }
+
+func renameDirectoryNoReplace(src, dst string) error {
+	return unix.RenamexNp(src, dst, unix.RENAME_EXCL)
+}

internal/fileops/rename_noreplace_linux.go

@@ -0,0 +1,11 @@
+//go:build linux
+
+package fileops
+
+import "golang.org/x/sys/unix"
+
+func checkAtomicDirectoryPromotionSupport() error { return nil }
+
+func renameDirectoryNoReplace(src, dst string) error {
+	return unix.Renameat2(unix.AT_FDCWD, src, unix.AT_FDCWD, dst, unix.RENAME_NOREPLACE)
+}

internal/fileops/rename_noreplace_other.go

@@ -0,0 +1,16 @@
+//go:build !linux && !darwin && !windows
+
+package fileops
+
+import (
+	"fmt"
+	"runtime"
+)
+
+func checkAtomicDirectoryPromotionSupport() error {
+	return fmt.Errorf("%w on %s", ErrAtomicDirectoryPromotionUnsupported, runtime.GOOS)
+}
+
+func renameDirectoryNoReplace(_, _ string) error {
+	return checkAtomicDirectoryPromotionSupport()
+}

internal/fileops/rename_noreplace_supported_test.go

@@ -0,0 +1,26 @@
+//go:build linux || darwin || windows
+
+package fileops
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestRenameDirectoryNoReplacePreservesExistingDestination(t *testing.T) {
+	root := t.TempDir()
+	src := filepath.Join(root, "source")
+	dst := filepath.Join(root, "destination")
+	mustWriteFile(t, filepath.Join(src, "value.txt"), []byte("source"), 0o644)
+	mustWriteFile(t, filepath.Join(dst, "value.txt"), []byte("existing"), 0o644)
+
+	if err := renameDirectoryNoReplace(src, dst); err == nil {
+		t.Fatal("renameDirectoryNoReplace() error = nil, want existing destination failure")
+	}
+	assertFileBytes(t, filepath.Join(src, "value.txt"), []byte("source"))
+	assertFileBytes(t, filepath.Join(dst, "value.txt"), []byte("existing"))
+	if info, err := os.Stat(src); err != nil || !info.IsDir() {
+		t.Fatalf("source directory was not preserved: info=%v err=%v", info, err)
+	}
+}

internal/fileops/rename_noreplace_windows.go

@@ -0,0 +1,19 @@
+//go:build windows
+
+package fileops
+
+import "golang.org/x/sys/windows"
+
+func checkAtomicDirectoryPromotionSupport() error { return nil }
+
+func renameDirectoryNoReplace(src, dst string) error {
+	from, err := windows.UTF16PtrFromString(src)
+	if err != nil {
+		return err
+	}
+	to, err := windows.UTF16PtrFromString(dst)
+	if err != nil {
+		return err
+	}
+	return windows.MoveFileEx(from, to, 0)
+}