Prepare the v2.2.0 release

.woodpecker/verify.yml

@@ -40,7 +40,7 @@ steps:
         set -eu
         output_dir="$(mktemp -d)"
         trap 'rm -rf "$output_dir"' EXIT
-        for target in linux/amd64 linux/arm64 darwin/amd64 darwin/arm64 windows/amd64 windows/arm64; do
+        for target in linux/amd64 linux/arm64 darwin/arm64; do
           goos="${target%/*}"
           goarch="${target#*/}"
           CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" go build -o "$output_dir/narratio-$goos-$goarch" ./cmd/narratio

docs/cli.md

@@ -20,7 +20,7 @@ Top-level commands:
 - `publish <session_id>`: force-run publish.
 - `clean <session_id>` or `clean --all`: remove local work/spool state.
 - `session <subcommand>`: session helper commands.
-- `config <subcommand>`: validate, display, source-trace, or compare resolved pipeline configuration.
+- `config <subcommand>`: validate, display, source-trace, catalog, or compare resolved pipeline configuration.
 
 Session subcommands:
 
@@ -45,6 +45,7 @@ Most session-aware commands accept:
 - `--session-id <session_id>`
 - `--previous-session-id <session_id>`
 - `--profile <name>`
+- `--secrets-dir <directory>`
 
 Rules:
 
@@ -53,10 +54,13 @@ Rules:
 - if both positional `<session_id>` and `--session-id` are provided, values must match.
 - `--previous-session-id` is a strict expectation: the selected session file
   must contain the same `previous_session_id`.
-- `--profile` selects a declared pipeline profile. It may be supplied once;
-  an explicit empty or unknown value fails configuration resolution. When it is
-  omitted, a declared `default_profile` is used. The same selection applies to
-  all common-flag commands, including `regenerate-artifacts`.
+- `--profile` selects `profiles/<name>.yml` relative to the pipeline root. It
+  may be supplied once; omitted, empty, and the exact name `default` select the
+  base configuration, while an unknown named profile fails. The same selection
+  applies to all common-flag commands, including `regenerate-artifacts`.
+- `--secrets-dir` selects ConfigKit credential-file bootstrap explicitly. A
+  relative value uses caller CWD. Without the flag, a present `secrets/`
+  directory beside the resolved root pipeline is used; an absent one is skipped.
 - `clean --all` cannot be combined with campaign/session selectors.
 - notification delivery is currently limited to the configured `noop` mode; see
   the [configuration reference](./config.md#notifications).
@@ -78,19 +82,24 @@ Commands with additional positionals keep their command-specific order:
 
 ## Command Reference
 
-### `config validate`, `config show`, `config sources`, and `config diff`
+### `config validate`, `config show`, `config sources`, `config artifacts`, and `config diff`
 
 ```bash
-narratio config validate [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>]
-narratio config show [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>]
-narratio config sources [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>]
-narratio config diff <left-profile> <right-profile> [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>]
+narratio config validate [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config show [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config sources [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config artifacts [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <directory>]
+narratio config diff <left-profile> <right-profile> [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--secrets-dir <directory>]
 ```
 
 These commands resolve the selected profile, defaults, ordinary paths, and—if
 a campaign is selected—the campaign-owned party. They neither discover or load
-a session nor create a workspace, manifest, run, lock, adapter, remote
-connection, or credential environment.
+a session nor create a workspace, manifest, run, lock, adapter, or remote
+connection. Like every pipeline-loading command, they bootstrap the selected or
+conventional secrets directory before resolving configuration; `config show`
+never includes bootstrap reports or secret values in its YAML output.
+Typed credential references, including `notarius.llm_api_key`, must therefore
+resolve even though these inspection commands never construct an adapter.
 
 Campaign selection is optional for a pipeline without party-driven artifact
 families. A pipeline with `scriptorium.artifact_families` needs a selected or
@@ -101,7 +110,7 @@ Session, range, force, and artifact-execution flags are not accepted.
 `config validate` writes a concise root-path, selected-profile (or `none`), and
 effective-digest summary after successful complete validation. `config show`
 writes one deterministic, secret-free YAML document containing defaulted and
-expanded concrete configuration. It omits composition declarations, artifact
+expanded concrete configuration. It omits ConfigKit directives, artifact
 family declarations, and runtime provenance.
 
 `config sources` reports the same fully validated resolution without printing
@@ -113,8 +122,19 @@ and generated family ownership. A generated party member has
 one family record and one party record at the same logical path. The output
 never reads or prints secret values.
 
-`config diff` resolves both supplied profile names from one parsed root source
-set and compares their fully resolved, secret-free effective mappings. It does
+`config artifacts` prints the concrete catalog used by analysis, ordered by
+artifact key. Its tab-separated header is `key`, `source`, `definitions`,
+`enabled`, `output`, `dependencies`, `family`, `character`, `family_source`, and
+`party_source`. Definitions and effective dependencies are sorted comma-separated
+lists, and control characters in fields are escaped. Family-member rows identify
+their family, character, defining family file, and party file. An empty catalog
+prints only the header. Ordinary artifacts can be inspected without a campaign;
+party-driven families retain the campaign requirement. See
+[Scriptorium Artifact Entries](./config.md#scriptorium-artifact-entries) for the
+configuration layout.
+
+`config diff` resolves both supplied profile names independently and compares
+their fully resolved, secret-free effective mappings. It does
 not accept `--profile`; the two positional names must be distinct, declared
 profiles. When party-driven families are present, both profiles must resolve to
 the same selected campaign and party. Use `--campaign-file` if profile-specific
@@ -157,6 +177,9 @@ Behavior:
 - rejects unknown endpoints and a `--from` endpoint after `--through`;
 - runs `render` before `extract`; an omitted or disabled Notarius
   configuration records an explicit `notarius_disabled` self-skip;
+- resolves every selected typed credential during the initial configuration
+  load, including a Notarius credential when extract is disabled, outside the
+  selected range, or later skipped as resumable;
 - skips already-succeeded stages unless `--force` is set or a stage-specific
   resume check finds its durable result obsolete;
 - applies `--force` only to stages in the selected range;
@@ -189,6 +212,10 @@ artifacts and any prerequisites required by those targets; without
 never run. Common session/configuration options and repeatable artifact values
 pass through unchanged.
 
+The forced extraction rebuilds the complete configured Notarius bundle,
+disables Notarius checkpoint loading, and requests chunk-plan refresh.
+`--artifacts` narrows only the analysis work; it never selects Notarius lanes.
+
 Because the expansion owns `--force`, `--from`, and `--through`, callers cannot
 supply those options. The shared `run` parser reports them as duplicate
 singleton flags. The alias has no private execution options or behavior, and
@@ -231,6 +258,10 @@ Equivalent to:
 narratio run-stage analyze <session_id> --force [...common config flags]
 ```
 
+This selects only `analyze`: it does not invoke Notarius, replace extraction,
+or remove existing extract outputs. `--artifacts` narrows the forced analysis
+targets and their required analysis prerequisites.
+
 ### `publish`
 
 ```bash
@@ -320,6 +351,7 @@ Options:
 
 - `--config <pipeline.yml>`
 - `--campaign <id>` or `--campaign-file <campaign.yml>`
+- `--secrets-dir <directory>`
 - `--previous-session-id <id>`
 - `--date <YYYY-MM-DD>`
 - `--title <text>`

docs/config.md

@@ -44,13 +44,16 @@ the durable copied session input.
 
 ### Read-only effective pipeline inspection
 
-`narratio config validate`, `narratio config show`, and `narratio config
-sources` use the same `--config`, `--campaign`, `--campaign-file`, and
+`narratio config validate`, `narratio config show`, `narratio config sources`,
+and `narratio config artifacts` use the same `--config`, `--campaign`,
+`--campaign-file`, and
 `--profile` selection rules as pipeline commands, but do not select, discover,
-or load a session. They do not read credential values or create runtime state.
+or load a session. They resolve configured typed credential references as part
+of the ordinary ConfigKit load but expose only their environment-variable names;
+they do not create runtime state.
 `narratio config diff <left-profile> <right-profile>` uses the same pipeline and
-campaign selectors, resolves each named profile independently from one parsed
-root source set, and does not accept a separate `--profile` flag.
+campaign selectors, resolves each named profile with an independent ConfigKit
+load, and does not accept a separate `--profile` flag.
 
 Campaign selection is optional only when the resolved pipeline has no
 `scriptorium.artifact_families`. When families are declared, Narratio selects a
@@ -59,10 +62,12 @@ then parses the campaign-owned party and expands concrete artifacts and any
 family publish rules before validation. `config validate` prints the resulting
 root, profile, and effective digest. `config show` emits the normalized
 effective pipeline YAML, with defaults and concrete expansion included but
-composition and family declarations omitted. `config sources` prints a stable
+ConfigKit directives and family declarations omitted. This review projection is
+not promised to be a reloadable authoring document. `config sources` prints a stable
 source projection instead of effective values: root/import/profile/default
 ownership plus campaign/party and generated-family records. Canonical derived
-players trace to the party. The [CLI reference](cli.md#config-validate-config-show-config-sources-and-config-diff)
+players trace to the party. `config artifacts` lists the concrete artifact map,
+effective dependencies, and defining sources. The [CLI reference](cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
 owns command syntax and output conventions.
 
 `config diff` compares normalized field values rather than YAML text or source
@@ -76,13 +81,14 @@ difference.
 ### Migrating to the maintained bundle
 
 Use the [production/testing bundle](../examples/production-testing/pipeline.yml)
-as the complete copyable migration reference. Split stable pipeline settings
-into explicit additive imports, place production/testing differences in one
-selected overlay, and retain a production `default_profile`. Convert campaign
-rosters to [canonical party input](integrations/party.md), then express
-character work as families. Inspect the result
-with `config validate`, `config show`, and `config sources`; use `config diff`
-to review profiles before running a session. Unversioned party documents are
+as the complete copyable migration reference. Keep production settings in the
+base, split them into explicit additive imports, and place testing differences
+in the fixed `profiles/testing.yml` overlay. Put ordinary artifacts and families
+in the recommended one-definition files shown by that bundle. Convert campaign
+rosters to [canonical party input](integrations/party.md), then express character
+work as families. Inspect the result with `config validate`, `config show`,
+`config sources`, and `config artifacts`; use `config diff default testing` to
+review the overlay before running a session. Unversioned party documents are
 rejected.
 
 ### Identity segments
@@ -96,20 +102,28 @@ remote state with an unsafe legacy identity must be migrated before use.
 
 ## Validation and Merge Rules
 
-- YAML decode is strict (`KnownFields(true)`) and accepts exactly one document:
-  unknown fields or trailing documents fail load.
+- ConfigKit performs strict typed YAML decoding and accepts exactly one document;
+  unknown fields, duplicate keys, and trailing documents fail load.
 - A pipeline file may explicitly import additive YAML fragments through the
-  root-only `composition.imports` list. Imported files contribute fields to one
+  root-only `configkit.imports` list. Imported files contribute fields to one
   logical pipeline document; they do not override fields supplied by the root
   or another import.
-- A root pipeline may declare named profiles. Exactly one profile is selected
-  by an option-aware caller or by `composition.default_profile`; a caller's
-  explicit selection takes precedence. Declaring profiles without either form
-  of selection is an error.
+- An omitted profile, an empty profile option, and the exact name `default`
+  select the base configuration. Any other name selects
+  `profiles/<name>.yml` relative to the pipeline root.
+- Campaign, session, and canonical-party documents are strict single-source
+  ConfigKit schemas. They do not enable imports, profiles, or `configkit`
+  directives.
 - Configured timeout and retry-delay durations must be positive. An omitted
   artifact timeout continues to inherit its configured Scriptorium timeout.
 - Session files must be concrete; unresolved `{{ ... }}` placeholders fail load.
+- Session and party inputs inspected before typed loading are limited to 8 MiB.
 - Pipeline defaults are applied before validation.
+- Defaults apply only to absent fields. Explicit empty strings, zero values,
+  `false`, and `null` retain ConfigKit's typed meaning and are then checked by
+  Narratio domain validation; they are not silently replaced by defaults.
+  Explicit null optional stage sections stay absent, while required sections
+  fail validation.
 - Campaign and session identities must agree.
 - Required stable files (`speakers_file`, `autocorrect_file`, `glossary_file`,
   `party_file`) are campaign-owned; the first three may be overridden by the
@@ -123,22 +137,16 @@ remote state with an unsafe legacy identity must be migrated before use.
   - local (`audio_dir` or `audio_files`), or
   - S3 (`audio_s3.prefix`).
 
-### Pipeline composition
+### Pipeline imports and profiles
 
 Large pipeline configurations may be split into explicitly named fragments and
-may declare one overlay per selectable profile:
+may use fixed-path profile overlays:
 
 ```yaml
-composition:
+configkit:
   imports:
-    - config/storage.yml
-    - config/integrations.yaml
-  default_profile: production
-  profiles:
-    production:
-      overlay: profiles/production.yml
-    testing:
-      overlay: profiles/testing.yml
+    - path: config/storage.yml
+    - path: config/integrations.yaml
 
 campaigns:
   root: /usr/local/share/narratio/campaigns
@@ -152,36 +160,29 @@ Imports are resolved relative to the directory containing the root pipeline
 file and are loaded in declaration order. Narratio does not scan directories or
 infer fragments. Each import must be a confined regular `.yml` or `.yaml` file:
 absolute paths, traversal, symlinks, directories, duplicate files, and an
-import of the root pipeline itself are rejected. Only the root pipeline may
-contain `composition`; nested composition is rejected.
+import of the root pipeline itself are rejected. Imported files cannot declare
+another `configkit` control section.
 
-Composition is additive. A map may be extended by multiple files when every
+Base merging is additive. A map may be extended by multiple files when every
 leaf is distinct, but a scalar, list, or map/list/scalar kind cannot be claimed
 more than once, even when the repeated values are identical. Conflict errors
-name the full field path and every source that claimed it. The assembled YAML is
-then decoded against the normal strict pipeline schema and defaults are applied
-once.
+name the full field path and every source that claimed it. Narratio additionally
+rejects an artifact or artifact-family key declared by more than one base file,
+even when those files contribute disjoint children.
 
-Profile names are case-sensitive, non-empty, trimmed, and cannot contain
-control characters. If `profiles` is present, it must contain at least one
-entry and every entry must contain only an `overlay` path. An explicit profile
-selection overrides `default_profile`; unknown and explicitly empty selections
-fail. Narratio never selects the first profile implicitly and does not read a
-profile selection from the environment.
+Profile names are case-sensitive and use ConfigKit's safe fixed-path naming
+rules. Only the selected overlay is opened; a malformed unselected overlay does
+not affect a base or different-profile load. Maps merge recursively, overlay
+scalars replace base scalars, and overlay lists replace base lists completely.
+Explicit `false`, zero, empty strings, empty containers, and null values remain
+explicit inputs and are not refilled by defaults. Profiles cannot inherit,
+stack, import files, or declare another ConfigKit control section.
 
-Every declared overlay is resolved relative to the root pipeline directory and
-must satisfy the same confined regular-YAML-file rules as an import. Narratio
-parses every declared overlay even when it is not selected, then applies only
-the selected one. Maps merge recursively, overlay scalars replace base scalars,
-and overlay lists replace base lists completely. Explicit `false`, zero, empty
-lists, and empty maps remain meaningful. YAML null cannot delete a value, and
-kind changes are rejected. Profiles cannot inherit from or stack with other
-profiles, and overlays cannot import files or declare profiles.
-
-After composition, Narratio strictly decodes the result, applies centralized
-defaults once, resolves ordinary paths, and computes a deterministic effective
+During ConfigKit loading, Narratio applies fresh typed defaults, recorded
+entry-level defaults and normalization, read-only file-local validation, and
+ordinary path resolution before computing a deterministic effective
 configuration digest. The digest represents the normalized, secret-free
-runtime pipeline mapping; it excludes composition declarations, source
+runtime pipeline mapping; it excludes ConfigKit directives, source
 provenance, profile identity, and raw environment secret values. Equivalent
 effective mappings therefore have the same digest regardless of how fields are
 split among the root and imports.
@@ -224,15 +225,31 @@ inputs:
 ## Secrets Handling
 
 - Do not place raw secrets in YAML.
-- Use env var names in config (for example `pipeline.audita.llm_api_key_env`).
-- Optionally load credential files from `pipeline.secrets.env_dir`. Each valid
-  environment-variable filename supplies one value; trailing CR/LF is removed.
+- Credential fields contain environment-variable names, represented internally
+  as ConfigKit typed secret references (for example
+  `pipeline.notarius.llm_api_key: NOTARIUS_LLM_API_KEY`). A configured reference
+  must resolve to a nonempty environment value during load.
+- Every surviving reference is resolved during ordinary pipeline loading,
+  regardless of whether its integration is disabled, outside the selected
+  stage range, or eligible for resume. Omit the field or set it to `null` when
+  that deployment should not require the credential.
+- Every command that loads a pipeline accepts `--secrets-dir <path>`. When the
+  flag is omitted, Narratio uses a present `secrets/` directory beside the
+  resolved root `pipeline.yml`; an absent conventional directory is ignored.
+  Explicit relative paths are resolved from the caller's working directory,
+  and a missing or empty explicit selection is an error.
+- Each immediate regular file whose name is a valid environment-variable name
+  supplies one value. ConfigKit removes trailing CR/LF, permits a symlink as the
+  selected directory, rejects non-hidden symlink entries, and limits input to
+  1 MiB per file, 8 MiB total, and 1024 entries. Dot-prefixed entries (including
+  files, directories, and symlinks) are ignored and omitted from bootstrap
+  reports, but still count toward the directory entry limit.
 - An existing process environment value takes precedence over a credential file.
-- Credential directories and files must not be symlinks and must be regular,
-  bounded files (at most 8 KiB per value). On POSIX, provision the directory
-  with no group/other access (normally `0700`) and files with no group/other
-  access (normally `0600`).
-- Commands that need storage/auth load filesystem secrets before constructing adapters.
+- On POSIX, operators must provision the directory with no group/other access
+  (normally `0700`) and files with no group/other access (normally `0600`);
+  ConfigKit validates type and bounds but does not enforce permission modes.
+- Bootstrap happens once per command, before any configuration load or adapter
+  construction. Safe reports contain filenames, paths, and statuses, never values.
 
 ## Publish Configuration Summary
 
@@ -274,22 +291,19 @@ Rules:
 
 | Field | Type | Required | Default / Rule |
 | --- | --- | --- | --- |
-| `composition.imports[]` | list of strings | No | explicit additive pipeline fragments relative to the root pipeline directory; `.yml` or `.yaml` regular files only |
-| `composition.default_profile` | string | Conditional | selected when profiles exist and no caller explicitly selects one; must name a declared profile |
-| `composition.profiles.<name>.overlay` | string | Conditional | required for every declared profile; one confined `.yml` or `.yaml` overlay relative to the root pipeline directory |
+| `configkit.imports[].path` | string | No | explicit additive pipeline fragment relative to the root pipeline directory; `.yml` or `.yaml` regular files only |
 | `pipeline.workspace.root` | string | No | `/var/lib/narratio` |
 | `pipeline.workspace.cleanup_after_publish` | bool | No | `false`; selects the published run's workspace target for final cleanup and requires effective `pipeline.publish.enabled=true` and `pipeline.publish.upload_run=true` |
 | `pipeline.campaigns.root` | string | No | `/usr/local/share/narratio/campaigns` |
 | `pipeline.campaigns.default_campaign_id` | string | No | empty |
-| `pipeline.secrets.env_dir` | string | No | empty |
 | `pipeline.storage.backend` | string | No | `local`; supported values are `local` and `s3` (case-insensitive) |
 | `pipeline.storage.s3.bucket` | string | Conditional | required when backend is `s3` and S3 session-audio or publish upload is enabled |
 | `pipeline.storage.s3.root_prefix` | string | No | `dnd` |
 | `pipeline.storage.s3.region` | string | No | empty |
 | `pipeline.storage.s3.endpoint` | string | No | empty |
 | `pipeline.storage.s3.force_path_style` | bool | No | `false` |
-| `pipeline.storage.s3.access_key_id_env` | string | No | `OBJECT_STORAGE_KEY_ID` |
-| `pipeline.storage.s3.secret_access_key_env` | string | No | `OBJECT_STORAGE_KEY` |
+| `pipeline.storage.s3.access_key_id` | secret reference | Conditional | unset; must be paired with `secret_access_key` |
+| `pipeline.storage.s3.secret_access_key` | secret reference | Conditional | unset; must be paired with `access_key_id` |
 | `pipeline.spool.root` | string | No | `/var/spool/narratio` |
 | `pipeline.spool.delete_audio_after_publish` | bool | No | `false`; selects the published run's spool-audio target for final cleanup and requires effective `pipeline.publish.enabled=true` and `pipeline.publish.upload_run=true` |
 | `pipeline.cache.root` | string | No | `/var/cache/narratio` |
@@ -317,7 +331,7 @@ Rules:
 | `pipeline.seriatim.env.filler_max_duration` | float | No | unset |
 | `pipeline.audita.binary` | string | No | `audita` |
 | `pipeline.audita.timeout` | duration | No | `3h` |
-| `pipeline.audita.llm_api_key_env` | string | No | empty |
+| `pipeline.audita.llm_api_key` | secret reference | No | unset |
 | `pipeline.audita.modules[]` | list[string] | No | empty |
 | `pipeline.audita.base_url` | string | No | empty |
 | `pipeline.audita.model` | string | No | empty |
@@ -345,10 +359,14 @@ Rules:
 | `pipeline.trim.seriatim.report` | bool | No | `false` |
 | `pipeline.notarius.enabled` | bool | No | `false` |
 | `pipeline.notarius.binary` | string | No | `notarius` |
+| `pipeline.notarius.llm_api_key` | secret reference | No | unset; exact environment-variable name forwarded to Notarius |
+| `pipeline.notarius.config_profile` | string | No | unset; exact Notarius ConfigKit profile selector |
 | `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.resume` | bool | No | `false`; request compatible Notarius checkpoint reuse during ordinary executed extraction |
+| `pipeline.notarius.chunk_cache` | string | No | empty; when set, exactly `auto`, `bypass`, or `refresh` |
 | `pipeline.notarius.references` | map[string]string | No | empty; maps normalized Notarius selectors to supported prepared Narratio source IDs; maximum 256 entries |
 | `pipeline.notarius.outputs` | map | Conditional | at least one entry when enabled |
 | `pipeline.render.enabled` | bool | No | `true` |
@@ -365,6 +383,73 @@ Rules:
 | `pipeline.scriptorium.artifact_families` | map | No | empty; expands one ordinary artifact per canonical party character |
 | `pipeline.notification.mode` | string | No | `noop`; the only supported notification mode until a provider is implemented |
 
+### Notarius Configuration Profile
+
+`pipeline.notarius.config_profile` optionally selects a Notarius ConfigKit
+configuration profile when extraction executes:
+
+```yaml
+notarius:
+  config_profile: downstream-economy
+```
+
+When the field is omitted or `null`, Narratio supplies no downstream profile
+selector and preserves Notarius's existing configuration-selection behavior.
+When present, Narratio forwards the exact string to Notarius, including an
+empty string or `default`, which explicitly select Notarius's composed base
+configuration. Notarius owns selector interpretation and validation; Narratio
+does not restrict profile names, resolve profile files, or rewrite the value.
+
+This selection is independent of Narratio's own `--profile` selection. A
+Narratio overlay may explicitly select a differently named Notarius overlay,
+but Narratio never forwards its own profile name automatically. It is also
+independent of `pipeline.notarius.llm_api_key`: an overlay may change either or
+both fields according to the deployed Notarius configuration. See the
+[Notarius integration contract](./integrations/notarius.md) for invocation and
+failure behavior.
+
+### Notarius Credential
+
+`pipeline.notarius.llm_api_key` optionally names the one environment variable
+to forward to the Notarius child process:
+
+```yaml
+notarius:
+  llm_api_key: OPENROUTER_API_KEY
+```
+
+The field contains the variable name, not a credential value. That name must
+match `api_key_env` in the Notarius/PromptKit profile selected by the deployed
+Notarius pipeline; Narratio does not infer or remap provider names. Omission or
+`null` forwards nothing. A non-null selected reference must resolve at pipeline
+load time through the process environment or existing secrets-directory
+bootstrap, even when Notarius is disabled, extract is outside the requested
+range, or a completed extraction can be reused.
+
+The Notarius child receives only Narratio's minimal execution environment plus
+this exact selected name and loaded value. Other credentials are not forwarded.
+Changing the value, name, presence, or absence does not invalidate extract's
+semantic resume evidence. The public name remains part of safe effective
+inspection, so renaming it may appear in `config diff` and change the effective
+configuration digest; rotating only its value does neither.
+
+See the [Notarius integration contract](./integrations/notarius.md) for the
+delivery and redaction boundary.
+
+### Notarius Cache Controls
+
+`pipeline.notarius.resume` defaults to `false`. When true, an executed ordinary
+extraction passes `--resume`; the deployed Notarius configuration must already
+enable checkpoint recording and should select an explicit persistent checkpoint
+root. Narratio does not inspect or generate that configuration.
+
+`pipeline.notarius.chunk_cache` is empty by default, leaving Notarius's
+configured mode in control. A nonempty value must be `auto`, `bypass`, or
+`refresh` and is passed as `--chunk_cache`. Forced extraction takes precedence
+over both fields: it omits `--resume` and passes `--chunk_cache refresh`.
+Neither field changes Narratio's semantic extraction fingerprint, so changing
+them alone does not rerun an otherwise complete durable extraction.
+
 ### Notarius Reference Bindings
 
 `pipeline.notarius.references` maps a Notarius CLI selector to a prepared
@@ -394,8 +479,8 @@ It validates only selector structure and the prepared source vocabulary;
 Notarius owns target-slot declarations and media compatibility.
 
 Before extraction, Narratio resolves every binding from the current prepared
-session manifest and streams it into a verified invocation-local snapshot whose
-absolute path is passed to Notarius. Missing, unsafe, empty,
+session manifest and streams it into a verified, content-addressed session
+snapshot whose absolute path is passed to Notarius. Missing, unsafe, empty,
 changed-during-copy, or checksum-inconsistent prepared evidence fails with
 guidance to force `prepare`. Bindings are sorted by normalized selector and are
 part of extraction fingerprint and resume identity. See the
@@ -430,7 +515,7 @@ For each `pipeline.scriptorium.artifacts.<name>`:
 | Field | Type | Required | Rule |
 | --- | --- | --- | --- |
 | `enabled` | bool | No | `false` if omitted |
-| `depends_on[]` | list[string] | No | must reference configured artifact keys; no self-reference; configured graph must be acyclic |
+| `depends_on[]` | list[string] | No | explicit ordering-only edges; must reference configured artifact keys, may not self-reference, and join inferred input edges in one acyclic graph |
 | `render_debug` | bool | No | per-artifact override |
 | `prompt_id` | string | Conditional | required when artifact is enabled |
 | `profile_id` | string | No | empty |
@@ -444,7 +529,7 @@ Narratio adds `session_id=narratio-session-<session_id>` to every Scriptorium re
 Without `--artifacts`, analyze executes enabled configured artifacts. With an
 explicit `--artifacts` list, the exact named configured artifacts are the
 one-invocation targets even if their `enabled` values are false. Analyze closes
-those targets over `depends_on`: a current prerequisite is reused, while a
+those targets over the effective dependency graph: a current prerequisite is reused, while a
 stale, missing, or failed prerequisite is rebuilt before its dependent.
 Unrelated artifacts are not executed. Named targets and any prerequisite that
 may require rebuilding must therefore have valid executable fields. This
@@ -462,6 +547,15 @@ For each artifact input `pipeline.scriptorium.artifacts.<name>.inputs.<input_nam
 loading. Use the canonical `source` identifier to select the input; Narratio
 does not provide adapter-specific input passthrough fields.
 
+Each `narratio.artifact.<name>` input source automatically contributes an
+effective dependency on that configured artifact. Narratio unions these inferred
+edges with explicit `depends_on` entries, then trims, deduplicates, and sorts the
+result during final configuration resolution. Authors use `depends_on` only for
+ordering constraints that are not already expressed by an artifact input.
+Use [`narratio config artifacts`](./cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
+to inspect keys, definition sources, outputs, and effective dependencies without
+loading a session or creating runtime state.
+
 ### Scriptorium Artifact Families
 
 `pipeline.scriptorium.artifact_families` declares a shared artifact template
@@ -493,9 +587,10 @@ Their resolved values are strings. A member variable may not reuse a static
 other Scriptorium artifact.
 
 Within a family only, an input source may use
-`narratio.member_artifact.<family>`. The referenced family must be named in
-that family's `member_dependencies`; resolution rewrites the source to the
-corresponding ordinary `narratio.artifact.<family>_<character_id>` source.
+`narratio.member_artifact.<family>`. Resolution rewrites the source to the
+corresponding ordinary `narratio.artifact.<family>_<character_id>` source and
+infers the concrete member dependency. Use `member_dependencies` only for
+additional same-character ordering constraints that are not already inputs.
 This syntax is rejected in explicit artifacts and never reaches runtime stages
 or Scriptorium.
 

docs/development.md

@@ -58,7 +58,7 @@ in the [Release Procedure](release.md).
 
 Woodpecker also runs `go test -race -shuffle=on -count=3 ./...` on its scheduled
 job to expose ordering and repeatability defects. Current runners cross-compile
-for macOS and Windows, but do not provide native macOS or Windows execution.
-Those cross-builds establish compilation only, not platform-equivalent runtime
-evidence. Add native checks only when official runner labels and successful
+the supported Linux amd64 and arm64 targets and the best-effort macOS arm64
+target. The macOS cross-build establishes compilation only, not native runtime
+evidence. Add a native check only when an official runner label and successful
 native-run evidence are available.

docs/integrations/audita.md

@@ -41,7 +41,6 @@ Construction fails for invalid static config values, including:
 
 Run fails for:
 - missing required request paths;
-- missing required credential env var when configured (`llm_api_key_env`);
 - subprocess execution failure;
 - invalid processed transcript JSON (`segments` array required);
 - invalid report JSON when reporting is enabled.
@@ -53,6 +52,9 @@ Failure results still include output/log/config/exit metadata for diagnostics.
 
 ## Deterministic Behavior
 - CLI args are built from runner config + request in a fixed order.
+- A configured typed `llm_api_key` reference is resolved during pipeline load;
+  the captured value is supplied only as the child's `AUDITA_LLM_API_KEY`
+  override and is never re-read from the parent environment.
 - Generated invocation YAML (`audita.generated.v1`) is emitted when requested.
 - Manifest writes are stage-owned; adapter itself is stateless.
 

docs/integrations/notarius.md

@@ -7,14 +7,14 @@ lanes from the final trimmed Seriatim transcript. Narratio owns invocation,
 safe bundle discovery, lane selection, and its own artifact metadata. Notarius
 owns pipeline definitions, lane schemas, the receipt, and bundle formats.
 
-Canonical Notarius v0.6.0 references:
+Canonical Notarius v0.7.0 references:
 
-- [CLI reference](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/cli.md)
-- [Subprocess consumer contract](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/subprocess.md)
-- [D&D pipeline and lane contracts](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/dnd-pipeline.md)
-- [Run-result receipt](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/run-result.md)
-- [JSON output bundle](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/json-output.md)
-- [D&D spell-catalog overlay](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/dnd-spell-catalog-overlays.md)
+- [CLI reference](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/cli.md)
+- [Subprocess consumer contract](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/consumers/subprocess.md)
+- [D&D pipeline and lane contracts](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/consumers/dnd-pipeline.md)
+- [Run-result receipt](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/integrations/run-result.md)
+- [JSON output bundle](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/integrations/json-output.md)
+- [D&D spell-catalog overlay](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/integrations/dnd-spell-catalog-overlays.md)
 
 The [complete Narratio example](../../examples/pipeline.full.annotated.yml)
 records the exact current constraints for all ten D&D lanes. Treat the linked
@@ -25,19 +25,38 @@ duplicate the complete schemas.
 
 When `pipeline.notarius.enabled` is true, Narratio resolves the executable,
 configuration path, input path, output directory, and working directory to
-absolute paths. Narratio requires the Notarius v0.6.0 CLI contract when
-references are configured and invokes each binding as a separate argument
-before `--json`:
+absolute paths. Configured references require support for the `--reference`
+contract introduced in Notarius v0.6.0. Cache controls require support for the
+selected `--resume` and `--chunk_cache` flags. Notarius v0.7.0 is the tested
+cache integration baseline, not an unconditional minimum version; Narratio
+does not impose a runtime version gate. Each reference binding is a separate
+argument before `--json`. When `pipeline.notarius.config_profile` is present,
+its exact value is another argument pair before `--json`:
 
 ```text
-notarius run <pipeline_id> --config <config_path> --input <trimmed_json> --output-dir <staging_dir> [--reference <selector>=<verified_snapshot_path>]... --json
+notarius run <pipeline_id> --config <config_path> --input <trimmed_json> --output-dir <staging_dir> [--resume] [--chunk_cache <mode>] [--reference <selector>=<verified_snapshot_path>]... [--config-profile <name>] --json
 ```
 
-Reference paths are absolute invocation-local snapshots streamed from the
-manifest-verified canonical files prepared inside the current Narratio session
-workspace. Narratio verifies snapshot checksum and size before and after the
-subprocess, and passes only configured bindings, ordered lexically by normalized
-selector, as direct argument-vector entries without shell interpretation. A CLI
+Notarius v0.7.0 is a known supporting version for `--config-profile`; using the
+option requires a compatible Notarius executable, but deployments that omit the
+field acquire no additional version requirement. Omitted or null configuration
+adds no flag. Supplied empty, `default`, and named selectors are forwarded
+without normalization so Notarius can apply its own ConfigKit semantics. This
+option selects a Notarius configuration overlay, not a PromptKit profile through
+`--llm-profile`, and Narratio never substitutes its own selected profile name.
+A missing or invalid downstream overlay is an ordinary Notarius subprocess
+failure: Narratio does not retry without the selector or fall back to another
+profile.
+
+Reference paths are absolute stable snapshots streamed from manifest-verified
+canonical files prepared inside the current Narratio session workspace. They
+live at `cache/notarius/references/<sha256>/<prepared-basename>` and retain the
+full lowercase content digest. Narratio verifies snapshot checksum and size
+before and after the subprocess, reconstructs missing or corrupt regular files
+from a still-valid prepared source, and passes only configured bindings,
+ordered lexically by normalized selector, as direct argument-vector entries
+without shell interpretation. Symlinks, directories, other non-regular
+conflicts, and prepared-source mismatches fail rather than being deleted. A CLI
 binding takes precedence over a matching external path in Notarius
 configuration. Narratio never emits `--without-reference`.
 
@@ -75,9 +94,41 @@ It does not pass a Narratio session ID or run `notarius config validate`
 automatically; the configured working directory and Narratio's minimal child
 environment apply to the subprocess.
 
+Ordinary extraction maps `pipeline.notarius.resume: true` to `--resume` and a
+nonempty `pipeline.notarius.chunk_cache` to `--chunk_cache <mode>`. Forced
+extraction always omits `--resume` and supplies `--chunk_cache refresh`.
+Notarius owns checkpoint enablement, recording, identity, and both cache roots;
+Narratio neither reads nor synthesizes its configuration. Use explicit
+persistent roots in Notarius when reuse must span subprocess invocations.
+Changing a reference path/digest or other upstream identity input correctly
+causes a cold miss. Chunk refresh bypasses lookup and publishes the accepted
+replacement for the active source; it does not delete unrelated plans or
+checkpoint state.
+
+When `pipeline.notarius.llm_api_key` is absent or null, Narratio adds no
+credential to that environment. When configured, the field names an environment
+variable whose already-loaded nonempty value is forwarded under that exact
+name. Operators must use the same name as the `api_key_env` in the selected
+Notarius/PromptKit profile. Narratio does not inspect the Notarius configuration
+to infer a provider or destination name, and it does not forward unrelated
+credentials from its parent environment or secrets directory.
+
+The selected name is registered with Narratio's shared streaming redactor.
+Credential bytes written by the child to captured stdout, captured stderr, or
+stderr-derived failure diagnostics are replaced with `<redacted>`, including
+custom names not known to the built-in sensitive-name list. Narratio does not
+serialize the credential into arguments, stage metadata, manifests, or
+generated configuration.
+
+This captured-stream protection is not a filesystem sandbox. Notarius remains
+responsible for keeping credentials out of files it writes directly, including
+bundle artifacts and its own diagnostics. Narratio does not scan or rewrite
+arbitrary downstream files and cannot promise redaction of transformed or
+encoded credentials.
+
 ## Accepted Result
 
-Narratio's supported invocation baseline is Notarius v0.6.0. The accepted
+Narratio's tested invocation baseline is Notarius v0.7.0. The accepted
 receipt remains `notarius.run-result.v2`; reference flags do not change the
 receipt or ten-lane output contract. The receipt
 must identify the configured pipeline, and its `index_file` must be exactly
@@ -102,6 +153,8 @@ version, and, when configured, module key. Missing, duplicate, rejected, or
 incompatible required lanes fail extraction even if Notarius exited zero. A
 configured lane whose v2 validation summary is `rejected` or `incomplete` also
 fails extraction.
+An accepted reused producer may report `producer_attempt_count: 0`; Narratio
+preserves that valid summary rather than treating it as a missing field.
 Unconfigured lanes may remain in the preserved bundle but do not become
 selectable Narratio sources.
 
@@ -130,3 +183,18 @@ 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.
+
+## Opt-in Integration Test
+
+The production-boundary cache test requires an absolute path to a locally
+built Notarius v0.7.0-compatible executable. It uses only a local
+OpenAI-compatible `httptest` provider and temporary cache/config/output roots:
+
+```sh
+NARRATIO_TEST_NOTARIUS_BINARY=/absolute/path/to/notarius \
+  GOWORK=off go test ./internal/stage \
+  -run '^TestNotariusCacheReuseIntegration$' -count=1 -v
+```
+
+When the variable is unset, the normal test suite skips this test and never
+searches `PATH`, a sibling checkout, the network, or a real provider.

docs/integrations/whisperx.md

@@ -17,11 +17,20 @@ Narratio sends an HTTP `POST` to the configured transcription URL using
 The server must return a `2xx` response whose body is valid JSON. Narratio does
 not currently require a more specific response schema at this boundary.
 
-The transcription URL must be an absolute `http` or `https` URL. The audio body
-is streamed through a fresh multipart writer for every attempt, so its memory
-use is bounded by the transport buffer rather than by the complete audio file.
+The transcription URL must be an absolute `http` or `https` URL. Each
+application attempt creates one bounded multipart framing and streams the audio
+from an open file between that framing; it does not buffer the complete audio
+or spool a multipart copy. The file field precedes the language field.
 WhisperX response acquisition is capped at 10 MiB.
 
+The initial request body and any transport-requested replay use the same
+multipart boundary and framing. Every body opens an independent audio handle
+and replay starts at the beginning. Narratio closes all bodies created for the
+attempt on response, failure, timeout, or cancellation, including when a
+transport abandons an earlier body. Genuine source read/close failures remain
+request failures; deliberate early closure of an abandoned original does not
+poison a valid replay.
+
 ## Request And Result Contract
 
 Each adapter request identifies a speaker, a readable audio file, and the
@@ -43,6 +52,11 @@ transcript output.
 - Narratio waits the configured retry delay between attempts and aborts that
   wait when the parent context is canceled.
 
+Transport replay occurs inside one application attempt and does not consume an
+entry from the configured retry budget or increment the reported attempt
+count. The standard HTTP redirect policy remains in effect; body replay can
+also support body-preserving redirects.
+
 ## Validation And Failure Semantics
 
 Client construction rejects a missing or non-HTTP(S) absolute transcription URL,
@@ -55,12 +69,19 @@ 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.
 
+Replay makes a fresh request body available to the HTTP transport; it does not
+provide an exactly-once server guarantee and does not prove recovery from every
+proxy or protocol-level failure. A server may have received work before the
+client learns that a request failed. WhisperX operations should therefore be
+safe to retry for the same input when practical.
+
 ## 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.
+parallel, bounded by the configured concurrency. Each validated speaker is
+checkpointed independently; any speaker failure still fails the aggregate
+stage, while committed earlier speakers remain reusable on an ordinary retry.
 
 ## Related Canonical Docs
 

docs/internal/adapters.md

@@ -65,6 +65,14 @@ Notarius is composed only when extraction is enabled; the extract stage owns
 prepared reference resolution, receipt, bundle, and configured-lane policy.
 The adapter validates the ordered selector/absolute-path pairs and is the sole
 owner of serializing them as repeated `--reference` arguments before `--json`.
+It also serializes an effective resume request as `--resume` and a nonempty
+validated chunk mode as `--chunk_cache <mode>`. The extract stage, not the
+adapter, resolves force precedence before constructing the request.
+Its request may also carry one already-resolved opaque ConfigKit secret
+reference. Only the Notarius adapter reads that value, rejects an unresolved or
+empty programmatic reference, and maps the exact configured name/value into the
+shared subprocess override and sensitive-name fields. It never rereads the
+parent environment. A nil reference preserves the previous invocation.
 
 Application-owned object-store construction loads configured filesystem
 secrets before invoking the invocation's storage factory. Production commands
@@ -78,15 +86,15 @@ in-memory backend without changing package state.
 - Subprocess adapters persist stage logs/generated configs through stage-managed paths.
 - Shared subprocess execution keeps private lifecycle mechanics separate from
   Narratio's environment, diagnostics, redaction, and error policy. It starts
-  an owned process group on Linux/macOS or a
-  kill-on-close job object on Windows. Every terminal path disposes of that
-  owned tree before returning. After a natural leader exit, Unix checks for
-  remaining group members and uses bounded graceful then forceful termination;
-  Windows closes the job so kill-on-close applies. Cancellation, deadlines, and
+  an owned process group on Linux and macOS. Every terminal path disposes of
+  that owned tree before returning. After a natural leader exit, the runner checks
+  for remaining group members and uses bounded graceful then forceful termination.
+  Cancellation, deadlines, and
   diagnostic limits use the same terminal disposal path without losing their
   original result classification. Child environments contain only the execution
-  baseline and adapter-specified values; configured credentials are explicit
-  sensitive values. Stdout and stderr are redacted while streaming into separate
+  baseline and adapter-specified values; configured credentials are resolved
+  once by ConfigKit and passed explicitly as sensitive values. Adapters do not
+  re-read their named environment references. Stdout and stderr are redacted while streaming into separate
   8 MiB diagnostic captures; a bounded wait closes a stream retained by a
   departed leader's descendant. Unsupported platforms reject owned command
   execution.

docs/internal/artifacts.md

@@ -58,10 +58,12 @@ Configured definitions are always registered. Without an explicit selection,
 the effective analyze set contains enabled definitions. With `--artifacts`, the
 exact named configured definitions become the effective set for that invocation,
 regardless of their `enabled` value. The effective-set resolver itself does not
-expand dependencies; the analyze work planner closes those targets over their
-configured prerequisite graph. Availability is separate from executability.
-Configuration may normalize a family selection into its concrete generated
-members before this resolver runs. The effective set retains optional family
+expand dependencies. Configuration has already materialized the sorted union of
+explicit ordering edges and configured-artifact input edges, and the analyze
+work planner closes targets over that one concrete prerequisite graph.
+Availability is separate from executability. Configuration may normalize a
+family selection into its concrete generated members before this resolver runs.
+The effective set retains optional family
 and character origin metadata, but its keys, catalog sources, and runtime
 lookups remain concrete configured-artifact identities.
 Family publish policies are likewise expanded into ordinary configured-source

docs/internal/configuration.md

@@ -8,30 +8,21 @@ pipeline-loading boundary implemented by `internal/config`.
 
 `LoadPipeline` assembles and validates a pipeline in this order:
 
-1. Parse the root YAML into a presence-aware composition tree. The tree retains
-   source names, full field paths, node kinds, declaration order, and explicit
-   zero, false, empty-map, and empty-list values.
-2. Decode the root-only `composition` envelope, validate its declaration shapes
-   and cross-field profile semantics, then clone the root without that envelope
-   as the ordinary pipeline body. The parsed source document is not mutated. A
-   load option retains the difference between omitted and explicitly empty
-   profile selection.
-3. Open each import relative to the root pipeline directory through the
-   confined regular-file boundary. Imports must use a `.yml` or `.yaml`
-   extension and cannot traverse, use symlinks, repeat a file, import the root,
-   or contain another composition envelope.
-4. Resolve and structurally parse every declared profile overlay through the
-   same confined regular-file boundary. Missing or malformed unselected
-   overlays fail the load. Overlays cannot contain a composition envelope.
-5. Additively merge the root body and imports. Distinct map leaves compose;
-   repeated scalar or list paths and node-kind disagreements are conflicts.
-6. Select exactly one declared profile from an explicit option or the default,
-   then recursively merge its overlay. Overlay leaves replace base leaves,
-   lists are atomic replacements, and null or kind changes fail.
-7. Emit deterministic canonical YAML and strictly decode it into
-   `PipelineConfig`.
-8. Apply pipeline defaults once, resolve ordinary relative pipeline paths from
-   the root pipeline file, and digest the normalized effective mapping.
+1. Give ConfigKit a fresh typed `PipelineConfig` default value, the root file,
+   enabled imports, and the caller's profile selection. Omitted, empty, and
+   exact `default` selections use the base; named profiles use the fixed
+   `profiles/<name>.yml` convention.
+2. ConfigKit parses the root and declared imports, additively merges base
+   sources, applies only the selected overlay, and strictly decodes the result.
+3. The definition resolver applies recorded entry-level and inherited defaults,
+   runs deterministic domain normalization, records derived publish destinations,
+   and resolves ordinary paths relative to the root pipeline file.
+4. The definition validator performs read-only checks that require only the
+   pipeline file. Checks requiring canonical-party family expansion remain at
+   the existing explicit pipeline/campaign resolution boundary.
+5. Narratio inspects public ConfigKit metadata to reject artifact and family keys
+   declared by multiple base files, adapts source ownership records, attaches
+   runtime-only state, and computes the normalized effective digest.
 
 This ordering preserves monolithic configuration behavior. Moving a field to
 an imported fragment changes its source ownership, not its path base, default,
@@ -53,25 +44,35 @@ download retain that exact pipeline object and its private provenance. Removing
 a temporary downloaded session file therefore cannot invalidate the resolved
 pipeline or campaign context.
 
+Campaign and session files use independent ConfigKit definitions with imports
+and profiles disabled. Campaign files normally load directly from their source;
+session files are first read through the 8 MiB bounded-read boundary because the
+application must reject unresolved template placeholders before passing an exact
+byte snapshot to ConfigKit. Local and remotely downloaded session bytes then use
+the same typed loader and retain the existing post-load identity expectations,
+stable-input precedence, and source-relative path semantics.
+
 The application also has a separate read-only inspection resolver for `config
-validate`, `config show`, and `config sources`. It uses the same production root/profile and
+validate`, `config show`, `config sources`, and `config artifacts`. It uses the same production root/profile and
 campaign selection functions, but never routes through session discovery,
-remote-session download, secret loading, adapter composition, workspace
+remote-session download, adapter composition, workspace
 initialization, manifest access, or cleanup. A pipeline with retained artifact
 family declarations must resolve its selected campaign before ordinary pipeline
 validation, which expands its canonical-party members and generated publish
 rules. A pipeline without those declarations may be validated by itself.
+Secret-directory bootstrap remains invocation-scoped and happens before the
+ordinary ConfigKit load because explicit typed references must resolve even for
+inspection.
 
 `MarshalEffectivePipeline` is the configuration-owned projection for `config
-show`. It serializes the typed, defaulted effective mapping through the
-deterministic composition renderer, then removes resolution-only artifact
-family declarations. The result contains no composition envelope or private
-provenance fields and has one trailing newline; commands do not marshal runtime
-objects directly.
+show`. It serializes one explicit typed safe projection with the existing YAML
+encoder. The projection has no slots for resolution-only artifact family
+declarations or private provenance. The result has one trailing newline;
+commands do not marshal runtime objects directly.
 
 `EffectivePipelineSources` and `EffectiveCampaignSources` provide the separate
 safe provenance projection for `config sources`. Pipeline ownership begins with
-the complete logical field paths retained during composition and classifies
+the complete logical field paths projected from the schema and classifies
 each contributor as root, import, profile, or centralized default. The
 projection replaces generated concrete member paths with paired family and
 canonical-party records, and does the same for generated publish rules.
@@ -79,12 +80,18 @@ Campaign records identify campaign-owned fields and party inputs; canonical
 derived players point to the party source. The application command only joins these sorted
 records with selection metadata and never reparses configuration files.
 
-`config diff` uses a paired profile loader that parses the root, imports, and
-declared overlays once, then clones the additive base before independently
-selecting, decoding, defaulting, and finalizing each profile. When campaign
+`EffectiveArtifactCatalog` projects the same finalized concrete artifact map for
+`config artifacts`. It derives source identity from each map key, carries sorted
+effective dependencies, aggregates ConfigKit definition contributors, and adds
+family/character/party provenance for generated members. It does not inspect
+files named after artifacts or infer another runtime graph.
+
+`config diff` uses a thin paired loader that performs two independent ConfigKit
+loads. When campaign
 resolution is needed, the command loads one selected campaign and party and
 expands both effective pipelines from that same party value. The configuration
-owner projects each normalized effective mapping into sorted logical paths;
+owner projects each normalized effective mapping directly from that same typed
+projection into sorted logical paths using JSON number preservation;
 mapping leaves are compared individually while sequence values remain atomic.
 Values are compact deterministic JSON representations for command output, not
 raw YAML fragments, ownership records, or secret material. A differing digest
@@ -98,13 +105,16 @@ configured roster authority; player and session-party fields are unsupported.
 
 ## Canonical Party Domain
 
-`ParseParty` is the package-owned boundary for parsing a party source. It
-requires a top-level `schema_version` and strictly validates the
-`narratio.party.v1` contract into ordered character domain values. The
-canonical value retains a separate exact byte copy of its source so consumers
-can materialize the authored party document without reserializing it. Its
-`PlayersYAML` method deterministically derives the versioned players-only
-projection.
+`ParseParty` is the package-owned boundary for parsing a party source. Its
+single-source ConfigKit definition disables imports and profiles, requires
+`schema_version` through `VersionRule`, and strictly decodes the
+character-keyed typed schema before applying Narratio's party-domain checks.
+Successful metadata coordinates recover authored character-map order without a
+second YAML parse; class and alias slice order comes directly from typed decode.
+The canonical value retains a separate exact byte copy of its source so
+consumers can materialize the authored party document without reserializing it.
+Its `PlayersYAML` method deterministically derives the ID-sorted, versioned
+players-only projection.
 
 An unversioned source is rejected before resolution, family expansion, or
 prepare-stage materialization.
@@ -117,58 +127,67 @@ for core pipeline/session rules, external adapters, publish/storage, Notarius,
 Scriptorium, and shared scalar/path checks. These are package-local ownership
 boundaries rather than extension points or a general validation framework.
 
-Validation fails fast and does not define which of several independent invalid
-fields is reported first. Map-backed declarations are nevertheless traversed
-deterministically. Notarius validates scalar policy, builds normalized reference
-bindings and output lanes independently, and installs all candidate values only
-after every check succeeds. Scriptorium validates artifact declarations and
-outputs, dependencies and inputs, variables, referenced outputs, and the final
+Validation is read-only, fails fast, and does not define which of several
+independent invalid fields is reported first. Map-backed declarations are
+nevertheless traversed deterministically. Reusable domain normalization is an
+explicit preceding operation for both loaded and programmatically constructed
+pipelines. Notarius normalization commits canonical scalar, reference, and output
+values only after the complete normalization succeeds; its validator then checks
+that candidate without changing it. Scriptorium validates artifact declarations
+and outputs, dependencies and inputs, variables, referenced outputs, and the final
 dependency graph in deterministic artifact order.
 
-Diagnostics identify the relevant semantic field. Composition conflicts also
+Diagnostics identify the relevant semantic field. Merge conflicts also
 retain every claiming source needed to correct an additive ownership conflict.
 Complete prose, punctuation, formatting, and private validation choreography are
 not compatibility contracts.
 
 ## Diagnostics And Runtime Metadata
 
-Syntax, duplicate-key, composition, conflict, and schema failures include the
+Syntax, duplicate-key, merge, profile, and schema failures include the
 relevant source name and full field path. Additive conflicts report every
 claiming source so operators can repair the split without repeatedly
 rediscovering additional conflicts.
 
 The loaded pipeline retains private runtime metadata for the absolute root
 path, ordered imports, selected profile name and selection source, selected
-overlay, contributing sources, effective digest, and leaf ownership. Base
+overlay, original ConfigKit metadata and source snapshots, effective digest,
+and leaf ownership. Base
 leaves retain their root/import owners, replaced leaves belong to the selected
 overlay, and centrally supplied values use the synthetic `default` owner. This
 metadata does not participate in YAML decoding or alter the public
 configuration model.
 
-The effective digest is SHA-256 over deterministic canonical YAML produced from
-the defaulted `PipelineConfig`. Runtime Notarius paths remain absolute for
+The effective digest is SHA-256 over a versioned deterministic JSON encoding of
+the same typed safe projection used by show and diff. Runtime Notarius paths remain absolute for
 execution, but the digest substitutes their normalized logical values captured
 before root-relative resolution, so relocating an equivalent configuration
-bundle does not change provenance. Because composition and resolution metadata
+bundle does not change provenance. Because ConfigKit and resolution metadata
 are private, the digest excludes source layout, profile name, and ownership.
-Configuration stores environment variable names rather than resolving raw
-credentials, so raw secret values are neither loaded nor hashed.
+Configuration stores credentials as ConfigKit typed references. A successful
+load resolves their values inside the opaque typed boundary, while this safe
+projection serializes only `EnvName()` identities. Credential-value changes are
+therefore never hashed; reference-name changes remain ordinary provenance.
+The optional Notarius credential follows the same path even for disabled or
+resumable extraction; normalization copies the opaque reference without reading
+or resolving it again. Direct YAML serialization has a configuration-owned safe
+projection so only the reference name can leave this boundary.
 `recomputePipelineEffectiveDigest` is the single package-owned refresh point
 for later runtime expansion.
 
 ## Test Surfaces
 
-`composition_test.go` protects the presence and merge algebra independently of
-the public schema. `pipeline_composition_test.go` exercises explicit imports,
-confinement, conflicts, strict decoding, metadata, and root-relative path
-behavior through `LoadPipeline`. `pipeline_profiles_test.go` covers selection,
-all-overlay validation, overlay behavior, provenance, option propagation, and
-effective-digest stability. Application configuration-loader tests protect the
+`pipeline_configkit_test.go` exercises real imports and fixed-path profiles,
+base selection, unselected-overlay isolation, source attribution, and Narratio's
+unique artifact/family base-declaration rule. Existing focused configuration
+tests protect defaults, explicit-value semantics, strict decoding, validation,
+metadata, and root-relative path behavior. Application configuration-loader tests protect the
 single-read boundary by changing the pipeline file after its initial load and
 confirming local session resolution retains the original pipeline. Other
 configuration tests continue to protect defaults and validation after assembly.
-`party_test.go` protects the versioned party schema, domain invariants, and
-deterministic players projection without involving campaign or runtime wiring.
+`party_test.go` protects the versioned typed party schema, domain invariants,
+authored character order, retained bytes, and deterministic players projection
+without involving campaign or runtime wiring.
 `party_resolution_test.go` protects campaign-owned party loading, canonical
 input restrictions, source provenance, and virtual players input selection.
 

docs/internal/manifest.md

@@ -50,6 +50,45 @@ The model admits these stage states:
 - `stale`
 - `interrupted`
 
+### Transcribe-owned speaker state
+
+A `transcribe` stage record may carry `transcribe_state_version: 1` and a
+`transcribe_speakers` map keyed by the existing derived speaker ID. The version
+is the authority marker: a versioned empty map means that no earlier speaker
+completion is reusable. Absence of both fields is accepted as legacy
+aggregate-only state so current Narratio can read older manifests.
+
+Each map entry represents one completed speaker only. It binds a versioned
+fingerprint to the session-relative audio path, audio SHA-256 and size, one
+canonical `transcript_raw` output with its path, SHA-256 and size, and the
+producer run ID. Paths must be canonical and session-relative, checksums must
+be lowercase SHA-256 values, and the record and output producer identities
+must agree. Other stages cannot carry this transcribe-owned state.
+Audio size is nonnegative; WhisperX determines whether that audio can be
+transcribed. Output size remains positive and the transcript must be valid JSON.
+
+Transcribe reconciles this collection against the current prepared audio set
+and observable WhisperX semantics. Missing or corrupt regular outputs become
+work; unsafe or non-regular evidence is rejected. It saves the reconciled
+collection before HTTP dispatch and saves a detached candidate after each
+validated completion. Only after a save succeeds is the candidate installed as
+live authority. Canonical bytes produced before an unsuccessful save are not a
+completion record.
+
+Aggregate transcribe status still describes the whole invocation. A failed,
+running, stale, or interrupted aggregate may therefore retain reusable speaker
+records, while aggregate outputs are declared only after the current
+collection is complete. Ordinary retry reconciles and reuses those records.
+Force first checkpoints an empty current collection, preventing later retries
+from reviving pre-force results.
+
+Older aggregate-only transcribe success is conservatively regenerated once
+when transcribe is selected because it does not bind individual outputs to the
+audio bytes used. The session-manifest schema identifier remains
+`narratio.session-manifest.v1`; this additive state is accepted by current
+Narratio, but older strict-decoding binaries may reject manifests after these
+fields have been written.
+
 ### Analyze-owned artifact state
 
 Every `analyze` stage record carries `analyze_state_version: 1` and an
@@ -156,10 +195,10 @@ establish a current immutable snapshot.
 
 Manifest replacement provides atomic visibility, not a crash- or power-loss
 durability guarantee. Callers reload after a returned save error before retrying.
-An analyze checkpoint error leaves the prior manifest as per-artifact authority.
-The runner then attempts to persist the bounded failed/stale retry collection;
-if that terminal save also fails, its error is joined with the originating
-failure and the next invocation must reconcile from the last readable manifest.
+A transcribe or analyze checkpoint error leaves the prior manifest collection
+as authority. The runner then attempts to persist the aggregate failure; if
+that terminal save also fails, its error is joined with the originating failure
+and the next invocation must reconcile from the last readable manifest.
 
 ## Execution Semantics
 
@@ -193,11 +232,11 @@ only fields returned by the new result. Marking a record stale does not clear
 those details because resume validation and diagnosis may still require them
 before execution begins.
 
-Aggregate lifecycle clearing deliberately preserves the analyze-owned
-per-artifact collection. This lets later reconciliation replace only evaluated
-entries without erasing unrelated current results. Other stages retain their
-existing aggregate-only lifecycle behavior and are forbidden from carrying the
-analyze-specific fields.
+Aggregate lifecycle clearing deliberately preserves both the transcribe-owned
+speaker collection and the analyze-owned per-artifact collection. This lets
+later reconciliation replace only non-current entries without erasing durable
+completions. Other stages retain aggregate-only lifecycle behavior and are
+forbidden from carrying either collection.
 
 A stage may explicitly return a skipped disposition and stable reason. The
 runner persists that outcome in the session manifest, clears older outputs,
@@ -228,7 +267,7 @@ evidence and focused durable validators:
 | Work | Reuse authority | Focused owners |
 | --- | --- | --- |
 | prepare | aggregate semantic fingerprint | [prepare](stage-prepare.md) |
-| transcribe | aggregate semantic fingerprint | [transcribe](stage-transcribe.md), [WhisperX](../integrations/whisperx.md) |
+| transcribe speakers | aggregate semantic fingerprint plus per-speaker input/output reconciliation | [transcribe](stage-transcribe.md), [WhisperX](../integrations/whisperx.md) |
 | merge | aggregate semantic fingerprint | [merge](stage-merge.md), [Seriatim](../integrations/seriatim.md) |
 | polish | aggregate semantic fingerprint | [polish](stage-polish.md), [Audita](../integrations/audita.md) |
 | normalize | aggregate semantic fingerprint | [normalize](stage-normalize.md), [Seriatim](../integrations/seriatim.md) |
@@ -310,6 +349,7 @@ where the last session transition can require operator interpretation.
 - Remote commit model and readers: `internal/artifacts/remote_commit.go`,
   `internal/artifacts/current_state_commit.go`,
 - Persistence and validation: `internal/manifest/store.go`
+- Transcribe speaker state: `internal/manifest/transcribe_state.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

@@ -37,7 +37,7 @@ failed stage or required manifest save prevents later stages from starting.
 | --- | --- | --- |
 | Executable | `cmd/narratio` | Process entry, standard stream wiring, argument handoff, and exit status. |
 | Application orchestration | `internal/app` | Command dispatch, configuration selection, secret-file environment loading, production composition, session locking, planning, execution, restore, and user-facing reporting. |
-| Configuration | [`internal/config`](configuration.md) | Presence-aware root/import/profile composition, canonical party and family expansion, strict YAML loading, defaults, normalization, session templating, and validation. |
+| Configuration | [`internal/config`](configuration.md) | ConfigKit-backed pipeline imports and profiles, canonical party and family expansion, strict YAML loading, defaults, normalization, session templating, and validation. |
 | Pipeline stages | `internal/stage` | Canonical stage registry, shared stage contract, execution dependencies, and implemented stage behavior. |
 | External boundaries | `internal/adapters`, `internal/audio` | WhisperX HTTP, downstream subprocesses, notification, object storage, and S3 audio materialization behind Narratio contracts. |
 | Manifests | `internal/manifest` | Durable session progress, stage transitions, validation, and atomic persistence. |
@@ -53,11 +53,12 @@ adapters. The normative rules for these relationships remain in
 [Architecture](../policy/architecture.md).
 
 Command execution normalizes one private dependency value per invocation.
-Configuration reads, secret-directory reads, remote downloads, object-store
-construction, and restore phase functions use that value throughout the
-command. Public command entry points supply production defaults, while focused
-tests replace only the external or sequencing boundary they exercise without
-mutating package state.
+Pipeline configuration-path discovery, configuration reads, secret-directory
+reads, remote downloads, object-store construction, and restore phase functions
+use that value throughout the command. Public command entry points supply the
+canonical discovery and production defaults, while focused tests may provide
+invocation-owned discovery candidates or replace only the external or sequencing
+boundary they exercise without mutating package state.
 
 Pipeline execution and `session plan` share the same inclusive contiguous-range
 model and the same pure pre-execution decision over recorded status, force, and
@@ -105,8 +106,8 @@ so any earlier replacement can stale a prior cleanup result.
 
 ## Focused Documentation
 
-- [Configuration Internals](configuration.md): pipeline composition, import
-  confinement, field ownership, decoding, and root-relative path semantics.
+- [Configuration Internals](configuration.md): pipeline imports and profiles,
+  field ownership, decoding, and root-relative path semantics.
 - [Adapter Internals](adapters.md): external adapter boundaries, composition,
   failure behavior, and test surfaces.
 - [Artifact Internals](artifacts.md): source identities, runtime catalog,

docs/internal/stage-analyze.md

@@ -10,7 +10,9 @@ dependency order, and safely materialize validated outputs.
 - ordinary configured artifacts from `pipeline.scriptorium.artifacts`; canonical
   party artifact families have already expanded into this map during
   configuration resolution, including corresponding member dependencies and
-  rewritten member-artifact input sources
+  rewritten member-artifact input sources. Each artifact's dependency list is
+  already the normalized union of explicit ordering edges and configured-artifact
+  input edges.
 - optional selected artifact keys supplied through the stage environment
 - built-in, configured, extraction, and previous-session source references in
   artifact inputs
@@ -51,6 +53,9 @@ Supported source families:
   manifest record and durable output pass the configured-artifact evidence
   contract; an incidental or stale canonical file is unavailable.
 - validates selected artifact dependency order (cycle-safe topo ordering).
+- consumes only the concrete dependency graph finalized by configuration;
+  planning, prerequisite closure, execution, reuse, invalidation, and semantic
+  evidence do not infer a second graph from inputs.
 - resolves required/optional inputs per artifact source definition into an
   ordered semantic identity. Each identity records the configured input name,
   canonical source ID, required policy, explicit presence, source contract,

docs/internal/stage-extract.md

@@ -19,12 +19,17 @@ procedures belong in [Operations](../operations.md).
 1. resolves the final trimmed transcript from the shared artifact catalog;
 2. resolves every configured prepared reference through the shared
    manifest-authoritative identity resolver before creating run-local output;
-3. streams each verified reference into an invocation-local snapshot and
-   rejects any source change observed while copying;
+3. resolves each reference to a stable content-addressed session snapshot,
+   reusing verified bytes or reconstructing a missing/corrupt regular entry,
+   and rejects any source change observed while copying;
 4. fingerprints the byte- and provenance-bearing Notarius invocation evidence,
-   including sorted reference identities;
+   including sorted reference identities and the explicit downstream
+   configuration-profile selection;
 5. creates a run-local staging directory and invokes the injected
-   `notarius.Runner`;
+   `notarius.Runner`, carrying the optional already-resolved credential
+   reference without exposing its value, the optional exact configuration
+   profile, plus effective resume and chunk-cache controls after force
+   precedence;
 6. revalidates the reference snapshots, then validates the v2 successful
    receipt, confined index, management documents, configured required lane
    descriptors, validation summaries, and regular payload files;
@@ -41,14 +46,22 @@ Narratio run ID, the resolved trimmed-input identity, and invocation
 fingerprint. The input identity binds the exact transcript bytes, canonical
 source ID, producer stage/output/run identity, and resolution provenance.
 Reference metadata contains only selector, source ID, canonical session-relative
-path, checksum, and size; adapter requests receive selector and absolute
-invocation-local snapshot path, never payload contents. Snapshot bytes must
+path, checksum, and size; adapter requests receive selector and absolute stable
+snapshot path, never payload contents. Snapshot bytes must
 match the prepared identity both before and after Notarius runs, so a concurrent
 prepared-file replacement cannot make recorded provenance describe different
 bytes from those supplied to Notarius.
 Validation completes before the single durable rename, so a rejected result
 cannot expose a partial durable bundle.
 
+Snapshot destinations are owned by `internal/artifacts` under
+`cache/notarius/references/<sha256>/<prepared-basename>`. They contain no run ID
+or timestamp and can be reused across invocations. Identical bytes and basename
+share a verified file; changed bytes select a different digest directory.
+Unsafe path components and non-regular conflicts fail without recursive
+deletion. Snapshot paths are delivery details and never become manifest output
+evidence or prerequisites for reuse of an already complete extract bundle.
+
 Any executed extraction outcome that replaces a different effective outcome
 marks succeeded analysis and delivery dependents stale. Render is an independent
 sibling and remains current. Repeating the same disabled self-skip with no
@@ -59,17 +72,28 @@ outputs is stable and does not repeatedly invalidate dependent stages.
 Before the focused validator runs, the application compares extract's versioned
 semantic fingerprint. It covers enablement, Notarius pipeline identity, sorted
 reference selector/source mappings, sorted declared output contracts, and each
-canonical `narratio.extraction.<key>` output identity. It excludes executable,
+canonical `narratio.extraction.<key>` output identity. It also covers the
+explicit Notarius configuration-profile selection; omitted, empty, and exact
+`default` selectors share the base-selection identity. It excludes executable,
 timeout, working directory, config path, and private Notarius config contents.
+Credential presence, reference name, and resolved value are likewise
+operational-only and excluded. Ordinary command loading still resolves a
+selected reference before resume evaluation, but a credential-only change does
+not rerun a valid completed extraction or stale its dependents.
 
 `internal/stage/extract_resume.go` then permits a skip only when the existing
 stage record still matches the current byte- and provenance-bearing invocation
 evidence. That evidence covers the current direct trimmed-transcript identity,
 sorted prepared-reference identities, pipeline identity, and configured output
-contracts. The same reference helper and transcript identity are resolved again
-for artifact evidence, so changing current transcript bytes, reference bytes,
-or producer identity makes the prior extraction obsolete. Operational runner
-settings do not invalidate otherwise current durable evidence.
+contracts, plus the same downstream profile-selection identity. The same
+reference helper and transcript identity are resolved again for artifact
+evidence, so changing current transcript bytes, reference bytes, producer
+identity, or a named downstream selection makes the prior extraction obsolete.
+Operational runner settings do not invalidate otherwise current durable
+evidence. This includes `resume` and `chunk_cache`; changing them alone cannot
+make a complete extraction non-resumable. If extraction executes, ordinary
+runs forward the configured values, while force disables resume and selects
+`refresh`.
 
 A valid prepared-reference change makes extraction non-resumable. Missing,
 unsafe, or checksum-inconsistent prepared evidence is a hard validation error
@@ -82,10 +106,12 @@ contracts and provenance, regular-file status, and stored checksums. Missing or
 obsolete results are non-resumable and run again; unsafe filesystem conditions
 return an error rather than silently accepting or replacing data.
 
-Neither contract can observe files imported by Notarius configuration, profile
-contents, prompt/module definitions, or other transitive inputs. Operators must
-force extraction after changing any such private input behind a stable
-identifier.
+This sensitivity is limited to the explicit downstream selector. Narratio's
+own profile label and whole effective-configuration digest are not blanket
+extraction invalidators. Neither contract can observe files imported by
+Notarius configuration, profile contents, prompt/module definitions, or other
+transitive inputs. Operators must force extraction after changing any such
+private input behind a stable identifier.
 
 ## Failure Behavior
 
@@ -110,9 +136,13 @@ recovery.
   `internal/stage/extract_resume.go`,
   `internal/stage/extract_test.go`,
   `internal/stage/semantic_contracts_delivery.go`
+- Stable snapshot materialization and verification:
+  `internal/stage/extract_references.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`
+- Opt-in Notarius v0.7.0 production-boundary coverage:
+  `internal/stage/notarius_cache_integration_test.go`

docs/internal/stage-merge.md

@@ -17,7 +17,11 @@ Normalize raw transcript inputs and merge into base transcript via Seriatim.
 
 ## Key Behavior
 
-- discovers and validates raw transcript inputs.
+- discovers and validates raw transcript inputs. For versioned transcribe
+  state, discovery requires aggregate transcribe success and consumes only the
+  aggregate outputs that exactly match the current per-speaker collection.
+  Partial checkpoints and orphaned files are not merge inputs. Legacy/manual
+  workflows without versioned speaker state retain directory discovery.
 - normalizes each raw transcript (`seriatim.Normalize`) into run-local scratch output.
 - merges normalized inputs (`seriatim.Run`) into base transcript.
 - validates merged transcript and optional report JSON.
@@ -26,6 +30,8 @@ Normalize raw transcript inputs and merge into base transcript via Seriatim.
 ## Invariants
 
 - merge always consumes normalized forms of raw inputs.
+- a versioned transcribe collection is usable only through its complete,
+  successful aggregate output declaration.
 - base transcript must validate before stage success.
 - report output is config-gated.
 

docs/internal/stage-polish.md

@@ -35,8 +35,9 @@ validation model, module set, transcript description, output schema, selected
 external configuration path, and canonical polished-transcript identity. Module
 ordering is normalized because the configured modules form a set. Audita's
 executable, timeouts, concurrency, report and debug behavior, work retention,
-and credential environment name are operational and do not invalidate a
-successful result.
+and resolved credential value are operational and do not invalidate a
+successful result. The public credential reference name may participate in
+configuration provenance but never exposes its resolved value.
 
 Narratio can fingerprint a selected model, module, or configuration identifier,
 but it cannot inspect content that Audita privately resolves behind that stable

docs/internal/stage-transcribe.md

@@ -14,12 +14,19 @@ Generate raw per-speaker transcripts from prepared audio using WhisperX.
 
 ## Key Behavior
 
-- discovers prepared audio from manifest inputs or canonical audio directory.
-- derives the transcript identity from the prepared `.flac` filename.
-- dispatches one WhisperX request per prepared file through a bounded errgroup.
-- validates each output as JSON.
-- writes run-local outputs then materializes canonical transcript outputs only
-  after every planned request succeeds.
+- discovers prepared audio from manifest inputs or the canonical audio
+  directory and derives each speaker identity from its `.flac` filename.
+- hashes the current audio and reconciles it with the versioned per-speaker
+  completion collection before starting HTTP work.
+- reuses only records whose audio identity, observable WhisperX semantics,
+  canonical output location, JSON, checksum, and size remain current.
+- durably checkpoints the reconciled collection before dispatch, then sends
+  only missing or obsolete speakers through the bounded worker group.
+- validates each successful run-local response, rechecks the audio bytes,
+  materializes the canonical raw transcript, and checkpoints that speaker
+  before accepting another completion.
+- rebuilds aggregate stage outputs in stable speaker order from the complete
+  current collection, including records produced by earlier invocations.
 
 ## Invariants
 
@@ -28,21 +35,48 @@ Generate raw per-speaker transcripts from prepared audio using WhisperX.
 - output path returned by adapter must match requested output path.
 - an empty adapter result path means the requested path; adapters cannot select
   an alternate destination.
-- each successful output is validated before stage success, and cancellation or
-  incomplete dispatch cannot be reported as a successful result.
+- candidate collection construction, manifest save, and installation into live
+  state are serialized. Expensive hashing and HTTP work remain outside that
+  critical section.
+- a canonical file has no completion authority until its manifest checkpoint
+  succeeds. A save failure preserves the last successfully committed
+  collection for ordinary retry.
+- each successful output is validated before its checkpoint, and cancellation,
+  incomplete dispatch, or a failed aggregate save cannot be reported as
+  aggregate stage success.
+- removed audio inputs are omitted from the current collection and aggregate
+  output list; reconciliation does not delete arbitrary transcript files.
+- merge cannot consume a partial versioned collection: it requires transcribe
+  aggregate success and an exact match between current speaker records and
+  declared aggregate outputs.
 
 ## Resume Evidence
 
-Transcribe records a versioned semantic-configuration fingerprint containing
-the Narratio-visible WhisperX service URL and recognition language. Changes to
-either rerun transcription and stale its fixed descendants while leaving
-prepare reusable. Retry count/delay, concurrency, timeout, credentials, and
-diagnostic locations are operational and do not change this evidence.
+Each completed speaker record binds the speaker ID, session-relative audio
+path, streamed audio checksum and size, the Narratio-visible WhisperX service
+URL and recognition language, and the canonical transcript path, checksum,
+size, and producer run. Changed audio bytes select that speaker for replacement;
+a changed URL or language selects every speaker. Either case stales fixed
+descendants through the normal runner decision. Retry count/delay, concurrency,
+timeout, credentials, and diagnostic locations are operational and do not
+change this evidence.
+
+An ordinary retry reuses every valid speaker checkpoint even when aggregate
+transcribe status is failed, running, or stale. A succeeded stage is skipped only
+when its stored speaker set and aggregate output evidence also match the current
+collection. Otherwise, execution repairs those records without retranscribing
+reusable speakers and stales dependents through the normal runner path.
+`--force` first durably replaces
+the collection with an empty versioned collection, so a later ordinary retry
+cannot resurrect pre-force results; speakers checkpointed by that forced
+attempt remain reusable.
 
 WhisperX models or private service configuration not exposed by Narratio's
 adapter contract cannot be fingerprinted; use `--force` after changing them.
-An existing successful transcribe record without evidence reruns once when
-selected.
+An older aggregate-only transcribe record has no trustworthy audio-to-output
+binding. When transcribe is next selected, every current speaker is therefore
+generated once and recorded in the versioned collection; later invocations can
+reuse it normally.
 
 ## Related Contracts And Tests
 
@@ -51,6 +85,9 @@ selected.
 - [Configuration](../config.md#pipeline) owns concurrency and other
   operator-selected values.
 - Implementation and tests: `internal/stage/transcribe.go`,
+  `internal/stage/transcribe_reconciliation.go`,
+  `internal/stage/transcribe_execution_state.go`,
+  `internal/stage/transcribe_resume.go`,
   `internal/stage/transcribe_test.go`,
-  `internal/stage/semantic_contracts_initial.go`, and
-  `internal/stage/semantic_contracts_initial_test.go`
+  `internal/stage/transcribe_reconciliation_test.go`, and
+  `internal/stage/transcribe_checkpoint_test.go`

docs/internal/workspace.md

@@ -37,6 +37,12 @@ validates and renames the external bundle to the unique immutable Notarius
 bundle path supplied by `internal/artifacts`. Exact physical paths belong in
 [Operations](../operations.md#extraction-workflow).
 
+Prepared references delivered to Notarius use session-durable paths owned by
+`internal/artifacts` beneath
+`cache/notarius/references/<sha256>/<prepared-basename>`. These verified
+content-addressed snapshots are outside run-local layout, survive ordinary
+published-run cleanup, and are excluded from publish and restore enumeration.
+
 ## Locking
 
 `artifacts.LocalStore` enforces the single-writer session lock via an

docs/operations.md

@@ -76,12 +76,15 @@ value from the last value persisted in the session manifest, which helps
 diagnose profile switches without changing resume authority.
 
 Before switching an operational profile, compare its effective meaning with the
-current selection through `narratio config diff <left-profile> <right-profile>`.
+production base through `narratio config diff default <profile>`.
 The command is read-only and succeeds whether it finds differences or not. Its
 sorted records describe defaulted, expanded concrete configuration—not source
 file layout—so it can be used to review model, artifact, and publish changes
-without creating a session or run. Select the same campaign explicitly when
-profiles could resolve different campaign paths; see the [CLI reference](cli.md#config-validate-config-show-config-sources-and-config-diff) for syntax and record format.
+without creating a session or run. Use `narratio config artifacts` to review the
+concrete artifact identities, definitions, outputs, and dependencies. Select the
+same campaign explicitly when profiles could resolve different campaign paths;
+see the [CLI reference](cli.md#config-validate-config-show-config-sources-config-artifacts-and-config-diff)
+for syntax and record format.
 
 ## Stage Execution and Continuation Behavior
 
@@ -113,8 +116,8 @@ Execution rules:
   perpetually rerun dependent work.
 
 Every aggregate stage except analyze currently provides semantic-configuration
-evidence; analyze retains its more precise per-artifact fingerprints and
-validator.
+evidence; transcribe additionally reconciles per-speaker completion evidence,
+while analyze retains its more precise per-artifact fingerprints and validator.
 Prepare additionally validates current stable/local/S3 source identity and the
 checksums of its durable prepared copies before reuse. Changed bytes, audio
 membership, S3 object identity, or missing/tampered copies rerun prepare and
@@ -184,7 +187,68 @@ shared session lifecycle. In particular, render does not require Notarius or
 Scriptorium, extract does not require Scriptorium, and analyze does not require
 the transcription, Seriatim, Audita, or Notarius adapters.
 
-For the common post-transcript development loop, use:
+### Transcription continuation and replacement
+
+Transcribe checkpoints each validated speaker independently. If a later
+speaker exhausts HTTP retries, the process is canceled, or the aggregate stage
+save fails, the aggregate stage is not successful and downstream merge cannot
+consume the partial set. The last successfully saved speaker records and their
+canonical `transcripts/raw/<speaker>.json` files remain durable, however.
+
+After correcting the underlying problem, retry normally:
+
+```bash
+narratio run-stage transcribe 2026-04-04
+```
+
+Narratio hashes the current prepared audio, validates recorded transcript JSON,
+checksum, size, and location, and calls WhisperX only for missing or obsolete
+speakers. A changed audio file replaces only its speaker; a changed WhisperX
+URL or language makes every current speaker non-reusable. Retry, timeout, and
+concurrency tuning do not invalidate completed speakers.
+
+Use force only when every current transcript must be replaced, including after
+a WhisperX model or private server-side configuration changes behind the same
+URL and language:
+
+```bash
+narratio run-stage transcribe 2026-04-04 --force
+```
+
+Force durably clears the reusable collection before sending requests. If that
+forced attempt fails, a later ordinary retry reuses only replacement speakers
+checkpointed after the reset; it does not revive pre-force results. Do not edit
+speaker checksums or copy older JSON into the canonical directory to influence
+reuse.
+
+The first current Narratio invocation that selects transcribe for an older
+aggregate-only manifest retranscribes every current speaker once because the
+old record cannot prove which audio bytes produced each file. Subsequent runs
+reuse the new versioned evidence normally. The overall session-manifest schema
+identifier is unchanged, but older strict-decoding Narratio binaries may reject
+a manifest after current Narratio writes the additive speaker fields; retain a
+backup or complete the upgrade before relying on downgrade.
+
+Restoring a manifest does not implicitly publish or restore every raw speaker
+file. If a recorded canonical transcript is absent locally, ordinary
+reconciliation schedules that speaker again. Removing run directories does not
+remove canonical checkpointed transcripts and does not by itself invalidate
+them.
+
+## Artifact Development Workflows
+
+Choose the narrowest workflow that matches the intended replacement:
+
+| Intent | Command | Preserved and rebuilt state |
+| --- | --- | --- |
+| Rebuild analysis only | [`narratio analyze 2026-04-04`](./cli.md#analyze) | Preserves transcripts and the current extraction bundle; force-runs selected analysis artifacts and their required analysis prerequisites. |
+| Rebuild extraction and analysis | [`narratio regenerate-artifacts 2026-04-04`](./cli.md#regenerate-artifacts) | Preserves transcripts; force-runs the complete configured extraction bundle, then selected analysis. `--artifacts` narrows analysis only. |
+| Continue bounded work normally | [`narratio run 2026-04-04 --from extract --through analyze`](./cli.md#run) | Uses ordinary resume decisions: current extraction and analysis can be skipped, while incomplete or obsolete work runs. |
+
+All three stop before publication. Inspect the results and run
+[`narratio publish`](./cli.md#publish) separately when delivery is intended.
+
+For the extraction-plus-analysis development loop, use:
 
 ```bash
 narratio regenerate-artifacts 2026-04-04
@@ -217,6 +281,30 @@ earlier checkpoints are reused and execution resumes with failed, stale, or
 missing work in dependency order. `--force` is unnecessary unless the operator
 intends to replace otherwise-current artifacts.
 
+### Developing With A Notarius Configuration Profile
+
+A Narratio development overlay can explicitly select a Notarius overlay without
+requiring the names to match. For example, `profiles/testing.yml` may contain:
+
+```yaml
+notarius:
+  config_profile: downstream-economy
+```
+
+Selecting Narratio's `testing` profile then supplies `downstream-economy` only
+when extraction executes. Switching that explicit selector makes a recorded
+extraction non-current and reruns it when `extract` is in the selected stage
+range; the usual dependent invalidation then applies. Selecting a profile does
+not create a separate workspace or retain a separate current extraction result.
+
+Narratio cannot observe edits inside Notarius configuration behind an unchanged
+selector. Force extraction after such an edit when a replacement is intended.
+Use `regenerate-artifacts` when both extraction and analysis must be rebuilt;
+use `analyze` when the existing extraction bundle should be preserved. Field
+semantics belong in [Configuration](./config.md#notarius-configuration-profile),
+and the exact downstream invocation belongs in the
+[Notarius integration contract](./integrations/notarius.md#invocation).
+
 ## Artifact Selection
 
 For a configured artifact family, selecting its family key expands to every
@@ -257,9 +345,14 @@ are not selectable or published implicitly.
 Configured Notarius references resolve only from the current manifest-backed
 prepared inputs. Their canonical locations are `inputs/party.yml`,
 `inputs/players.yml`, `inputs/glossary.yml`, and, when configured,
-`inputs/spell_catalog.json`. Extraction supplies Notarius with verified copies
-under `runs/<run_id>/extract/references/` so a concurrent refresh of canonical
-prepared files cannot change the bytes consumed by an in-flight invocation.
+`inputs/spell_catalog.json`. Extraction supplies Notarius with verified,
+content-addressed copies under
+`cache/notarius/references/<sha256>/<prepared-basename>`. The full lowercase
+SHA-256 digest identifies the reference bytes, so retries and later invocations
+reuse the same path while changed bytes select a new path. Older snapshots are
+retained and are not current artifact evidence. A concurrent refresh of
+canonical prepared files therefore cannot change the bytes consumed by an
+in-flight invocation.
 For a canonical party, preparation retains the validated authored party bytes
 at `inputs/party.yml` and generates `inputs/players.yml` from that roster.
 The manifest records their checksums separately, with the players input marked
@@ -294,6 +387,28 @@ Notarius extraction validates its run-local bundle and moves it once to the
 unique durable bundle path. A pre-existing durable destination is a collision;
 no duplicate promotion tree is created.
 
+`pipeline.notarius.resume` and `pipeline.notarius.chunk_cache` control work
+inside an executed Notarius invocation. Ordinary extraction forwards their
+configured values. Forced extraction always disables checkpoint loading and
+requests chunk-plan `refresh`, even when the configured mode is `bypass`.
+Refresh regenerates and replaces only the active input's accepted chunk plan;
+it is not cache deletion.
+
+Notarius checkpoint recording must be enabled in the deployed Notarius
+configuration. Narratio neither creates a Notarius pipeline/configuration nor
+reads it to infer checkpoint availability. Configure explicit persistent
+Notarius chunk-plan and checkpoint roots when reuse must survive separate
+Narratio subprocesses. A changed reference, transcript, selected profile,
+profile content, endpoint, or other Notarius checkpoint identity input can
+produce a legitimate cold miss.
+
+There are two independent resume boundaries: Narratio may skip a completed
+`extract` stage without invoking Notarius, while `--resume` allows Notarius to
+reuse compatible accepted work only after Narratio has decided extraction must
+execute. Deleting Narratio's local reference snapshots does not invalidate a
+complete durable extraction; missing snapshots are reconstructed only when a
+later extraction actually runs.
+
 ## External Command Lifecycle
 
 When an external command is cancelled or times out, Narratio terminates its
@@ -334,13 +449,13 @@ narratio run-stage extract 2026-04-04 --force
 Narratio automatically reruns extraction when its recorded invocation contract,
 prepared Narratio reference identities, or durable output validation changes.
 The semantic portion covers Notarius enablement, pipeline identity, declared
-reference mapping, and output contracts. Executable, timeout, working directory,
-and private config-file paths are operational and do not invalidate a current
-result.
-It cannot fingerprint configuration files, profiles, prompts, modules, or
+reference mapping, output contracts, and the explicit Notarius configuration
+profile selector. Executable, timeout, working directory, and private
+config-file paths are operational and do not invalidate a current result.
+It cannot fingerprint configuration or profile contents, prompts, modules, or
 other references loaded transitively by Notarius itself. Force extraction after
-changing any of those inputs, even when the top-level Narratio and Notarius
-config paths remain the same. A forced extract
+changing any of those inputs behind an unchanged selector, 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.
@@ -506,6 +621,7 @@ Durable session paths:
 - `logs/**`
 - `config/**`
 - `runs/**`
+- `cache/notarius/references/<sha256>/<prepared-basename>`
 
 `manifest.json` is the versioned session-progress document. Narratio accepts
 only `narratio.session-manifest.v1`. Do not edit unsupported session state to
@@ -516,6 +632,13 @@ relevant workflow requires.
 Validated Notarius bundles live below `artifacts/notarius/{run_id}/`; receipt,
 stderr, and pre-promotion output remain in the producing run's `extract`
 directory as described in [Extraction Workflow](#extraction-workflow).
+Stable reference snapshots live in the session cache path above. Ordinary
+published-run cleanup removes its selected run-local target but retains this
+cache, and publish/restore never enumerate it as an artifact. Remove a snapshot
+or digest directory manually only while no command is active for the session;
+regular missing or corrupt content is reconstructed on the next executed
+extraction, while symlinks, directories, and other unsafe conflicts must be
+investigated and removed deliberately.
 
 Run-local layout:
 
@@ -550,18 +673,28 @@ converge on `0664`. Narratio explicitly applies these modes so a restrictive
 caller umask does not remove group write or setgid. It does not change file or
 directory ownership: the configured workspace's existing group is inherited.
 
-Windows does not implement POSIX mode bits or setgid semantics. Configure the
-workspace, spool, and cache locations with an ACL that grants the collaborating
-group read/write access, and configure credential locations with an ACL limited
-to the intended credential owner. Do not use POSIX mode displays as evidence of
-Windows access control.
-
 API keys are credentials, not ordinary workspace data. Store them outside the
 shared workspace or in a separately restricted credential location; ordinary
 workspace group access must never be treated as authorization to read keys.
-On POSIX, provision a credential directory as `0700` and credential files as
-`0600`; Narratio rejects group- or other-readable configured credential paths.
-On Windows, restrict the directory and files with ACLs to the credential owner.
+Provision a credential directory as `0700` and credential files as `0600`;
+these permission modes are an operator requirement and are not enforced by
+ConfigKit.
+
+Every pipeline-loading command bootstraps credentials once, before the first
+configuration load and before any adapter is constructed. `--secrets-dir`
+selects an explicit directory (relative paths use caller CWD). Otherwise a
+present `secrets/` beside the resolved root pipeline is used, while an absent
+conventional directory is skipped. Explicit missing/empty selections and
+invalid existing directories fail the command. ConfigKit follows a symlink used
+as the selected directory, but immediate entries must be regular non-symlink
+files. Existing environment variables—including empty values—always win.
+
+S3 `access_key_id` and `secret_access_key` are an optional pair of typed
+environment references. When both are absent, the AWS SDK retains its default
+credential provider chain. When both are configured, Narratio supplies their
+load-time resolved values through an explicit static provider. Audita's optional
+`llm_api_key` is likewise captured at pipeline load and supplied only through
+the child process's `AUDITA_LLM_API_KEY` override.
 
 External adapter results are individually bounded before Narratio validates or
 materializes them. These per-file limits do not reserve disk space: prevent hard
@@ -624,4 +757,5 @@ normal migration path.
 - Publish requires prerequisite stages through `render` and `analyze` to be succeeded.
 - Markdown publish defaults require render outputs (`transcripts/final.md` and `transcripts/final.trimmed.md`).
 - Restore requires configured object storage and committed remote current state.
-- Storage-backed commands load filesystem secrets before object-store initialization.
+- Pipeline-loading commands bootstrap filesystem secrets once before configuration
+  resolution and object-store initialization.

docs/policy/architecture.md

@@ -146,7 +146,7 @@ Configuration is strict, explicit, centralized, and operator-oriented.
 - YAML decoding rejects unknown fields.
 - Defaults are centralized and testable.
 - Empty configured values do not silently replace meaningful defaults.
-- Validation rejects invalid composition before stage execution where
+- Validation rejects invalid configuration before stage execution where
   practical.
 - Root-owned imports and one selected profile resolve deterministically through
   the configuration owner; commands do not implement their own merge rules.
@@ -234,20 +234,26 @@ configuration, logs, reports, diagnostics, and Notarius artifacts—is
 intentionally shareable with the configured workspace group. API-key material
 is sensitive and is not covered by the ordinary workspace-sharing policy.
 
-On POSIX systems, Narratio-created ordinary workspace directories converge on
+Narratio supports Linux on amd64 and arm64. macOS arm64 is best effort: it has
+a cross-build gate and release asset, but no native runtime release gate.
+Windows and macOS amd64 are unsupported and have no compatibility promise or
+release asset.
+
+On supported systems, Narratio-created ordinary workspace directories converge on
 setgid `02775` and ordinary workspace files on `0664`, even when the caller's
 umask is restrictive. This preserves the existing workspace group for nested
 creation and atomic replacements without changing ownership. API-key storage
-uses a separate restrictive contract. On Windows, POSIX mode bits and setgid
-are not authoritative; operators must provide the equivalent shared-group and
-credential-restricted ACLs described in [Operations](../operations.md#workspace-permissions).
+uses a separate restrictive contract.
 
 Raw secrets must not be stored in pipeline, campaign, or session YAML or written
 to manifests, logs, generated configuration, reports, publish metadata,
-documentation, or examples. 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.
+documentation, or examples. Pipeline configuration names environment variables;
+command-owned startup may populate previously unset variables from ConfigKit's
+explicit or conventional secrets directory. This bootstrap occurs once per
+invocation before any typed configuration load or adapter construction and is
+not hidden inside library loaders. Safe bootstrap diagnostics may identify
+files and statuses but never values. Diagnostics should avoid transcript and
+prompt content unless a deliberate, bounded inspection mechanism requires it.
 
 Logs, reports, generated invocation files, generated configuration, and render
 debug files are diagnostics, not canonical pipeline products. They should be

docs/release.md

@@ -58,14 +58,14 @@ scripts/check-release-candidate.sh vMAJOR.MINOR.PATCH
 
 It validates the version and matching note, module hygiene, formatting,
 whitespace, uncached tests, race tests, static checks, documentation, examples,
-and six official cross-build assets. It uses `GOWORK=off`, does not contact
+and three official cross-build assets. It uses `GOWORK=off`, does not contact
 application services, CI, or Gitea, and does not create tags or modify tracked
 source. Fix any failure on `main`, commit it, push it normally, and rerun the
 checker.
 
-The checker builds Linux, macOS, and Windows assets for `amd64` and `arm64`.
-Cross-builds prove compilation; they are not native macOS or Windows runtime
-evidence.
+The checker builds Linux assets for `amd64` and `arm64` and a macOS asset for
+`arm64`. The macOS cross-build proves compilation; it is not native macOS
+runtime evidence.
 
 ## Publish the tag
 

docs/releases/README.md

@@ -22,5 +22,7 @@ See the [release procedure](../release.md) for creating a candidate and tag.
 When asynchronous publication succeeds, the corresponding Gitea release is the
 canonical source for downloadable binaries and checksums.
 
+- [v2.0.0](v2.0.0.md)
+- [v1.7.0](v1.7.0.md)
 - [v1.6.0](v1.6.0.md)
 - [v1.5.0](v1.5.0.md)

docs/releases/v2.0.0.md

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

docs/releases/v2.1.0.md

@@ -0,0 +1,70 @@
+# Narratio v2.1.0
+
+This release adds explicit Notarius credential forwarding and cache reuse,
+with predictable forced regeneration and clearer artifact-development guidance.
+
+## Summary
+
+Narratio can pass one typed API-key reference to Notarius and opt into reuse of
+its accepted execution checkpoints. Stable, verified reference snapshots allow
+unchanged references to retain their cache identity across Narratio invocations.
+Configuration loading now uses ConfigKit v0.3.0.
+
+## Compatibility
+
+The new `notarius.llm_api_key`, `notarius.resume`, and `notarius.chunk_cache`
+fields are optional. Existing v2 configuration needs no schema migration.
+Configured credential references must resolve during configuration loading,
+even when extraction will not execute; omitted references add no requirement.
+
+Forced extraction now omits Notarius checkpoint loading and requests chunk-plan
+refresh. This includes `regenerate-artifacts`, which remains an alias for a
+forced run from extract through analyze. `analyze` continues to force analysis
+only, preserving extraction results and transcripts.
+
+Notarius v0.7.0 is the tested cache integration baseline, not an unconditional
+minimum-version gate. The executable must support the flags selected for the
+invocation, including `--chunk_cache refresh` for forced extraction. Reference
+bindings retain the contract introduced in Notarius v0.6.0. Receipt and artifact
+schemas are unchanged; reused producers may report zero production attempts.
+
+Supported release platforms remain Linux amd64/arm64 and best-effort macOS
+arm64. macOS validation is cross-compilation, not native runtime testing.
+
+## Upgrade
+
+No special migration is required for existing v2 configuration. To use the new
+features:
+
+- Declare `notarius.llm_api_key` with the environment-variable name expected by
+  the deployed Notarius profile, supplying its value through the existing
+  environment or secrets-directory mechanism.
+- Enable checkpoint recording in Notarius configuration before setting
+  Narratio's `notarius.resume: true`. Prefer explicit persistent Notarius cache
+  roots for service deployments. An optional `notarius.chunk_cache` overrides
+  its ordinary chunk-plan policy.
+- Expect an initial checkpoint miss when moving from old run-local reference
+  paths to stable session snapshots. No migration or deletion of old caches is
+  required. Force refreshes the current input's plan, not the whole cache.
+- Use `narratio analyze SESSION` for analysis-only iteration, and
+  `narratio regenerate-artifacts SESSION` when extraction should also be rebuilt.
+
+See [Configuration](../config.md), the
+[Notarius integration contract](../integrations/notarius.md), and
+[Artifact Development Workflows](../operations.md#artifact-development-workflows)
+for current details.
+
+## Changes
+
+- Upgraded ConfigKit from v0.1.2 to v0.3.0 and isolated pipeline-discovery tests
+  from machine configuration.
+- Added optional typed Notarius credentials, exact-name environment delivery,
+  safe configuration projections, and captured-stream redaction for custom names.
+- Added Notarius resume and chunk-cache controls with force taking precedence.
+- Added verified, content-addressed session reference snapshots that survive
+  ordinary run cleanup without becoming published artifacts.
+- Preserved relative-workspace support for snapshot delivery and confinement.
+- Kept credentials and cache-policy controls outside extract's semantic resume
+  identity, so operational changes alone do not invalidate completed results.
+- Added an opt-in real Notarius integration test using a local provider to
+  verify reuse, partial-failure recovery, and forced refresh without paid calls.

docs/releases/v2.1.1.md

@@ -0,0 +1,33 @@
+# Narratio v2.1.1
+
+This release fixes configuration loading for publish rules that explicitly set
+requiredness while relying on a derived destination.
+
+## Summary
+
+Publish rules with a derivable source and omitted `dest` now load correctly
+whether `required` is omitted, explicitly true, or explicitly false. Previously,
+explicit requiredness could trigger ConfigKit's `resolve.unrecorded_change`
+diagnostic instead of accepting valid configuration.
+
+## Compatibility
+
+This is a backward-compatible bookkeeping correction. Publish schemas,
+destination derivation, requiredness defaults, and YAML syntax support are
+unchanged. Explicit destinations and explicit false values remain intact.
+No ConfigKit upgrade or publication/storage behavior change is involved.
+
+## Upgrade
+
+No special upgrade action or configuration migration is required. Operators
+need not add a derivable `dest`, remove explicit `required` values, or change
+flow-style YAML to block-style YAML. See the
+[configuration reference](../config.md#publish-configuration-summary).
+
+## Changes
+
+- Recorded derived publish destinations at ConfigKit's atomic sequence boundary
+  while retaining field-level resolution records.
+- Added real load/validation and safe effective-configuration regression coverage
+  for built-in and configured-artifact sources, explicit requiredness, and an
+  explicit destination that must remain unchanged.

docs/releases/v2.1.2.md

@@ -0,0 +1,26 @@
+# Narratio v2.1.2
+
+This release allows secrets directories to contain hidden metadata without
+preventing credential loading.
+
+## Summary
+
+ConfigKit v0.3.1 ignores dot-prefixed entries such as `.git`, `.gitignore`, and
+`.DS_Store` during secrets bootstrap. These entries no longer block otherwise
+valid credentials from loading and are omitted from bootstrap reports.
+
+## Compatibility
+
+This is a backward-compatible dependency bug fix. Configuration schemas,
+existing-environment precedence, and validation of non-hidden entries are
+unchanged. Hidden entries still count toward the directory entry limit.
+
+## Upgrade
+
+No special upgrade action or configuration migration is required. See the
+[credential configuration reference](../config.md) for secrets-directory usage.
+
+## Changes
+
+- Updated ConfigKit from v0.3.0 to v0.3.1.
+- Clarified hidden-entry handling in the secrets-directory documentation.

docs/releases/v2.2.0.md

@@ -0,0 +1,53 @@
+# Narratio v2.2.0
+
+This release adds resumable per-speaker transcription, replayable WhisperX
+uploads, and explicit Notarius configuration profile selection.
+
+## Summary
+
+Successfully checkpointed speaker transcripts survive later transcription
+failures. Ordinary retry reuses valid completed speakers and processes only the
+remaining work. WhisperX uploads can now be replayed by the HTTP transport
+without buffering whole audio files.
+
+## Compatibility
+
+Existing CLI commands, speaker identities, transcript formats, and pipeline
+ordering are unchanged. Session manifests gain optional per-speaker completion
+evidence. Older manifests remain readable, but older Narratio binaries may
+reject manifests containing the new fields; downgrading after transcription
+is not transparent.
+
+The optional `pipeline.notarius.config_profile` field selects a downstream
+Notarius configuration profile independently of Narratio's own profile. Using
+it requires a Notarius version supporting `--config-profile`; v0.7.0 is a known
+supporting version. Omitting the field retains downstream default selection.
+
+## Upgrade
+
+No configuration migration is required. When transcribe is next selected for
+an older aggregate-only manifest, all current speakers are transcribed once
+to establish reusable evidence. Previously unrecorded results from failed runs
+are not automatically imported.
+
+Use ordinary run/resume to retain valid checkpoints. Forcing transcribe resets
+reuse of prior results; an ordinary retry after a failed forced invocation can
+retain replacements already checkpointed by that attempt. See the
+[operations guide](../operations.md) and
+[Notarius profile configuration](../config.md#notarius-configuration-profile).
+
+## Changes
+
+- Persist validated speaker completions independently of aggregate stage status,
+  with serialized saves for concurrent work and input/output integrity checks.
+- Rebuild aggregate transcript references when speakers are removed or those
+  references need repair, without retranscribing otherwise-current speakers.
+- Keep removed and incomplete speaker results out of versioned merge inputs.
+- Permit checkpoint evidence for zero-byte audio when WhisperX accepts it;
+  service rejection still fails normally.
+- Provide fresh streaming multipart bodies for HTTP transport replay while
+  preserving application retries, cancellation, and resource cleanup. This
+  addresses the missing-body-replay limitation, not persistent upstream
+  protocol failures or exactly-once service execution.
+- Pass explicit Notarius configuration profiles through the CLI and include
+  their selection in extraction reuse decisions.

docs/roadmap/feature-request-notarius-config-profile.yml

@@ -0,0 +1,102 @@
+---
+title: Select a Notarius configuration profile through Narratio
+status: proposed
+consumer: rakestraw.dnd Ansible collection and its pipeline role
+
+use_case: >-
+  The Ansible pipeline role deploys Narratio and Notarius as independently
+  configured applications. Operators select Narratio's testing profile to
+  develop prompts and other pipeline assets with inexpensive models while
+  retaining production settings in the base configuration. That selection
+  currently cannot select a Notarius configuration profile, leaving extraction
+  on its production settings.
+
+desired_outcome: >-
+  Narratio's testing overlay can explicitly select Notarius's testing overlay.
+  Notarius owns the resulting model choices and any other testing-specific
+  settings. The operator selects testing once when invoking Narratio, without
+  editing production files or separately invoking Notarius.
+
+rationale: >-
+  Both applications already use ConfigKit configuration overlays. Selecting
+  the downstream overlay preserves each application's configuration ownership
+  and supports future Notarius testing adjustments without adding a Narratio
+  configuration field for every downstream setting. This is explicit profile
+  composition, not a requirement for arbitrary recursive orchestration.
+
+requested_interface:
+  field: pipeline.notarius.config_profile
+  value: Optional Notarius ConfigKit configuration profile name
+  supplied: >-
+    Invoke Notarius with --config-profile and the selected name when extraction
+    executes. The setting is available in Narratio's base configuration and
+    its ordinary configuration overlays.
+  omitted_or_null: >-
+    Do not supply a profile-selection flag. Preserve existing Notarius
+    configuration selection behavior.
+  downstream_semantics: >-
+    Retain Notarius's existing selector semantics, including the default
+    selector for its base configuration. Notarius resolves profiles relative
+    to its own configuration root, not Narratio's root.
+  explicit_selection: >-
+    Do not automatically forward Narratio's profile name. The selected
+    Notarius profile may have the same name or a different name.
+  distinction: >-
+    This selects a ConfigKit configuration overlay through --config-profile,
+    not a PromptKit LLM profile through --llm-profile. The Notarius overlay
+    can itself select the desired LLM profiles using native Notarius settings.
+
+illustrative_consumer_configuration:
+  note: >-
+    These are separate files managed by the Ansible suite, not a proposed
+    combined configuration document. The Notarius pipeline and LLM profile
+    names are illustrative; the referenced LLM profile must be available to
+    Notarius.
+  narratio_profiles_testing_yml: |
+    notarius:
+      config_profile: testing
+  notarius_profiles_testing_yml: |
+    pipelines:
+      dnd-session:
+        llm_profile: gemini-flash-latest
+  operator_workflow: >-
+    Select --profile testing on the ordinary Narratio invocation. Narratio
+    selects Notarius's testing configuration overlay when it runs extraction;
+    other Narratio testing overrides continue to apply normally.
+
+observable_behavior:
+  - >-
+    Notarius owns overlay discovery, composition, validation, and the effective
+    downstream settings. A missing or invalid selected overlay must not silently
+    fall back to production or be reported as successful extraction.
+  - >-
+    Changing the selected Notarius configuration profile must not silently
+    reuse Narratio's completed extraction from the previous selection.
+    Existing downstream invalidation behavior applies when extraction changes.
+  - >-
+    Editing Notarius overlay contents behind an unchanged selector retains
+    the existing operator responsibility to force extraction when needed.
+    This request does not require inspecting or tracking private Notarius files.
+  - >-
+    Existing credential forwarding remains independently configurable through
+    notarius.llm_api_key, including in the Narratio testing overlay when the
+    selected Notarius settings require a different credential reference.
+
+consumer_follow_up: >-
+  After Narratio releases this interface, the standalone Narratio Ansible role
+  can expose an optional variable and omit the field when unset. The suite role
+  can deploy Notarius's testing overlay and select it from Narratio's testing
+  overlay, then distribute the changes through the collection. These consumer
+  updates are separate from this upstream feature request.
+
+scope_boundaries:
+  - No Notarius change is currently required; v0.7.0 supports --config-profile.
+  - No separate Narratio-side Notarius LLM-profile selector is requested.
+  - No automatic profile-name propagation or downstream configuration rewriting.
+  - No new credential discovery, copying, or provider-name remapping.
+  - This request specifies consumer behavior, not internal Narratio code changes.
+
+references:
+  narratio_configuration: ../config.md
+  narratio_notarius_integration: ../integrations/notarius.md
+  notarius_cli: https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.7.0/docs/cli.md

docs/roadmap/fix-publish-output-destination-resolution.md

@@ -0,0 +1,171 @@
+# Bugfix Request: Preserve Publish Destination Derivation with Explicit Rules
+
+Status: Implemented; release target v2.1.1
+
+## Summary
+
+Narratio should successfully derive a publish destination when a valid
+`publish.outputs` rule omits `dest`, regardless of whether that rule explicitly
+sets `required`. Narratio v2.1.0 currently rejects this valid configuration in
+some cases with a ConfigKit `resolve.unrecorded_change` diagnostic.
+
+This is a configuration-resolution bookkeeping defect, not a YAML formatting
+restriction. The fix should preserve the existing public configuration
+contract: operators may omit a derivable destination and may independently set
+`required` to either `true` or `false`.
+
+## Observed Behavior
+
+A pipeline containing a rule equivalent to the following fails during native
+configuration loading:
+
+```yaml
+publish:
+  outputs: [{"source": "narratio.transcript.final", "required": true}]
+```
+
+Narratio reports:
+
+```text
+error [resolve.unrecorded_change] /publish/outputs: application resolution changed a value without a covering record
+```
+
+The same failure is possible with ordinary block-style YAML when `required` is
+explicit and `dest` is omitted. Flow-style YAML is valid input and is not the
+cause of the defect.
+
+The existing destination-derivation tests do not expose this case because
+their rules also omit `required`. Narratio then supplies the default value for
+`required` and records a change covering the complete `publish.outputs`
+sequence. That broader record happens to cover the destination mutation as
+well, masking the missing resolution record that occurs when `required` is
+already explicit.
+
+## Expected Contract
+
+The current configuration contract remains unchanged:
+
+- `publish.outputs[].source` is required;
+- `publish.outputs[].dest` may be omitted when Narratio can derive it from the
+  selected built-in or configured artifact source;
+- `publish.outputs[].required` defaults to `true`, but an operator may set it
+  explicitly to either `true` or `false`; and
+- equivalent valid YAML representations have equivalent configuration
+  semantics.
+
+For example, both rules below should load and validate, retain their explicit
+requiredness, and receive the canonical derived destination:
+
+```yaml
+publish:
+  outputs:
+    - source: narratio.transcript.final
+      required: true
+    - source: narratio.transcript.final_markdown
+      required: false
+```
+
+## Root Cause
+
+ConfigKit intentionally treats sequences as atomic when comparing values
+before and after application resolution. Changing a field within one
+`publish.outputs` entry is therefore observed as a change to the
+`/publish/outputs` sequence.
+
+Narratio currently derives each missing destination by recording a
+normalization at the individual `publish.outputs[index].dest` path. That child
+record does not cover the atomic parent-sequence change. A separate
+container-level record is currently created only when Narratio also defaults
+an omitted `required` value, so configurations with explicit `required` values
+lose that incidental coverage.
+
+## Recommended Fix
+
+Make Narratio's publish-output resolution bookkeeping conform to ConfigKit's
+atomic-sequence contract. Whenever resolution derives one or more destinations
+inside `publish.outputs`, the recorded resolution changes must cover the
+`/publish/outputs` container. Destination derivation must not depend on an
+unrelated `required` default being applied in the same list.
+
+The implementation may retain more specific provenance where useful, but it
+must provide valid container-level coverage for the sequence mutation. Do not
+change ConfigKit's sequence comparison semantics merely to accommodate this
+call site.
+
+## Scope
+
+- Correct destination derivation for publish rules that explicitly set
+  `required` and omit a derivable `dest`.
+- Preserve explicit `required: true` and `required: false` values.
+- Preserve canonical destination derivation for built-in transcript sources
+  and configured artifact sources.
+- Add focused regression coverage through Narratio's real configuration load
+  and validation boundary.
+- Prepare a backward-compatible patch release after the fix passes the normal
+  repository validation.
+
+## Non-goals
+
+- Changing the `publish.outputs` schema, defaults, source vocabulary, or
+  destination derivation rules.
+- Requiring operators or configuration generators to supply `dest` when it is
+  derivable.
+- Requiring operators to omit explicit `required: true` values.
+- Treating flow-style YAML or JSON-compatible YAML values as unsupported.
+- Changing ConfigKit's documented atomic treatment of sequences.
+- Adding storage, publication, or network integration behavior.
+
+## Regression Coverage
+
+Add one focused package-level behavioral test that loads and validates a
+pipeline with omitted destinations and explicit requiredness. A representative
+list should include both `required: true` and `required: false`, then assert
+that:
+
+- loading and validation succeed;
+- each destination is derived from its source; and
+- each explicit `required` value is preserved.
+
+Using the compact flow-style representation that exposed the integration
+failure is reasonable, while the existing block-style cases continue to cover
+omitted `required`. The test should exercise the public configuration behavior,
+not ConfigKit call choreography or Narratio's private recording mechanism.
+
+No network service, object store, credential, or test double is required.
+
+## Target End State
+
+All valid publish rules with derivable sources behave consistently across
+these cases:
+
+| `dest` | `required` | Result |
+| --- | --- | --- |
+| omitted | omitted | destination derived; required defaults to `true` |
+| omitted | `true` | destination derived; explicit `true` retained |
+| omitted | `false` | destination derived; explicit `false` retained |
+| explicit | either value | explicit destination and requiredness retained |
+
+`narratio config validate`, `narratio config show`, and ordinary pipeline
+loading accept the corrected cases without an unrecorded-change diagnostic.
+No operator configuration migration or workaround is required.
+
+## Release
+
+This is a backward-compatible correction to documented configuration behavior
+and should be released as Narratio `v2.1.1`. The release note should identify
+the corrected explicit-required/derived-destination interaction and state that
+no special upgrade action is required.
+
+## Acceptance Criteria
+
+- The focused regression test fails against v2.1.0 behavior and passes after
+  the fix.
+- Publish rules with explicit `required: true` or `required: false` and omitted
+  derivable destinations load and validate successfully.
+- Derived destinations and explicit requiredness appear correctly in the safe
+  effective configuration.
+- Existing destination derivation, publish validation, imports, profiles, and
+  configuration inspection tests remain green.
+- The complete repository validation and release-candidate checks pass.
+- The correction is published as the immutable `v2.1.1` tag with the required
+  release note.

docs/roadmap/implementation.md

@@ -0,0 +1,431 @@
+# Implementation Plan: Speaker Checkpoints and Replayable Uploads
+
+Status: Implementation complete; Stages 1-7 complete.
+
+## Scope and Instructions
+
+Implement [the feature roadmap](transcribe-checkpoints-and-replay.md). This plan
+replaces the completed Notarius-profile plan; that work remains in Git history.
+Complete the numbered stages in order, one bounded implementation prompt per
+stage for gpt-5.6-terra. Record completion and actual verification results here.
+Intermediate stages are not a finished feature and must not be released alone.
+
+Before working, read `AGENTS.md`, `docs/development.md`, all three policies in
+`docs/policy/`, the roadmap, and the stage-specific preparation below. Use the
+maintained Go 1.26 toolchain. Preserve operator choices, speaker filenames,
+bounded concurrency, streaming, and existing credentials/settings. Do not add
+new commands, dependencies, service requirements, arbitrary speaker limits,
+strict transcript schemas, or a generic checkpoint framework. No release or
+upstream service work is part of this plan.
+
+## Settled Design Decisions
+
+### Progress Representation and Recovery
+
+Add optional `transcribe_state_version` (version 1) and `transcribe_speakers`
+to `manifest.StageRecord`. The latter is a map keyed by the existing derived
+speaker ID, containing completed results only. Absence means that work must be
+done; a separate pending/running/failed state machine is unnecessary. Aggregate
+stage status/error remains authoritative for invocation failure. A versioned
+empty collection is meaningful: no earlier completions are reusable.
+
+Use a typed `TranscribeSpeakerRecord` with fingerprint version/digest,
+session-relative audio path, audio SHA-256 and size, one `ArtifactRecord` output
+of kind `transcript_raw`, output size, and producer run ID. The output carries
+its session-relative canonical path and SHA-256. No new transcript source IDs,
+synthetic schema contracts, timestamps, or historical attempt collections are
+needed. Existing aggregate diagnostics can retain attempt/status/duration data
+for work performed in the current invocation; do not fabricate that information
+for reused outputs.
+
+Keep the session-manifest schema identifier unchanged for this optional additive
+state; validate supported per-stage versions and retain strict decoding. New
+Narratio reads old manifests. Old binaries may reject the added fields under
+their existing strict decoding; document this downgrade limitation.
+
+For pre-feature manifests with no speaker evidence, regenerate transcription
+once when transcribe is next selected, even if the aggregate status succeeded.
+Old records do not bind individual transcript results to the audio bytes they
+used. Do not guess that association from current files, old `per_file` metadata,
+or abandoned run directories. This is the conservative upgrade path, not a new
+general migration system. No other stage or unselected range is forcibly run.
+
+### Reuse and Persistence
+
+Compute a versioned per-speaker fingerprint from speaker ID, session-relative
+audio identity, actual audio SHA-256/size, and the existing transcribe semantic
+configuration fingerprint (URL and language). Do not include invocation ID,
+absolute workspace root, concurrency, retry settings, timeout, or credentials.
+Stream input hashing with cancellation; do not trust stored input checksums
+alone. Recheck input bytes before accepting a new result to catch changes
+observed during transcription; fail that job without checkpointing mismatched
+evidence. This is not a hostile-filesystem snapshot or concurrent-editor
+transaction guarantee and requires no extra audio copies or file locks.
+
+Validate reusable output identity, canonical location, JSON, checksum, and size.
+Use existing confinement/file primitives. Missing or corrupt regular outputs
+become work; unsafe paths/non-regular conflicts follow existing safety errors,
+not recursive deletion. Missing required audio is an input error, not permission
+to reuse an unverifiable transcript. Removed inputs are removed from the current
+collection and aggregate outputs without deleting arbitrary files.
+
+One stage-owned mutex serializes candidate collection construction, manifest
+save, and installation of committed state. Save a detached candidate and only
+then update live authority, following analyze's pattern. Workers must not mutate
+the live manifest outside that serialized section. Keep expensive HTTP work
+outside the mutex. Stop dispatch/acceptance on checkpoint failure, join workers
+before return, and preserve the last successfully committed collection.
+
+Before launching requests, durably checkpoint the reconciled collection. For
+force, that collection is empty: this prevents later ordinary retries from
+resurrecting pre-force outputs. If this initial save fails, do not start requests
+or overwrite outputs. Force again clears the current collection again; ordinary
+retry retains replacements already checkpointed after the previous force.
+
+Save each validated result at its canonical raw-transcript location before
+checkpointing it. Run-local files remain staging/diagnostics, not reuse
+dependencies. Bytes left by an unsuccessful checkpoint have no completion
+authority. A worker may checkpoint a completed result using the parent stage
+context even if a sibling canceled the work context; parent cancellation or a
+checkpoint failure may prevent that save. Guarantee retention of committed
+work, not all responses that happened to finish during cancellation.
+
+### Multipart Replay Mechanism
+
+Replace the pipe/producer upload mechanism with a small pull-based multipart
+body backed by the audio reader. Use `mime/multipart.Writer` to encode a bounded
+metadata prefix and suffix around the file payload: capture the prefix after
+`CreateFormFile`, then encode the language field and closing boundary as the
+suffix. Generate this framing once per application attempt. Each body combines
+fresh readers for prefix, freshly opened audio, and suffix using `io.MultiReader`.
+No full-audio buffer or temporary multipart spool is needed.
+
+The initial body and `Request.GetBody` use the same factory/framing. Each owns
+an independent audio handle. Give bodies idempotent Close and context-triggered
+closure (for example `context.AfterFunc`), without holding a mutex across a
+blocking Read. Read/Close coordination must release a blocked source and retain
+genuine source-read/close errors while not treating deliberate early closure as
+source failure. The attempt owns cleanup of all created bodies, including
+replays; reject new body creation after its context expires.
+
+Remove the old producer `Wait` dependency. Track body-local source errors with
+synchronization so a swallowed upload read error cannot become a successful
+transcript, and a deliberately closed original body cannot poison a successful
+replay. Keep the outer retry policy and per-attempt context. Do not add replay
+counters to the user retry budget or depend on Go's exact internal retry count.
+Preserve ordinary HTTP redirect policy; body replay can also enable Go's
+body-preserving redirects. Do not introduce a new redirect restriction.
+
+## Stage 1: Add Durable Speaker Completion State
+
+Status: Complete.
+
+Completion marker (2026-09-21): Added optional versioned transcribe-owned
+per-speaker completion records, strict load/save validation, isolated collection
+cloning, legacy aggregate-only compatibility, and lifecycle-preservation
+coverage. Verified with `go test ./internal/manifest` and `git diff --check`.
+
+Preparation: `internal/manifest/manifest.go`, `store.go`, `analyze_state.go`,
+their tests, and `docs/internal/manifest.md`.
+
+1. Implement the additive fields, record type, version constants, validation,
+   and deep-copy helper described above in a focused transcribe state file.
+2. Validate ownership by the transcribe stage, supported versions, required
+   completed-record evidence, output kind, safe relative paths, SHA-256 values,
+   sizes, and producer identity consistency. Use existing speaker IDs as-is;
+   do not apply analysis-artifact key regexes or copy unrelated analysis limits.
+3. Integrate validation with both load and save. Preserve speaker state through
+   running/failed/stale transitions and aggregate result clearing. Generic stage
+   invalidation does not discard checkpoints; reconciliation checks their data.
+4. Test round-trip populated and versioned-empty collections, old manifests
+   without fields, malformed evidence, deep-copy isolation, and lifecycle
+   preservation. Avoid imposing JSON formatting or map order as contracts.
+
+Verify: `go test ./internal/manifest`.
+
+Done when durable storage can represent partial completion independently of
+aggregate status, without activating new stage behavior yet.
+
+## Stage 2: Implement Read-Only Speaker Reconciliation
+
+Status: Complete.
+
+Completion marker (2026-09-21): Extracted shared deterministic transcribe job
+construction and added read-only, cancellation-aware per-speaker input and
+output reconciliation. Covered unchanged reuse, semantic and operational
+settings, changed/added/removed inputs, legacy and forced work, missing/corrupt
+outputs, unsafe evidence, and missing audio without connecting runtime skip or
+checkpoint behavior. Verified with
+`go test ./internal/stage ./internal/manifest` and `git diff --check`.
+
+Preparation: `internal/stage/transcribe.go`, `semantic_contracts_initial.go`,
+`analyze_reconciliation.go`, `extract_resume.go`, their tests, and the artifact
+path/integrity helpers they use.
+
+1. Extract the existing audio discovery/job construction into a shared,
+   transcribe-local reconciliation path. Preserve current input selection and
+   speaker identity behavior. Build sorted jobs and classify completed results
+   against the current input set and evidence defined above.
+2. Add pure fingerprint construction and streamed input/output inspection.
+   Resolve manifest input paths with existing session-relative/absolute reading
+   helpers; do not change input precedence or assume the current working directory
+   is the session root. Newly written checkpoint paths are always relative.
+3. Return current records and work-to-do without writing files or manifest state.
+   Force selects all work and an empty starting collection; old unversioned
+   transcribe state supplies no reusable records. Do not connect aggregate skip
+   validation until execution can produce the new evidence in Stage 3.
+4. Test unchanged reuse; one changed audio among unchanged speakers; URL/language
+   changes; operational-setting exclusions; added/removed speakers; missing and
+   corrupt regular outputs; unsafe record paths; missing audio; and force.
+   Prefer compact tables and real temporary files over mocked hashing.
+
+Verify: `go test ./internal/stage ./internal/manifest`.
+
+Done when one read-only decision path describes exactly which speakers need
+work, without altering existing runtime behavior prematurely.
+
+## Stage 3: Execute and Checkpoint Speakers Incrementally
+
+Status: Complete.
+
+Completion marker (2026-09-21): Connected reconciliation to bounded execution,
+durably checkpointed the starting and per-speaker collections through detached
+manifest candidates, preserved committed work across failures and cancellation,
+rebuilt deterministic aggregate outputs from reusable evidence, and added
+speaker-aware resume validation. Updated production-style fixtures and covered
+serialized concurrent saves, force/retry replacement, input rechecks, save
+failures, partial recovery, and aggregate reconstruction. Verified with
+`go test ./internal/stage ./internal/app`, `go test -race ./internal/stage`, and
+`git diff --check`.
+
+Preparation: `internal/stage/transcribe.go`, `analyze_execution_state.go`,
+`run_local.go`,
+`internal/app/runner.go`, and transcribe/checkpoint test fixtures.
+
+1. Wire reconciliation into Run, require the existing ManifestStore collaborator,
+   and checkpoint the starting collection before launching HTTP work. Implement
+   detached candidate saves and serialized live-state updates. Audit Store.Save's
+   normalization side effects so a failed candidate save cannot mutate shared
+   authority; use detached records/collections wherever it may write.
+2. Retain bounded errgroup execution and fail-fast cancellation. Check the work
+   context before beginning each queued request. After success, validate adapter
+   output path/JSON, recheck audio evidence, materialize canonical bytes, compute
+   output evidence, and checkpoint the new record. Check a checkpoint-failure
+   latch under the serialization lock before accepting further completions.
+3. Preserve committed successes on all error/cancellation exits, including
+   starting-save, canonical-write, per-speaker-save, and aggregate-save failures.
+   Return meaningful speaker/error context and let the runner retain its existing
+   terminal failure behavior. Do not report partial aggregate outputs as success.
+4. Assemble final outputs in deterministic speaker order from the reconciled
+   completed records, including reused speakers, with original producer IDs and
+   checksums. If all records are current but aggregate status is failed/running,
+   finish without WhisperX calls and rebuild the aggregate success.
+5. Implement `transcribeStage.ValidateResume` using the same read-only
+   reconciliation, so a succeeded stage cannot skip changed audio or invalid
+   transcripts. Retain the existing URL/language semantic check and runner
+   invalidation behavior; no whole-config digest invalidation.
+6. Adapt existing tests that require no canonical partial files: the new contract
+   permits checkpointed files but forbids aggregate success. Update fixtures to
+   provide real manifest stores and records rather than a production fallback
+   that silently disables checkpointing.
+7. Test A/B completion followed by C failure and D unattempted (concurrency 1),
+   reload and retry only C/D; concurrently finishing speakers without lost saves;
+   failure and cancellation after a checkpoint; save failure retaining earlier
+   authority; and force failure followed by ordinary retry without old-result reuse.
+
+Verify: `go test ./internal/stage ./internal/app` and
+`go test -race ./internal/stage`.
+
+Done when incremental progress and aggregate success work together safely under
+ordinary, concurrent, canceled, and forced execution.
+
+## Stage 4: Verify Runner, Merge, and Durable Lifecycle Integration
+
+Status: Complete.
+
+Completion marker (2026-09-22): Added real-persistence application coverage for
+cross-invocation retry, interrupted running state, failed aggregate saves,
+force/reset persistence, downstream invalidation, and one-time legacy upgrade.
+Versioned merge input discovery now requires aggregate success and an exact
+match between current speaker evidence and declared aggregate outputs, while
+legacy/manual directory fallback remains available. Restore coverage preserves
+session-relative speaker paths, leaves unpublished raw transcripts absent, and
+canonical transcripts remain reusable after run-directory cleanup. Verified
+with `go test ./internal/app ./internal/stage ./internal/manifest`,
+`go test -race ./internal/app`, and `git diff --check`.
+
+Preparation: `internal/app/runner.go`, `runner_runtime.go`,
+`analyze_checkpoint_recovery_test.go`, `runner_terminal_failure_test.go`,
+`restore_manifest_paths.go`, `internal/stage/merge.go`, and existing publish,
+restore, and cleanup tests routed from `docs/internal/overview.md`.
+
+1. Add application-level tests with real manifest persistence and fake WhisperX:
+   failed stage across separate invocations; all speakers committed before a
+   failed aggregate save; interrupted running state; force/reset persistence;
+   and unchanged upstream/succeeded downstream invalidation behavior. Inject
+   errors at durable boundaries, not every private method call.
+2. Close `discoverRawTranscripts`' directory-scan fallback for versioned
+   transcribe state: incomplete aggregate state is not usable, and completed
+   state consumes only declared current aggregate outputs. Keep legacy/manual
+   fallback behavior for manifests without the new state rather than banning
+   previously allowed standalone workflows. Test that partial files and removed
+   speaker files cannot leak into a versioned merge, including direct stage calls.
+3. Audit manifest copies/projections, remote manifest publication, restored path
+   handling, and cleanup. Preserve relative speaker paths on restore and verify
+   they resolve under the destination session, not the producer machine. Reuse
+   existing confinement validation, not a second path-policy layer.
+4. Add focused round-trip coverage through existing publish/restore fixtures for
+   speaker records. Restored outputs absent from the published selection are
+   non-reusable locally; do not silently expand remote publication to every raw
+   speaker or upload audio as a new default. Removing run directories must not
+   remove canonical checkpointed transcripts or make reuse require run files.
+5. Verify old succeeded manifests are readable and rerun once when transcribe is
+   selected, then reuse normally. Keep old manifests usable for other commands
+   and bounded ranges. Do not automatically import old unrecorded JSON files.
+
+Verify: `go test ./internal/app ./internal/stage ./internal/manifest` and
+`go test -race ./internal/app`.
+
+Done when the new authority survives runner and storage lifecycles and cannot
+expose incomplete transcripts downstream.
+
+## Stage 5: Implement Replayable Streaming Multipart Bodies
+
+Status: Complete.
+
+Completion marker (2026-09-22): Replaced the producer-goroutine/pipe upload
+with bounded standard-library multipart framing and independently opened,
+pull-based streaming bodies shared by the initial request and `GetBody`.
+Bodies are tracked, idempotently closed, canceled with their attempt context,
+and retain genuine read/close failures without treating an intentionally
+abandoned original as a replay failure. Added focused coverage for identical
+framing, partial-read restart, simultaneous handles, source failures, and
+blocked-reader release. Verified with `go test ./internal/adapters/whisperx`,
+`go test -race ./internal/adapters/whisperx` (including ten repeated runs), and
+`git diff --check`.
+
+Preparation: `internal/adapters/whisperx/http.go`, `http_test.go`,
+`docs/integrations/whisperx.md`, and the installed Go net/http Request/GetBody
+contract. Keep this change inside the WhisperX adapter.
+
+1. Implement the pull-based framing/body factory described above, using standard
+   multipart encoding rather than hand-escaped MIME headers. Add no new payload
+   cap or upload-size setting. Preserve file-before-language wire semantics and
+   the existing form names.
+2. Install the initial body and GetBody with the same per-attempt boundary.
+   Track and close all opened bodies even if a custom transport returns early;
+   make cleanup/context callbacks safe against overlapping reads and closes.
+   A GetBody failure must be returned, not converted into an empty upload.
+3. Replace pipe-specific Wait/error paths in doTranscribeAttempt with source-error
+   handling appropriate to pull bodies. Genuine source errors remain visible;
+   intentional closure during transport retry is not itself a source error.
+   Keep response bounds, JSON validation, output installation, retry
+   classification/delay, cancellation, and application attempt reporting intact.
+4. Remove obsolete producer-only helpers rather than retaining two upload paths.
+   Adapt tests tied to producer goroutines to assert streaming and resource
+   release, which are the actual contracts.
+5. Add focused body tests: original/replayed bytes parse identically; replay
+   after partial consumption starts from the beginning; simultaneous bodies use
+   independent handles; source failures propagate; Close/cancel releases a
+   blocked reader without deadlock. Use synthetic audio and gated readers.
+
+Verify: `go test ./internal/adapters/whisperx` and
+`go test -race ./internal/adapters/whisperx`.
+
+Done when replay is available without whole-file buffering or producer-lifetime
+ambiguity, and all existing HTTP behavior tests still pass.
+
+## Stage 6: Verify Replay Through the HTTP Adapter Boundary
+
+Status: Complete.
+
+Completion marker (2026-09-22): Added a deterministic transport regression
+that abandons a partially consumed original request, obtains and parses a fresh
+multipart replay from `GetBody`, and installs the returned transcript within one
+application attempt. Covered replay open/read failures, early responses,
+parent cancellation during replay, and disposal of every opened source, while
+proving an abandoned original does not poison a valid replay. Added a real HTTP
+retry-exhaustion integration that durably retains an earlier speaker and resumes
+only the uncommitted speakers. This controlled boundary verifies Narratio's
+replay contract, not recovery from every real proxy protocol failure. Verified
+with `go test ./internal/adapters/whisperx ./internal/app ./internal/stage`,
+`go test -race ./internal/adapters/whisperx ./internal/app ./internal/stage`,
+ten repeated race-enabled adapter runs, and `git diff --check`.
+
+Preparation: Existing injectable HTTPClient/RoundTripper seam and local-server
+tests in `internal/adapters/whisperx/http_test.go`; app WhisperX wiring tests.
+
+1. Add a deterministic transport-boundary regression: consume part of the
+   original request, close it as a transport would, call GetBody, consume and
+   parse the replay using the request Content-Type, then return valid JSON.
+   Verify exact audio/language, installed output, and one application attempt.
+   This must fail before replay support; merely checking non-nil GetBody is
+   insufficient. Do not assert Go's internal retry count or require a flaky
+   handcrafted HTTP/2 peer in the default suite.
+2. Cover replay-open/read failure, early response, parent cancellation during
+   replay, and disposal of all body handles. Reuse Stage 5 body coverage rather
+   than repeat every low-level case. Prove an intentionally abandoned original
+   body does not turn a valid replay response into failure.
+3. Retain local HTTP tests for application retries, non-retryable responses,
+   response limits, invalid JSON, and bounded streaming; exercise that an
+   exhausted HTTP failure still leaves earlier speaker checkpoints reusable
+   using the existing stage/app seams.
+4. Record the verification boundary honestly: controlled transport replay proves
+   Narratio's body contract, not recovery from every real proxy PROTOCOL_ERROR.
+
+Verify: `go test ./internal/adapters/whisperx ./internal/app ./internal/stage`
+and race tests for changed packages.
+
+Done when the replay defect is protected at the adapter boundary and the
+application retry/checkpoint behavior remains intact.
+
+## Stage 7: Document and Validate the Completed Feature
+
+Status: Complete.
+
+Completion marker (2026-09-22): Updated the canonical transcribe, merge,
+manifest, WhisperX, operations, and troubleshooting documentation for durable
+speaker reconciliation, serialized checkpoints, aggregate gating, durable
+force reset, legacy upgrade and downgrade limits, replayable streaming bodies,
+resource ownership, and the absence of an exactly-once remote guarantee.
+Reviewed the completed implementation against the feature roadmap and
+implementation-discipline guidance. Verified with `go test ./...`,
+`go test -race ./...`, `go vet ./...`, `go build ./...`,
+`go test ./internal/doccheck`,
+`go test ./internal/config -run '^TestExamplesLoadAndValidate$'`, and
+`git diff --check`.
+
+1. Update canonical transcribe and manifest internals for the collection,
+   reconciliation, durable-reset force behavior, serialized saves, and aggregate
+   gate. Remove obsolete all-speakers-before-canonical-write descriptions.
+2. Update the WhisperX contract for replayable bounded streaming, application
+   attempts versus transport replay, resource ownership, and the lack of an
+   exactly-once server guarantee. Do not claim the remote protocol issue is fixed.
+3. Update operations/troubleshooting with ordinary resume versus force, durable
+   partial success, old-manifest one-time retranscription on selecting transcribe,
+   downgrade limitations, and private-model-change guidance. No new CLI flags
+   or promises of automatic historical-file recovery. Keep detailed wire and
+   storage contracts in their canonical owners.
+4. Review changes against the feature roadmap and the implementation-discipline
+   guidance. Do not elevate selected storage encoding, internal ordering, test
+   synchronization, or exact error wording into new product guarantees.
+5. Run the full repository gates below; document results and update completion
+   statuses only after they pass. Leave release preparation for a separate task.
+
+```sh
+go test ./...
+go test -race ./...
+go vet ./...
+go build ./...
+go test ./internal/doccheck
+go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+git diff --check
+```
+
+## Open Questions
+
+None requiring user input. The agreed roadmap supplies the product decisions;
+this plan selects a completed-only manifest collection, conservative old-record
+handling, and pull-based replay as implementation choices. No stronger remote
+reliability guarantee, new operator restriction, or generic orchestration
+surface is proposed.

docs/roadmap/notarius-config-profile.md

@@ -0,0 +1,151 @@
+# Notarius Configuration Profile Selection
+
+Status: Implemented.
+
+## Goal and Authority
+
+Implement the consumer behavior in the
+[feature request](feature-request-notarius-config-profile.yml): let a Narratio
+configuration overlay explicitly select a Notarius configuration overlay.
+Operators can select testing once in Narratio while each application retains
+ownership of its own model choices and other settings.
+
+This roadmap describes the target state and required change areas, not an
+implementation sequence; the [implementation plan](implementation.md) owns
+the ordered work and implementation decisions. The feature request is the scope
+authority; the mechanisms suggested below do not establish additional
+user-facing contracts.
+
+## Underlying Issue
+
+Narratio can select its own testing profile, but extraction currently invokes
+Notarius without a configuration-profile selector. Consequently, a Narratio
+testing invocation cannot explicitly select the independently deployed Notarius
+testing overlay through pipeline configuration.
+
+Adding individual downstream model or tuning fields would duplicate Notarius's
+configuration responsibilities. Passing its existing profile selector provides
+the requested flexibility without that duplication.
+
+## Targeted End State
+
+### Explicit Configuration Composition
+
+`pipeline.notarius.config_profile` is an optional Notarius configuration-profile
+name, usable in Narratio's base configuration and ordinary overlays. For
+example, Narratio's `profiles/testing.yml` may contain:
+
+```yaml
+notarius:
+  config_profile: testing
+```
+
+Selecting Narratio's `--profile testing` then selects that explicit downstream
+profile whenever extraction executes. The two profile names need not match.
+Narratio does not automatically forward its own profile name.
+
+When supplied, the selection is passed through Notarius's `--config-profile`
+argument. Omitted or null values add no profile-selection flag and preserve
+existing downstream selection behavior. Notarius's selector semantics remain
+authoritative, including empty and `default` selections for the composed base
+configuration. Narratio does not introduce a stricter profile-name policy.
+
+This field selects a ConfigKit configuration overlay, not a PromptKit profile
+through `--llm-profile`. The Notarius overlay owns any resulting LLM-profile
+selection. Notarius resolves overlays relative to its own configuration root;
+Narratio neither resolves those files against its own root nor rewrites them.
+
+### Execution, Reuse, and Failure
+
+A changed downstream profile selection must not silently reuse Narratio's
+completed extraction from the previous selection. Existing dependent-stage
+invalidation applies when extraction changes; the feature does not change the
+pipeline order or transcript-generation behavior.
+
+Notarius continues to own downstream overlay discovery, composition, validation,
+and cache compatibility. A missing or invalid selected overlay must fail
+extraction through the existing subprocess error path, without silently falling
+back to production or reporting successful extraction.
+
+Existing resume and chunk-cache controls, including force precedence, remain
+in place. Profile selection does not itself mean forced regeneration or cache
+deletion. Narratio's decision to execute extraction and Notarius's decision to
+reuse compatible work remain separate responsibilities.
+
+Changes to private Notarius configuration behind an unchanged selector retain
+the existing operator responsibility to force extraction when needed. There is
+no new downstream file inspection or content-tracking requirement.
+
+Credential forwarding remains independently configured through
+`notarius.llm_api_key`. A Narratio testing overlay may select a different
+credential reference alongside the downstream profile; no new credential
+discovery or provider-name mapping is introduced.
+
+## Required Change Areas
+
+### Configuration and Inspection
+
+Extend the typed Notarius configuration and its existing safe effective-YAML
+projection to carry the optional field. Preserve the ordinary configuration
+composition behavior for base values, overlay overrides, and null values.
+Configuration inspection must reflect the selected field without exposing
+credential values. No replacement configuration loader or downstream preflight
+loader is needed.
+
+### Stage and Adapter Boundary
+
+Carry the selection from resolved configuration through the extraction request
+to the subprocess adapter. Construct the optional CLI flag using the existing
+argument-vector mechanism, alongside the current configuration path, references,
+credential forwarding, and cache controls. Retain normal Notarius error handling.
+
+### Extraction Reuse Evidence
+
+Account for the downstream selection in the existing extraction reuse evidence.
+The current semantic-configuration and invocation fingerprints are the relevant
+change points. Keep their treatment of selection consistent so that a profile
+change cannot bypass the required rerun and downstream invalidation.
+
+This roadmap does not require a new manifest schema, separate profile-indexed
+workspaces, historical per-profile result retention, or an upgrade-wide forced
+rerun.
+
+### Verification and Documentation
+
+Use focused behavioral coverage for configuration composition and inspection,
+explicit argument forwarding and omission, changed-selection reuse decisions,
+and ordinary subprocess failure handling. Reuse existing test coverage where
+it already protects unchanged credentials, force behavior, and invalidation;
+do not duplicate Notarius's complete profile-validation or cache test suites.
+
+When implemented, update the canonical
+[configuration reference](../config.md),
+[Notarius integration contract](../integrations/notarius.md),
+[extract internals](../internal/stage-extract.md), and relevant maintained
+examples. The development workflow in [Operations](../operations.md) should
+explain explicit profile composition and the existing need to force extraction
+after private downstream edits. Clarify that profile selection does not create
+an isolated testing workspace or separate current extraction state.
+
+## Compatibility and Scope Boundaries
+
+Notarius v0.7.0 already documents `--config-profile`; no upstream change is
+required. Document the capability needed when the option is used, without adding
+an unconditional runtime version gate for existing unconfigured deployments.
+
+This feature adds no Narratio CLI selector, automatic profile propagation,
+Notarius LLM-profile selector, downstream configuration rewriting, new
+credential mechanism, or generalized recursive orchestration. Existing output
+and receipt contracts remain unchanged.
+
+The consumer's Ansible role and suite updates are follow-up work outside this
+repository. They can deploy downstream overlays and populate this field after
+the interface is released.
+
+## Scope Review
+
+No stronger requirements or new invariants are proposed beyond the feature
+request and existing repository contracts. In particular, minimizing one-time
+upgrade invalidation and avoiding reruns between equivalent base selectors are
+implementation considerations, not additional acceptance guarantees imposed
+by this roadmap. No user decisions are currently required before planning.

docs/roadmap/remediation.md

@@ -0,0 +1,104 @@
+# Transcription Checkpoint Review Remediation
+
+Status: Complete.
+
+Reviewed commits `7b0bf99` through `96b4e73` against the speaker-checkpoint and
+upload-replay roadmap and the original user request. No production changes were
+made during review. The two failures below were reproduced with temporary tests
+using the existing stage fixtures; those temporary tests were removed afterward.
+
+## 1. Reconcile Aggregate State Even When No Speaker Needs HTTP Work
+
+Priority: P2. Status: Complete.
+
+Implemented read-only aggregate identity/evidence checks before resume skips.
+Regression tests cover removed inputs, missing/duplicate/mismatched outputs,
+repair without HTTP requests, downstream invalidation, and exclusion of removed
+speakers from merge without deleting their files. Equivalent path forms and
+output ordering remain reusable. Focused app/stage tests and documentation
+checks pass.
+
+Location: `internal/stage/transcribe_resume.go:21` and the reconciliation result
+consumed there.
+
+`ValidateResume` returns reusable whenever `reconciliation.work` is empty. That
+only establishes that the remaining speakers need no transcription; it does
+not establish that the stored speaker collection and aggregate outputs match
+the current input set. Removing a speaker while the remaining inputs are
+unchanged produces an empty work list, so a succeeded record can be skipped
+without pruning the removed speaker. Its old aggregate output can then remain
+available to merge. Likewise, missing/inconsistent aggregate output references
+are not repaired when all individual checkpoints remain valid. A prepare rerun
+may already stale transcribe in some workflows, but the resume validator cannot
+rely on that to satisfy its own current-input/output checks.
+
+Reproduction: create complete alice/bob checkpoint records and aggregate
+outputs with the existing fixture, mark the stage succeeded, remove bob from
+the manifest's audio inputs, and call `ValidateResume`. It returns reusable
+despite still retaining bob in the stored collection and aggregate outputs.
+
+Required change:
+
+- Distinguish no HTTP work from no reconciliation work. Validate that the
+  committed speaker set and aggregate transcript references represent the
+  reconciled current set, including relevant identity/evidence, before skipping.
+- When only collection/output repair is needed, return non-resumable and let
+  normal execution checkpoint the pruned collection and rebuild aggregate
+  outputs without retranscribing reusable speakers. Preserve existing dependent
+  invalidation; do not force otherwise-current speaker jobs.
+- Keep resume inspection read-only. Do not fix this by deleting incidental
+  transcript files or changing stage selection rules.
+
+Verification: add removed-only input coverage (not just removal combined with
+an added speaker), plus incomplete/inconsistent aggregate output coverage.
+Exercise the runner skip path and assert zero additional WhisperX requests,
+correct rebuilt outputs, and no removed speaker reaching merge.
+
+## 2. Remove the Unrequested Positive Audio-Size Requirement
+
+Priority: P2. Status: Complete.
+
+Audio evidence now permits zero size while still rejecting negative values.
+Regression tests verify zero-size manifest round trips, successful service
+results being checkpointed and reused, and service rejection being propagated
+without creating a completion record. Nonempty valid-JSON output validation is
+unchanged. Both new success regressions failed before the fix.
+
+Location: `internal/manifest/transcribe_state.go:95`.
+
+The new record validator requires `AudioSize > 0`. Previously Narratio accepted
+a regular `.flac` file and delegated transcription acceptance to WhisperX;
+neither the user request nor the roadmap introduced minimum audio size or a new
+audio-content validation policy. Zero-byte content still has a valid checksum
+and can be represented by the checkpoint mechanism. The current check runs
+after a successful response and output materialization, causing the entire
+stage to fail instead of retaining the completed work.
+
+Reproduction: prepare one speaker, replace its prepared audio with an empty
+regular file, and use the existing successful fake WhisperX client. Stage
+execution fails at checkpoint validation with `audio_size must be positive`.
+This demonstrates the extra Narratio restriction; it does not assert that a
+particular production WhisperX deployment accepts empty audio.
+
+Required change:
+
+- Permit zero audio size and reject negative sizes. Keep byte hashing and
+  size equality checks. Let the downstream service decide whether the input
+  can be transcribed, as before.
+- Do not remove the nonempty valid-JSON output requirement, which has a separate
+  existing justification. Do not add a new early empty-audio rejection instead.
+- Update any new tests/documentation that promote strictly positive input size
+  to a contract.
+
+Verification: cover a zero-size record round-trip and successful adapter result
+being checkpointed/reused for that input; retain negative-size rejection and
+ordinary downstream-failure tests.
+
+## Verification After Remediation
+
+Completed on 2026-09-22 with Go 1.26.8: focused stage/application tests,
+changed-package race tests, `go test ./...`, `go test -race ./...`,
+`go vet ./...`, `go build ./...`, `go test ./internal/doccheck`,
+`go test ./internal/config -run '^TestExamplesLoadAndValidate$'`, and
+`git diff --check` all passed. Existing replay and cancellation regression
+tests remain intact; the HTTP adapter required no changes.

docs/roadmap/transcribe-checkpoints-and-replay.md

@@ -0,0 +1,183 @@
+# Resumable Speaker Transcription and Replayable Uploads
+
+Status: Implemented.
+
+## Goal and User Intent
+
+Preserve completed speaker transcriptions when another speaker fails, so an
+ordinary run or resume does not repeat expensive successful work. Separately,
+allow Go's HTTP transport to replay WhisperX uploads when it encounters a
+recoverable connection or stream failure.
+
+The motivating incident exhausted four transcription attempts for one speaker
+after three other speakers had completed. The final error reported an HTTP/2
+peer `PROTOCOL_ERROR` and an inability to replay the request body without
+`Request.GetBody`.
+
+These are related reliability improvements, not a promise to eliminate all
+service failures. This document owns the target state and scope; the
+[implementation plan](implementation.md) owns ordered changes and concrete
+storage and replay decisions.
+
+## Pre-implementation Limitations
+
+The [transcribe stage](../internal/stage-transcribe.md) writes successful
+responses into the invocation's run directory. It installs canonical transcripts
+and returns its per-file metadata only after every speaker succeeds. A failure
+therefore leaves no durable per-speaker progress that ordinary resume can use,
+even though some completed JSON files may remain on disk.
+
+The [WhisperX adapter](../integrations/whisperx.md) already retries requests at
+the application level, reopening audio for each attempt. Its multipart request
+body is streamed through a pipe and has no replay callback. Transport-level
+recovery within an attempt cannot reconstruct a consumed body. Providing that
+capability addresses the replay limitation, not the underlying peer protocol
+error; persistent failures can still require service or proxy investigation.
+
+## Targeted End State
+
+### Durable Per-Speaker Completion
+
+Each successfully transcribed speaker file is validated, saved, and recorded
+durably before that work is treated as reusable. The session manifest remains
+the progress authority. Follow the existing analyze-stage checkpoint pattern
+without turning speakers into pipeline stages or introducing a general task
+engine.
+
+Ordinary execution reconciles the current prepared audio set with recorded
+speaker results. It reuses valid completed work and schedules failed,
+unattempted, missing, or obsolete results. A result is not reusable merely
+because a matching filename exists. Evidence binds it to the speaker/audio
+identity, audio content, Narratio-visible recognition settings, and the saved
+transcript's integrity.
+
+The relevant recognition settings remain the WhisperX service URL and language.
+Retry count/delay, concurrency, timeout, credentials, and diagnostic locations
+remain operational settings rather than reasons to discard completed results.
+Changes to private service models or configuration remain an operator reason
+to force transcription; Narratio does not inspect the service's private state.
+
+Changes to individual audio inputs require replacement of affected results,
+not automatic retranscription of unrelated unchanged speakers. New speakers
+need transcription; removed speakers do not remain in the current aggregate
+output set. Records must describe the inputs actually used, not grant reuse to
+old bytes under a changed identity.
+
+### Aggregate Completion, Failure, and Force
+
+The overall stage succeeds only when every currently required speaker has a
+validated, durably recorded result. Incomplete per-speaker progress does not
+make the aggregate stage successful or authorize downstream consumption of an
+incomplete collection. Existing dependent-stage invalidation remains in effect.
+
+Retain bounded concurrency and the current fail-fast behavior: a speaker
+failure cancels outstanding work. Results already checkpointed remain reusable
+after failure, cancellation, or process interruption. A later ordinary run or
+resume retries the remaining work; it does not need a new CLI command.
+
+Checkpoint writes must be serialized so concurrently finishing speakers cannot
+overwrite each other's progress. A failed persistence operation must not
+advertise an unrecorded result as complete. Previously committed progress remains
+the recovery authority, and checkpoint errors must be surfaced rather than
+reported as successful transcription.
+
+`--force` on transcribe regenerates all currently required speakers instead of
+reusing their prior completions. Successful replacements are checkpointed too,
+so an ordinary retry after a failed forced invocation can retain the newly
+completed work. Old results for speakers not successfully regenerated must not
+satisfy that unfinished forced replacement on the next ordinary retry.
+
+This deliberately relaxes the current all-speakers-before-canonical-installation
+implementation: individual durable results may exist while the stage remains
+failed or incomplete. The all-speakers-before-stage-success boundary remains.
+
+### Replayable WhisperX Requests
+
+Provide `Request.GetBody` so the HTTP transport can obtain a fresh body for the
+same request. Replay reopens the audio and reproduces the multipart boundary,
+filename, fields, and framing used by the original request. Independent bodies
+must not share a consumed reader or mutable read position.
+
+Preserve bounded-memory streaming rather than buffering entire audio files in
+memory. Correctly close every original and replay body and its audio resources;
+preserve source-read errors and cancellation, including partial uploads and
+early server responses. Successful response handling must not accidentally wait
+on, or report errors solely from, an abandoned body instead of the replay that
+produced the response.
+
+Keep the existing application-level retry count, delay, classification, and
+per-attempt timeout semantics. Transport replays occur within an application
+attempt and its context; they are not additional Narratio-configured attempts.
+Do not add another custom retry loop, silently increase retries, or disable
+HTTP/2 globally as a substitute for replay support.
+
+The adapter still validates and installs only successful JSON responses.
+Request replay does not provide exactly-once server execution: a lost response
+can cause repeated transcription work, as can existing application retries.
+No new service-side idempotency protocol is required by this feature.
+
+## Required Change Areas
+
+- **Manifest and persistence:** Add transcribe-owned per-speaker progress and
+  evidence, with safe checkpoint updates and preservation through runner failure
+  handling. Review manifest validation, copying, serialization, and existing
+  publish/restore handling wherever the new records pass through those paths.
+- **Stage execution and reuse:** Reconcile speaker records before scheduling,
+  checkpoint validated completions, rebuild aggregate outputs from the current
+  set, and account for partial progress in ordinary and forced execution.
+  Review successful-stage reuse as well as failed-stage retries so stale audio
+  or missing/corrupt transcript evidence is not bypassed by an aggregate skip.
+- **Artifact lifecycle:** Use existing path, installation, integrity, and manifest
+  mechanisms. Ensure checkpointed results do not depend on an ephemeral run
+  directory surviving cleanup. Do not introduce a second authoritative ledger
+  or infer completion by scanning abandoned runs.
+- **HTTP adapter:** Make multipart bodies reproducible and correctly manage
+  every replay's lifetime, without changing the WhisperX HTTP schema or requiring
+  an upstream service change.
+- **Tests:** Cover per-speaker recovery after a later failure, interruption and
+  checkpoint failure; changed/missing inputs or outputs; force followed by an
+  ordinary retry; concurrent successful completions; and aggregate downstream
+  gating. Extend existing fixtures rather than duplicating analyze's complete
+  test suite. Exercise actual body reconstruction after partial consumption,
+  multipart equivalence, cancellation/resource cleanup, and existing bounded
+  streaming and retry behavior—not just the presence of a callback. Default
+  tests remain offline and use local/fake HTTP boundaries, never paid services.
+- **Documentation:** Update the canonical
+  [stage guide](../internal/stage-transcribe.md),
+  [manifest guide](../internal/manifest.md),
+  [WhisperX contract](../integrations/whisperx.md),
+  [operations guide](../operations.md), and
+  [troubleshooting guide](../troubleshooting.md) when implemented. Explain
+  partial completion, ordinary retry versus force, and the distinction between
+  request replay limitations and persistent peer protocol failures.
+
+## Compatibility and Scope Boundaries
+
+Existing speaker identities, raw transcript formats, and the fixed pipeline
+order remain unchanged. Existing manifests remain readable, but must not be
+treated as containing per-speaker evidence they never recorded. Recovery relies
+on verified recorded evidence, not guessed associations between old transcripts
+and current audio. No general legacy-migration framework is part of the feature.
+
+Automatic adoption of unrecorded outputs from previously failed runs is outside
+this feature. Such files may be recoverable after verifying their inputs, but
+their presence alone does not prove completion. No recovery/import command,
+speaker-selection CLI, historical invocation ledger, new service configuration
+discovery, or broader orchestration refactor is included.
+
+The feature does not diagnose the remote service/proxy's underlying HTTP/2
+problem, guarantee recovery from persistent protocol errors, or change the
+policy for continuing unrelated work after failure.
+
+## Completion Criteria
+
+For speakers A, B, C, and D: if A and B are checkpointed, C fails, and D has not
+completed, ordinary resume retains A and B, retries C, and completes D. No
+downstream stage receives an incomplete successful transcription collection.
+Force replaces all four, while an ordinary retry can reuse checkpoints from
+that forced replacement attempt.
+
+The HTTP transport can reconstruct the upload after consuming the original
+body, without whole-file memory buffering, lost source errors, or stranded
+upload resources. A persistent peer error still fails normally after the
+configured application attempts, without discarding checkpointed speakers.

docs/troubleshooting.md

@@ -213,6 +213,134 @@ input validation.
 
 Relevant reference: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior).
 
+## Transcription retries speakers that previously finished
+
+Symptom:
+
+- an ordinary transcribe retry calls WhisperX for a speaker that appeared to
+  finish earlier; or
+- an older successful session retranscribes all speakers once after upgrade.
+
+Likely causes:
+
+- the earlier response or canonical file was not followed by a successful
+  speaker manifest checkpoint;
+- the prepared audio bytes/path, WhisperX URL, or language changed;
+- the canonical transcript is missing, unsafe, non-regular, invalid JSON, or
+  no longer matches its recorded size/checksum;
+- the previous invocation used `--force`, which durably cleared pre-force
+  speaker authority; or
+- the manifest has only legacy aggregate transcribe state and cannot bind old
+  outputs to their source audio.
+
+Safe fix:
+
+- inspect the reported speaker and filesystem error, repair the prepared input
+  or canonical-output path when needed, then retry transcribe normally;
+- allow a legacy aggregate-only record to regenerate once; and
+- do not edit the manifest or copy an old JSON file into place as a substitute
+  for a successful checkpoint.
+
+```bash
+narratio run-stage transcribe 2026-04-04
+```
+
+Valid earlier speaker checkpoints remain reusable even while aggregate
+transcribe status is failed or interrupted. Merge remains gated until
+transcribe reaches aggregate success with the complete current set.
+
+Relevant reference: [Operations: Transcription continuation and replacement](./operations.md#transcription-continuation-and-replacement).
+
+## WhisperX model changed but transcription was reused
+
+Symptom:
+
+- WhisperX model weights, prompts, or private server configuration changed at
+  the same configured URL, but Narratio still considers speaker transcripts
+  current.
+
+Likely cause:
+
+- Narratio can fingerprint its configured URL and language, not private state
+  loaded behind the service boundary.
+
+Safe fix:
+
+- explicitly force transcribe to establish a durable reset, then use ordinary
+  retry if the replacement attempt is interrupted.
+
+```bash
+narratio run-stage transcribe 2026-04-04 --force
+```
+
+Transport replay and application retries do not provide an exactly-once remote
+execution guarantee. Investigate server-side failures or duplicate work using
+WhisperX diagnostics; do not infer completion from a server log without the
+local speaker checkpoint.
+
+Relevant references: [WhisperX integration](./integrations/whisperx.md) and
+[Transcribe internals](./internal/stage-transcribe.md).
+
+## Transcription manifest checkpoint failed
+
+Symptom:
+
+- transcribe reports that a speaker completed but the session manifest could
+  not be saved.
+
+Likely causes:
+
+- the workspace is unwritable or out of space;
+- the manifest path or parent directory is unsafe or the wrong type; or
+- a transient filesystem error interrupted atomic replacement.
+
+Safe fix:
+
+- correct the reported filesystem or permission problem and retry transcribe
+  normally. Narratio reconciles from the last readable collection and reuses
+  only successfully checkpointed speakers.
+
+A canonical raw transcript left by the failed checkpoint is incidental and
+does not become reusable authority. Preserve the manifest and diagnostics if
+the failure repeats; do not advertise the file by editing manifest evidence.
+
+Relevant references: [Manifest persistence](./internal/manifest.md#persistence-semantics)
+and [Transcribe internals](./internal/stage-transcribe.md).
+
+## Notarius credential reference unavailable
+
+Symptom:
+
+- configuration loading reports `secret.unset`, `secret.empty`, or an invalid
+  name for `pipeline.notarius.llm_api_key`, including during inspection or a
+  run that would not execute extraction.
+
+Likely causes:
+
+- the named environment variable is absent or empty;
+- the selected secrets directory lacks a file with that exact name;
+- a process environment variable exists but is empty, so bootstrap correctly
+  leaves it in place; or
+- the Narratio reference name does not match `api_key_env` in the selected
+  Notarius/PromptKit profile.
+
+Safe fix:
+
+- set the exact nonempty environment variable or provide an equivalently named
+  file through the conventional or explicit secrets directory;
+- remove or set the field to `null` when that deployment needs no credential;
+  and
+- use `config show`, `config sources`, and `config diff` to inspect the public
+  reference name and ownership without exposing its value.
+
+Narratio resolves selected credential references during ordinary pipeline
+loading, before stage selection or resume. Credential changes alone do not
+invalidate a completed extraction, but the current reference must still resolve
+for the command to proceed.
+
+Relevant references: [Secrets](./config.md#secrets-handling) and
+[Notarius credential](./config.md#notarius-credential).
+
 ## Notarius executable missing
 
 Symptom:
@@ -289,6 +417,59 @@ Do not point Notarius directly at the original source path or edit the manifest
 checksum. Relevant references: [Notarius reference configuration](./config.md#notarius-reference-bindings)
 and [Operations: Extraction Workflow](./operations.md#extraction-workflow).
 
+## Notarius reference snapshot cache conflict
+
+Symptom:
+
+- extraction reports that a path beneath
+  `cache/notarius/references/<sha256>/` is a symlink, directory, or another
+  non-regular or unsafe entry.
+
+Safe fix:
+
+- stop commands for the affected session and inspect the exact reported path;
+- investigate unexpected links or special files as possible tampering; and
+- remove only the reported conflicting leaf or digest directory after
+  confirming it belongs to the reconstructible reference cache.
+
+A missing snapshot or corrupt regular snapshot is reconstructed automatically
+from currently verified prepared input. Do not recursively delete the session
+root, edit the manifest checksum, or remove cache entries while extraction is
+active.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Notarius resume executes work instead of reusing it
+
+Symptom:
+
+- Narratio executes `extract`, and Notarius makes provider calls despite
+  `pipeline.notarius.resume: true`.
+
+Likely causes:
+
+- checkpoint recording is disabled in the Notarius configuration;
+- Notarius is using a nonpersistent or different checkpoint root;
+- transcript/reference bytes, reference paths, pipeline/profile content,
+  local-backend endpoint, selected lanes, or another checkpoint identity input
+  changed; or
+- Narratio extraction was forced, which deliberately omits `--resume`.
+
+Safe fix:
+
+- validate the deployed Notarius configuration and configure explicit
+  persistent chunk-plan and checkpoint roots;
+- inspect Notarius checkpoint decision reason codes for a legitimate cold miss;
+  and
+- retry without force only when reuse is intended.
+
+Remember that Narratio skipping a completed `extract` stage is separate from
+Notarius reusing work inside an extraction that does execute. Narratio does not
+read the Notarius configuration to infer checkpoint enablement.
+
+Relevant references: [Notarius cache controls](./config.md#notarius-cache-controls)
+and [Notarius integration](./integrations/notarius.md).
+
 ## Notarius reference selector or generated-handoff collision
 
 Symptom:
@@ -361,10 +542,10 @@ Symptom:
 
 Likely causes:
 
-- the executable/config path, pipeline ID, timeout, working directory, or
-  configured output contracts changed;
+- the pipeline ID or configured output contracts changed;
 - a configured prepared reference selector, source, path, checksum, or byte
   size changed;
+- the direct trimmed transcript identity or bytes changed; or
 - the durable bundle, index, lane set, provenance, regular-file status, or
   checksum no longer validates.
 
@@ -374,6 +555,11 @@ Safe fix:
   an unsafe path or bundle symlink error as filesystem corruption or tampering
   and investigate it rather than replacing files manually.
 
+Executable, config path, timeout, working directory, and Notarius credential
+changes are operational and do not by themselves invalidate reusable extraction
+evidence. A newly selected credential must nevertheless resolve during command
+configuration loading.
+
 ## Notarius transitive configuration changed
 
 Symptom:
@@ -488,18 +674,22 @@ Symptom:
   or unsupported; or
 - restore rejects a committed remote session manifest for the same reason.
 
-Likely cause:
+Likely causes:
 
 - the manifest predates the supported canonical state contract. Narratio
   rejects it at the schema boundary rather than interpreting aggregate output
-  lists or incidental files as current authority.
+  lists or incidental files as current authority; or
+- an older strict-decoding Narratio binary is reading a current manifest that
+  contains additive transcribe speaker state it does not recognize.
 
 Safe fix:
 
 - for unsupported local state, move the old manifest aside and regenerate, or
   restore a verified current committed snapshot when one is authoritative;
 - for unsupported committed remote state, regenerate healthy local state as
-  needed and republish the session.
+  needed and republish the session; and
+- for a downgrade rejection, return to the current binary or restore a verified
+  pre-upgrade backup. Do not delete only the speaker fields by hand.
 
 ```bash
 narratio run 2026-04-04
@@ -511,7 +701,7 @@ old files into canonical paths merely to influence selection; availability is
 manifest-owned.
 
 Relevant references: [Operations: Stage Execution and Continuation Behavior](./operations.md#stage-execution-and-continuation-behavior)
-and [Manifest Internals](./internal/manifest.md#analyze-owned-artifact-state).
+and [Manifest Internals](./internal/manifest.md#transcribe-owned-speaker-state).
 
 ## Scriptorium private input changed without a rerun
 
@@ -717,7 +907,8 @@ Symptom:
 
 Likely causes:
 
-- invalid `pipeline.secrets.env_dir`;
+- missing, unreadable, or invalid explicit `--secrets-dir`;
+- invalid entries in the conventional `secrets/` directory beside the root pipeline;
 - missing credential environment variables;
 - invalid S3 endpoint/bucket settings.
 
@@ -725,12 +916,12 @@ Diagnostics:
 
 ```bash
 ls -la /path/to/secrets_dir
-env | sed 's/=.*//' | grep -E 'OBJECT_STORAGE|AWS|AUDITA|SCRIPTORIUM'
+env | sed 's/=.*//' | grep -E 'OBJECT_STORAGE|AWS|AUDITA|NOTARIUS|SCRIPTORIUM'
 ```
 
 Safe fix:
 
-- correct secret-file path and permissions;
+- correct `--secrets-dir`, secret-file type/name, and operator-managed permissions;
 - provide required env vars;
 - keep secret values out of YAML.
 

examples/README.md

@@ -11,18 +11,26 @@ in the [configuration reference](../docs/config.md).
   WhisperX URL.
 - [Production-shaped pipeline](pipeline.production.yml): S3 storage, publish,
   external tools, and configured Scriptorium artifacts.
-- [Production/testing split bundle](production-testing/pipeline.yml): explicit
-  `conf.d` imports, a production default, and selectable production/testing
-  overlays. It also demonstrates canonical-party artifact families and a
-  testing-only disabled artifact. Validate it with:
+- [Production-base/testing-overlay bundle](production-testing/pipeline.yml):
+  explicit `conf.d` imports, one ordinary artifact per `artifacts/<key>.yml`,
+  one family per `artifact-families/<key>.yml`, and the fixed
+  `profiles/testing.yml` overlay. Production values live in the base. The
+  bundle demonstrates an input-only artifact dependency, two character families
+  sharing the campaign party, and a testing-only disabled artifact. Inspect it
+  offline with:
 
   ```sh
   narratio config validate --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml
-  narratio config diff production testing --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml
+  narratio config show --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config sources --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config artifacts --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml --profile testing
+  narratio config diff default testing --config examples/production-testing/pipeline.yml --campaign-file examples/campaigns/sample-campaign/campaign.yml
   ```
 
-  `config show` and `config sources` accept the same selectors and remain
-  read-only.
+  These inspection commands load configuration and resolve explicit credential
+  references, but do not load a session or create runtime state. This bundle is
+  intentionally local-storage and credential-reference free so the walkthrough
+  needs no tools or cloud services.
 - [Full annotated pipeline](pipeline.full.annotated.yml): every implemented
   pipeline section with explanatory comments.
 - [Extraction subset pipeline](pipeline.extraction-subset.yml): a focused
@@ -52,11 +60,20 @@ with the sample campaign and a compatible local- or S3-audio session.
   [glossary](campaigns/sample-campaign/glossary.yml), and canonical
   [party](campaigns/sample-campaign/party.yml) fixture, plus an optional
   [spell-catalog overlay](campaigns/sample-campaign/spell_catalog.json) that
-  follows the Notarius v0.6 contract. Narratio derives the players projection
+  follows the Notarius v0.7 contract. Narratio derives the players projection
   from this party source.
 - [Sample speaker audio](audio/sample-speaker.flac) is a text placeholder that
   reserves the expected filename and directory shape. Replace it with a real
   FLAC file before running transcription.
 
-The examples contain environment-variable names but no credential values. They
-use fictional campaign content and reserved example domains.
+The examples contain environment-variable names but no credential values. Set
+`OBJECT_STORAGE_KEY_ID`, `OBJECT_STORAGE_KEY`, `AUDITA_LLM_API_KEY`, and
+`NOTARIUS_LLM_API_KEY` to nonempty synthetic or real values before loading the
+examples that reference them. They use fictional campaign content and reserved
+example domains.
+
+For a deployed bundle, keep credential files out of source control in a
+`secrets/` directory beside the root `pipeline.yml`, one file per referenced
+environment-variable name, or select another directory with `--secrets-dir`.
+Static S3 credentials are optional: omit both S3 credential references to use
+the AWS SDK discovery chain.

examples/pipeline.full.annotated.yml

@@ -7,9 +7,8 @@ workspace:
   # Optional: remove run-scoped workdir after successful publish commit.
   cleanup_after_publish: false
 
-# Optional: local secret file loader (directory of ENV_VAR_NAME files).
-# secrets:
-#   env_dir: ./secrets
+# Credential files, when used, belong in a sibling secrets/ directory or in the
+# directory selected with --secrets-dir; they are not pipeline configuration.
 
 storage:
   # Defaults to "local". Use "s3" explicitly for publish + S3 audio workflows.
@@ -24,8 +23,8 @@ storage:
     endpoint: ""
     force_path_style: false
     # Optional; defaults shown explicitly.
-    access_key_id_env: OBJECT_STORAGE_KEY_ID
-    secret_access_key_env: OBJECT_STORAGE_KEY
+    access_key_id: OBJECT_STORAGE_KEY_ID
+    secret_access_key: OBJECT_STORAGE_KEY
 
 campaigns:
   # Optional; defaults to /usr/local/share/narratio/campaigns.
@@ -94,7 +93,7 @@ audita:
   # Optional overrides; defaults shown explicitly where applicable.
   binary: audita
   timeout: 3h
-  llm_api_key_env: AUDITA_LLM_API_KEY
+  llm_api_key: AUDITA_LLM_API_KEY
   modules: [glossary, homophones, spoken_word, grammar]
   base_url: ""
   model: ""
@@ -129,13 +128,26 @@ trim:
     report: false
 
 notarius:
-  # Optional structured extraction between trim and render.
+  # Optional structured extraction after render and before analyze.
   enabled: true
   binary: notarius
+  # Optional environment-variable reference. Match this name to api_key_env in
+  # the selected Notarius/PromptKit profile; never put the credential here.
+  llm_api_key: NOTARIUS_LLM_API_KEY
+  # Optionally select a Notarius ConfigKit overlay when extraction runs. This
+  # is independent of Narratio's selected profile and may use a different name.
+  # See docs/config.md and docs/integrations/notarius.md for the contracts.
+  # config_profile: downstream-economy
   config_path: /usr/local/etc/notarius/config.yml
   pipeline_id: dnd-session
   timeout: 3h
   working_directory: /usr/local/etc/notarius
+  # Reuse compatible accepted Notarius checkpoints during executed extraction.
+  # The Notarius config must enable recording and provide persistent cache roots.
+  resume: true
+  # auto uses or creates a source-addressed chunk plan. Forced Narratio
+  # extraction overrides this with refresh and disables checkpoint loading.
+  chunk_cache: auto
   # External campaign references use prepared Narratio source IDs. Omit an
   # optional binding when the selected Notarius pipeline does not need it.
   references:

examples/pipeline.production.yml

@@ -8,8 +8,8 @@ storage:
     bucket: my-dnd-archive
     root_prefix: dnd
     region: us-east-1
-    access_key_id_env: OBJECT_STORAGE_KEY_ID
-    secret_access_key_env: OBJECT_STORAGE_KEY
+    access_key_id: OBJECT_STORAGE_KEY_ID
+    secret_access_key: OBJECT_STORAGE_KEY
 
 campaigns:
   root: /usr/local/share/narratio/campaigns
@@ -57,7 +57,7 @@ seriatim:
 audita:
   binary: audita
   timeout: 3h
-  llm_api_key_env: AUDITA_LLM_API_KEY
+  llm_api_key: AUDITA_LLM_API_KEY
   modules: [glossary, homophones, spoken_word, grammar]
   output_schema: audita-v1
   work_dir_retention: auto

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

@@ -1,23 +1,11 @@
 scriptorium:
-  binary: scriptorium
-  config_path: ./scriptorium/config.yml
   artifact_families:
-    character_meta:
-      enabled: true
-      for_each: party.characters
-      prompt_id: dnd.character_meta
-      output_path_pattern: artifacts/characters/{character_id}/meta.md
-      member_vars:
-        character_id: character_id
-        character_name: character.name
-        player_name: player.name
-        class_summary: character.class_summary
     character_items:
       enabled: true
       for_each: party.characters
       prompt_id: dnd.character_items
+      profile_id: production-placeholder
       output_path_pattern: artifacts/characters/{character_id}/items.md
-      member_dependencies: [character_meta]
       inputs:
         character_meta:
           source: narratio.member_artifact.character_meta

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

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

examples/production-testing/artifacts/player_handout.yml

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

examples/production-testing/artifacts/session_recap.yml

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

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

@@ -10,6 +10,8 @@ audita:
   binary: audita
   modules: [glossary, grammar]
   output_schema: audita-v1
+  model: narratio-production-model-placeholder
+  validation_model: narratio-production-validator-placeholder
 
 normalize:
   output_path: transcripts/final.json

examples/production-testing/pipeline.yml

@@ -1,15 +1,14 @@
-# Copyable production/testing pipeline entry point. Every fragment is named
-# explicitly; Narratio never scans conf.d automatically.
-composition:
+# Copyable production-base/testing-overlay entry point. ConfigKit expands only
+# these explicit path and glob imports; Narratio never scans directories itself.
+configkit:
   imports:
-    - conf.d/platform.yml
-    - conf.d/storage.yml
-    - conf.d/transcript.yml
-    - conf.d/artifacts.yml
-    - conf.d/publish.yml
-  default_profile: production
-  profiles:
-    production:
-      overlay: profiles/production.yml
-    testing:
-      overlay: profiles/testing.yml
+    - path: conf.d/platform.yml
+    - path: conf.d/storage.yml
+    - path: conf.d/transcript.yml
+    - path: conf.d/publish.yml
+    - glob: artifacts/*.yml
+    - glob: artifact-families/*.yml
+
+scriptorium:
+  binary: scriptorium
+  config_path: ./scriptorium/config.yml

examples/production-testing/profiles/production.yml

@@ -1,10 +0,0 @@
-audita:
-  model: narratio-production-model-placeholder
-  validation_model: narratio-production-validator-placeholder
-
-scriptorium:
-  artifact_families:
-    character_meta:
-      profile_id: production-placeholder
-    character_items:
-      profile_id: production-placeholder

examples/production-testing/profiles/testing.yml

@@ -4,6 +4,10 @@ audita:
 
 scriptorium:
   artifacts:
+    session_recap:
+      profile_id: testing-placeholder
+    player_handout:
+      profile_id: testing-placeholder
     testing_notes:
       enabled: false
       output_path: artifacts/testing-notes.md

go.mod

@@ -3,6 +3,7 @@ module gitea.maximumdirect.net/eric/narratio
 go 1.26.0
 
 require (
+	gitea.maximumdirect.net/eric/ConfigKit v0.3.1
 	github.com/aws/aws-sdk-go-v2/config v1.32.17
 	github.com/aws/aws-sdk-go-v2/credentials v1.19.16
 	github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0
@@ -28,4 +29,5 @@ require (
 	github.com/aws/aws-sdk-go-v2/service/sso v1.30.17 // indirect
 	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21 // indirect
 	github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 // indirect
+	go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect
 )

go.sum

@@ -1,3 +1,5 @@
+gitea.maximumdirect.net/eric/ConfigKit v0.3.1 h1:etfTI/dxO3NnS18La0V6IeUh/Iv+3PjVfEyaN8hTuOM=
+gitea.maximumdirect.net/eric/ConfigKit v0.3.1/go.mod h1:tYWpsrbLyu05r8+1MapWp5fN6ZrX9I8hU2xEPedYmYk=
 github.com/aws/aws-sdk-go-v2 v1.41.7 h1:DWpAJt66FmnnaRIOT/8ASTucrvuDPZASqhhLey6tLY8=
 github.com/aws/aws-sdk-go-v2 v1.41.7/go.mod h1:4LAfZOPHNVNQEckOACQx60Y8pSRjIkNZQz1w92xpMJc=
 github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10 h1:gx1AwW1Iyk9Z9dD9F4akX5gnN3QZwUB20GGKH/I+Rho=
@@ -36,6 +38,8 @@ github.com/aws/smithy-go v1.25.1 h1:J8ERsGSU7d+aCmdQur5Txg6bVoYelvQJgtZehD12GkI=
 github.com/aws/smithy-go v1.25.1/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
+go.yaml.in/yaml/v4 v4.0.0-rc.6 h1:1h7H1ohdUh93/FyE4YaDa1Zh64K6VVbjF4K6WUxMtH4=
+go.yaml.in/yaml/v4 v4.0.0-rc.6/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0=
 golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
 golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=

internal/adapters/audita/subprocess.go

@@ -5,7 +5,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/url"
-	"os"
 	"strconv"
 	"strings"
 	"time"
@@ -24,7 +23,8 @@ const MaxReportOutputBytes int64 = 16 * 1024 * 1024
 type SubprocessRunnerConfig struct {
 	Binary                   string
 	Timeout                  time.Duration
-	LLMAPIKeyEnv             string
+	LLMAPIKey                string
+	LLMAPIKeyReference       string
 	Modules                  []string
 	BaseURL                  string
 	Model                    string
@@ -43,7 +43,8 @@ type SubprocessRunnerConfig struct {
 type SubprocessRunner struct {
 	binary                   string
 	timeout                  time.Duration
-	llmAPIKeyEnv             string
+	llmAPIKey                string
+	llmAPIKeyReference       string
 	modules                  []string
 	baseURL                  string
 	model                    string
@@ -62,7 +63,8 @@ type SubprocessRunner struct {
 func NewSubprocessRunnerFromConfigValues(
 	binary string,
 	timeout string,
-	llmAPIKeyEnv string,
+	llmAPIKey string,
+	llmAPIKeyReference string,
 	modules []string,
 	baseURL string,
 	model string,
@@ -86,7 +88,8 @@ func NewSubprocessRunnerFromConfigValues(
 	return NewSubprocessRunner(SubprocessRunnerConfig{
 		Binary:                   binary,
 		Timeout:                  parsedTimeout,
-		LLMAPIKeyEnv:             llmAPIKeyEnv,
+		LLMAPIKey:                llmAPIKey,
+		LLMAPIKeyReference:       llmAPIKeyReference,
 		Modules:                  modules,
 		BaseURL:                  baseURL,
 		Model:                    model,
@@ -110,6 +113,9 @@ func NewSubprocessRunner(cfg SubprocessRunnerConfig) (*SubprocessRunner, error)
 	if cfg.Timeout <= 0 {
 		return nil, fmt.Errorf("audita timeout must be > 0")
 	}
+	if (cfg.LLMAPIKey == "") != (strings.TrimSpace(cfg.LLMAPIKeyReference) == "") {
+		return nil, fmt.Errorf("audita credential value and reference must be provided together")
+	}
 	for i, module := range cfg.Modules {
 		if strings.TrimSpace(module) == "" {
 			return nil, fmt.Errorf("audita module at index %d is empty", i)
@@ -152,7 +158,8 @@ func NewSubprocessRunner(cfg SubprocessRunnerConfig) (*SubprocessRunner, error)
 	return &SubprocessRunner{
 		binary:                   strings.TrimSpace(cfg.Binary),
 		timeout:                  cfg.Timeout,
-		llmAPIKeyEnv:             strings.TrimSpace(cfg.LLMAPIKeyEnv),
+		llmAPIKey:                cfg.LLMAPIKey,
+		llmAPIKeyReference:       strings.TrimSpace(cfg.LLMAPIKeyReference),
 		modules:                  modules,
 		baseURL:                  strings.TrimSpace(cfg.BaseURL),
 		model:                    strings.TrimSpace(cfg.Model),
@@ -195,16 +202,10 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 	}
 	args := r.buildArgs(req, reqModules)
 
-	credentialEnvVar := strings.TrimSpace(r.llmAPIKeyEnv)
-	credentialPresent := false
+	credentialPresent := r.llmAPIKeyReference != ""
 	env := map[string]string{}
-	if credentialEnvVar != "" {
-		credential, ok := os.LookupEnv(credentialEnvVar)
-		if !ok || strings.TrimSpace(credential) == "" {
-			return PolishResult{}, fmt.Errorf("audita: required credential environment variable %s is not set", credentialEnvVar)
-		}
-		env["AUDITA_LLM_API_KEY"] = credential
-		credentialPresent = true
+	if credentialPresent {
+		env["AUDITA_LLM_API_KEY"] = r.llmAPIKey
 	}
 
 	if req.GeneratedConfigPath != "" {
@@ -269,7 +270,7 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 			"total_llm_concurrency":      r.totalLLMConcurrency,
 			"proposal_llm_concurrency":   r.proposalLLMConcurrency,
 			"validation_llm_concurrency": r.validationLLMConcurrency,
-			"credential_env_var":         r.llmAPIKeyEnv,
+			"credential_env_var":         r.llmAPIKeyReference,
 			"credential_present":         credentialPresent,
 		},
 	}, nil
@@ -299,7 +300,7 @@ func (r *SubprocessRunner) failureResult(req PolishRequest, modules []string, ru
 			"total_llm_concurrency":      r.totalLLMConcurrency,
 			"proposal_llm_concurrency":   r.proposalLLMConcurrency,
 			"validation_llm_concurrency": r.validationLLMConcurrency,
-			"credential_env_var":         r.llmAPIKeyEnv,
+			"credential_env_var":         r.llmAPIKeyReference,
 			"credential_present":         credentialPresent,
 		},
 	}
@@ -375,7 +376,7 @@ func (r *SubprocessRunner) writeInvocationConfig(req PolishRequest, args []strin
 		"output_path":                req.OutputProcessedPath,
 		"report_path":                req.ReportPath,
 		"work_dir":                   req.WorkDir,
-		"credential_env_var":         r.llmAPIKeyEnv,
+		"credential_env_var":         r.llmAPIKeyReference,
 		"credential_present":         credentialPresent,
 	}
 	return writeGeneratedYAML(req.GeneratedConfigPath, payload)

internal/adapters/audita/subprocess_test.go

@@ -6,16 +6,12 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
 func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
@@ -31,7 +27,8 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	runner, err := NewSubprocessRunner(SubprocessRunnerConfig{
 		Binary:                   wrapper,
 		Timeout:                  mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv:             "OPENAI_KEY_SOURCE",
+		LLMAPIKey:                "super-secret",
+		LLMAPIKeyReference:       "OPENAI_KEY_SOURCE",
 		Modules:                  []string{"glossary", "homophones", "glossary"},
 		BaseURL:                  "https://openrouter.ai/api/v1",
 		Model:                    "openrouter/google/gemma-4-31b-it",
@@ -48,6 +45,7 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	if err != nil {
 		t.Fatalf("NewSubprocessRunner() error = %v", err)
 	}
+	t.Setenv("OPENAI_KEY_SOURCE", "changed-after-resolution")
 
 	dir := t.TempDir()
 	req := PolishRequest{
@@ -132,47 +130,31 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	}
 }
 
-func TestSubprocessRunnerMissingConfiguredCredentialFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
-	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "MISSING_AUDITA_KEY",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+func TestSubprocessRunnerRejectsIncompleteResolvedCredential(t *testing.T) {
+	_, err := NewSubprocessRunner(SubprocessRunnerConfig{
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKeyReference: "MISSING_AUDITA_KEY",
 	})
-	req := auditaReqForTest(t, false)
-
-	_, err := runner.Run(context.Background(), req)
-	if err == nil {
-		t.Fatal("expected error, got nil")
-	}
-	if !strings.Contains(err.Error(), "required credential environment variable MISSING_AUDITA_KEY is not set") {
-		t.Fatalf("error = %q, want missing credential guidance", err.Error())
+	if err == nil || !strings.Contains(err.Error(), "credential value and reference must be provided together") {
+		t.Fatalf("NewSubprocessRunner() error = %v, want credential pair error", err)
 	}
 }
 
 func TestSubprocessRunnerUnconfiguredCredentialEnvOmitsCredential(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		Modules:   []string{"glossary"},
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
 	})
 	req := auditaReqForTest(t, false)
 
@@ -190,9 +172,6 @@ func TestSubprocessRunnerUnconfiguredCredentialEnvOmitsCredential(t *testing.T)
 }
 
 func TestSubprocessRunnerOmitsUnspecifiedParentEnvironment(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	t.Setenv("AUDITA_INHERITED_MARKER", "inherited-from-parent")
@@ -200,13 +179,13 @@ func TestSubprocessRunnerOmitsUnspecifiedParentEnvironment(t *testing.T) {
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
-		LLMAPIKeyEnv: "",
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		Modules:   []string{"glossary"},
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
+		LLMAPIKey: "",
 	})
 	req := auditaReqForTest(t, false)
 	if _, err := runner.Run(context.Background(), req); err != nil {
@@ -220,21 +199,18 @@ func TestSubprocessRunnerOmitsUnspecifiedParentEnvironment(t *testing.T) {
 }
 
 func TestSubprocessRunnerOmitsModulesFlagWhenNotConfigured(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "",
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		BaseURL:   "https://openrouter.ai/api/v1",
+		Model:     "openrouter/google/gemma-4-31b-it",
+		Report:    false,
 	})
 	req := auditaReqForTest(t, false)
 	if _, err := runner.Run(context.Background(), req); err != nil {
@@ -249,19 +225,16 @@ func TestSubprocessRunnerOmitsModulesFlagWhenNotConfigured(t *testing.T) {
 }
 
 func TestSubprocessRunnerOmitsBaseURLAndModelFlagsWhenNotConfigured(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "",
-		Report:       false,
+		Binary:    writeAuditaHelperWrapper(t),
+		Timeout:   mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey: "",
+		Report:    false,
 	})
 	req := auditaReqForTest(t, false)
 	if _, err := runner.Run(context.Background(), req); err != nil {
@@ -279,22 +252,20 @@ func TestSubprocessRunnerOmitsBaseURLAndModelFlagsWhenNotConfigured(t *testing.T
 }
 
 func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "fail")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -310,25 +281,26 @@ func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
 	if !strings.Contains(err.Error(), req.StdoutLogPath) || !strings.Contains(err.Error(), req.StderrLogPath) {
 		t.Fatalf("error = %q, want stdout/stderr log paths", err.Error())
 	}
+	if strings.Contains(err.Error(), "super-secret") {
+		t.Fatalf("failure diagnostic exposed credential value: %q", err.Error())
+	}
 }
 
 func TestSubprocessRunnerSubprocessFailureAddsStderrDescriptorHint(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "fail_bad_fd")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -341,22 +313,20 @@ func TestSubprocessRunnerSubprocessFailureAddsStderrDescriptorHint(t *testing.T)
 }
 
 func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "missing_output")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -369,22 +339,20 @@ func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "invalid_output")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -397,22 +365,20 @@ func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerSegmentsMissingFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "segments_missing")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       false,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             false,
 	})
 	req := auditaReqForTest(t, false)
 	_, err := runner.Run(context.Background(), req)
@@ -425,22 +391,20 @@ func TestSubprocessRunnerSegmentsMissingFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
 	t.Setenv("AUDITA_HELPER_MODE", "invalid_report")
 	t.Setenv("OPENAI_KEY_SOURCE", "super-secret")
 	t.Setenv("AUDITA_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
 
 	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
-		Binary:       writeAuditaHelperWrapper(t),
-		Timeout:      mustParseAuditaDuration(t, "2s"),
-		LLMAPIKeyEnv: "OPENAI_KEY_SOURCE",
-		Modules:      []string{"glossary"},
-		BaseURL:      "https://openrouter.ai/api/v1",
-		Model:        "openrouter/google/gemma-4-31b-it",
-		Report:       true,
+		Binary:             writeAuditaHelperWrapper(t),
+		Timeout:            mustParseAuditaDuration(t, "2s"),
+		LLMAPIKey:          "super-secret",
+		LLMAPIKeyReference: "OPENAI_KEY_SOURCE",
+		Modules:            []string{"glossary"},
+		BaseURL:            "https://openrouter.ai/api/v1",
+		Model:              "openrouter/google/gemma-4-31b-it",
+		Report:             true,
 	})
 	req := auditaReqForTest(t, true)
 	_, err := runner.Run(context.Background(), req)
@@ -453,21 +417,22 @@ func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerConstructorValidation(t *testing.T) {
-	_, err := NewSubprocessRunnerFromConfigValues("", "3h", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
+	_, err := NewSubprocessRunnerFromConfigValues("", "3h", "synthetic-secret", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
 	if err == nil {
 		t.Fatal("expected binary validation error")
 	}
-	_, err = NewSubprocessRunnerFromConfigValues("audita", "bad", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
+	_, err = NewSubprocessRunnerFromConfigValues("audita", "bad", "synthetic-secret", "AUDITA_LLM_API_KEY", []string{"glossary"}, "https://openrouter.ai/api/v1", "openrouter/google/gemma-4-31b-it", "", "", "", "", nil, nil, "", nil, true)
 	if err == nil {
 		t.Fatal("expected timeout parse error")
 	}
 	_, err = NewSubprocessRunner(SubprocessRunnerConfig{
-		Binary:       "audita",
-		Timeout:      mustParseAuditaDuration(t, "1s"),
-		LLMAPIKeyEnv: "AUDITA_LLM_API_KEY",
-		Modules:      []string{"glossary"},
-		BaseURL:      "://",
-		Model:        "openrouter/google/gemma-4-31b-it",
+		Binary:             "audita",
+		Timeout:            mustParseAuditaDuration(t, "1s"),
+		LLMAPIKey:          "synthetic-secret",
+		LLMAPIKeyReference: "AUDITA_LLM_API_KEY",
+		Modules:            []string{"glossary"},
+		BaseURL:            "://",
+		Model:              "openrouter/google/gemma-4-31b-it",
 	})
 	if err == nil {
 		t.Fatal("expected base url validation error")

internal/adapters/notarius/fake.go

@@ -15,6 +15,10 @@ func (f *FakeRunner) Run(ctx context.Context, req RunRequest) (RunResult, error)
 		return RunResult{}, err
 	}
 	copyRequest := req
+	if req.ConfigProfile != nil {
+		profile := *req.ConfigProfile
+		copyRequest.ConfigProfile = &profile
+	}
 	copyRequest.References = append([]ReferenceBinding(nil), req.References...)
 	f.Requests = append(f.Requests, copyRequest)
 	if f.Err != nil {

internal/adapters/notarius/runner.go

@@ -4,6 +4,8 @@ package notarius
 import (
 	"context"
 	"time"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 )
 
 const ReceiptSchemaVersion = "notarius.run-result.v2"
@@ -22,7 +24,10 @@ type ReferenceBinding struct {
 
 // RunRequest contains the resolved inputs and diagnostic destinations for one invocation.
 type RunRequest struct {
-	Binary           string
+	Binary string
+	// LLMAPIKey is an optional already-resolved credential reference.
+	LLMAPIKey        *configkit.Secret
+	ConfigProfile    *string
 	ConfigPath       string
 	PipelineID       string
 	InputPath        string
@@ -31,6 +36,8 @@ type RunRequest struct {
 	ReceiptPath      string
 	LogPath          string
 	Timeout          time.Duration
+	Resume           bool
+	ChunkCache       string
 	References       []ReferenceBinding
 }
 

internal/adapters/notarius/subprocess.go

@@ -52,6 +52,20 @@ func (r *SubprocessRunner) Run(ctx context.Context, req RunRequest) (RunResult,
 	if err != nil {
 		return RunResult{}, err
 	}
+	var envOverrides map[string]string
+	var sensitiveEnvNames []string
+	if req.LLMAPIKey != nil {
+		if !req.LLMAPIKey.IsResolved() {
+			return RunResult{}, fmt.Errorf("notarius LLM API credential must be resolved and nonempty")
+		}
+		name := req.LLMAPIKey.EnvName()
+		value := req.LLMAPIKey.Value()
+		if value == "" {
+			return RunResult{}, fmt.Errorf("notarius LLM API credential must be resolved and nonempty")
+		}
+		envOverrides = map[string]string{name: value}
+		sensitiveEnvNames = []string{name}
+	}
 
 	args := []string{
 		"run", req.PipelineID,
@@ -59,18 +73,29 @@ func (r *SubprocessRunner) Run(ctx context.Context, req RunRequest) (RunResult,
 		"--input", req.InputPath,
 		"--output-dir", req.OutputRoot,
 	}
+	if req.Resume {
+		args = append(args, "--resume")
+	}
+	if req.ChunkCache != "" {
+		args = append(args, "--chunk_cache", req.ChunkCache)
+	}
 	for _, reference := range references {
 		args = append(args, "--reference", reference.Selector+"="+reference.Path)
 	}
+	if req.ConfigProfile != nil {
+		args = append(args, "--config-profile", *req.ConfigProfile)
+	}
 	args = append(args, "--json")
 	processResult, err := r.run(ctx, subprocess.RunRequest{
-		Executable:      req.Binary,
-		Args:            args,
-		WorkingDir:      req.WorkingDirectory,
-		Timeout:         req.Timeout,
-		DiagnosticOwner: "notarius",
-		StdoutLogPath:   req.ReceiptPath,
-		StderrLogPath:   req.LogPath,
+		Executable:        req.Binary,
+		Args:              args,
+		WorkingDir:        req.WorkingDirectory,
+		Timeout:           req.Timeout,
+		DiagnosticOwner:   "notarius",
+		StdoutLogPath:     req.ReceiptPath,
+		StderrLogPath:     req.LogPath,
+		EnvOverrides:      envOverrides,
+		SensitiveEnvNames: sensitiveEnvNames,
 	})
 	baseResult := RunResult{
 		ReceiptPath: req.ReceiptPath,
@@ -143,6 +168,9 @@ func validateRunRequest(req RunRequest) ([]ReferenceBinding, error) {
 	if req.Timeout <= 0 {
 		return nil, fmt.Errorf("notarius timeout must be positive")
 	}
+	if !validChunkCacheMode(req.ChunkCache) {
+		return nil, fmt.Errorf("notarius chunk cache mode must be empty, auto, bypass, or refresh")
+	}
 	for label, path := range map[string]string{
 		"config":            req.ConfigPath,
 		"input":             req.InputPath,
@@ -201,6 +229,15 @@ func validateRunRequest(req RunRequest) ([]ReferenceBinding, error) {
 	return references, nil
 }
 
+func validChunkCacheMode(mode string) bool {
+	switch mode {
+	case "", "auto", "bypass", "refresh":
+		return true
+	default:
+		return false
+	}
+}
+
 type receiptDocument struct {
 	SchemaVersion             string                      `json:"schema_version"`
 	RunID                     string                      `json:"run_id"`
@@ -248,7 +285,7 @@ func validateValidationSummaries(documents []validationSummaryDocument) ([]Valid
 		if document.Status != "complete" && document.Status != "rejected" && document.Status != "incomplete" {
 			return nil, fmt.Errorf("notarius validation summary status %q is invalid", document.Status)
 		}
-		if document.ProducerAttemptCount == nil || *document.ProducerAttemptCount <= 0 || !validTerminalAction(document.TerminalAction) {
+		if document.ProducerAttemptCount == nil || *document.ProducerAttemptCount < 0 || !validTerminalAction(document.TerminalAction) {
 			return nil, fmt.Errorf("notarius validation summary is missing required fields")
 		}
 		summaries = append(summaries, ValidationSummary{

internal/adapters/notarius/subprocess_test.go

@@ -11,6 +11,7 @@ import (
 	"testing"
 	"time"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	sharedsubprocess "gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
 )
 
@@ -76,6 +77,174 @@ func TestSubprocessRunnerBuildsExactInvocationAndDiscoversBundle(t *testing.T) {
 	}
 }
 
+func TestSubprocessRunnerMapsCacheControlsToArguments(t *testing.T) {
+	tests := []struct {
+		name       string
+		resume     bool
+		chunkCache string
+		want       []string
+	}{
+		{name: "resume", resume: true, want: []string{"--resume"}},
+		{name: "auto", chunkCache: "auto", want: []string{"--chunk_cache", "auto"}},
+		{name: "bypass", chunkCache: "bypass", want: []string{"--chunk_cache", "bypass"}},
+		{name: "refresh", chunkCache: "refresh", want: []string{"--chunk_cache", "refresh"}},
+		{name: "resume and mode", resume: true, chunkCache: "auto", want: []string{"--resume", "--chunk_cache", "auto"}},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.Resume = test.resume
+			req.ChunkCache = test.chunkCache
+			var captured sharedsubprocess.RunRequest
+			runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				captured = processReq
+				writeValidBundleAndReceipt(t, req, false)
+				return sharedsubprocess.RunResult{ExitCode: 0}, nil
+			}}
+
+			if _, err := runner.Run(context.Background(), req); err != nil {
+				t.Fatalf("Run() error = %v", err)
+			}
+			want := []string{
+				"run", req.PipelineID, "--config", req.ConfigPath, "--input", req.InputPath,
+				"--output-dir", req.OutputRoot,
+			}
+			want = append(want, test.want...)
+			want = append(want, "--json")
+			if !reflect.DeepEqual(captured.Args, want) {
+				t.Fatalf("subprocess args = %#v, want %#v", captured.Args, want)
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerMapsConfigProfileToArgumentPair(t *testing.T) {
+	tests := []struct {
+		name     string
+		supplied bool
+		value    string
+	}{
+		{name: "omitted"},
+		{name: "empty", supplied: true},
+		{name: "default", supplied: true, value: "default"},
+		{name: "named", supplied: true, value: "inexpensive"},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			if test.supplied {
+				req.ConfigProfile = &test.value
+			}
+			var captured sharedsubprocess.RunRequest
+			runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				captured = processReq
+				writeValidBundleAndReceipt(t, req, false)
+				return sharedsubprocess.RunResult{ExitCode: 0}, nil
+			}}
+
+			if _, err := runner.Run(context.Background(), req); err != nil {
+				t.Fatalf("Run() error = %v", err)
+			}
+			flagIndex := -1
+			for index, arg := range captured.Args {
+				if arg == "--config-profile" {
+					flagIndex = index
+					break
+				}
+			}
+			if !test.supplied {
+				if flagIndex >= 0 {
+					t.Fatalf("subprocess args unexpectedly contain config profile: %#v", captured.Args)
+				}
+				return
+			}
+			if flagIndex < 0 || flagIndex+2 >= len(captured.Args) || captured.Args[flagIndex+1] != test.value || captured.Args[flagIndex+2] != "--json" {
+				t.Fatalf("config profile argument boundary = %#v, want flag, exact value, then --json", captured.Args)
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerRejectsInvalidChunkCacheBeforeLaunch(t *testing.T) {
+	for _, mode := range []string{"AUTO", "invalid", " auto "} {
+		t.Run(mode, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.ChunkCache = mode
+			started := false
+			runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				started = true
+				return sharedsubprocess.RunResult{}, nil
+			}}
+			_, err := runner.Run(context.Background(), req)
+			if err == nil || !strings.Contains(err.Error(), "chunk cache mode") {
+				t.Fatalf("Run() error = %v, want chunk cache validation failure", err)
+			}
+			if started {
+				t.Fatal("subprocess started with invalid chunk cache mode")
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerForwardsLoadedCredentialWithoutRereadingEnvironment(t *testing.T) {
+	const envName = "NARRATIO_NOTARIUS_CUSTOM_LLM_KEY"
+	const loadedValue = "loaded-notarius-credential"
+	credential := loadTestCredential(t, envName, loadedValue)
+	t.Setenv(envName, "changed-parent-credential")
+
+	req := validRunRequest(t)
+	req.LLMAPIKey = credential
+	var captured sharedsubprocess.RunRequest
+	runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		captured = processReq
+		writeValidBundleAndReceipt(t, req, false)
+		return sharedsubprocess.RunResult{ExitCode: 0}, nil
+	}}
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	if len(captured.EnvOverrides) != 1 || captured.EnvOverrides[envName] != loadedValue {
+		t.Fatal("subprocess did not receive exactly the loaded credential override")
+	}
+	if !reflect.DeepEqual(captured.SensitiveEnvNames, []string{envName}) {
+		t.Fatalf("sensitive environment names = %#v, want configured reference name", captured.SensitiveEnvNames)
+	}
+	wantArgs := []string{
+		"run", req.PipelineID, "--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)
+	}
+}
+
+func TestSubprocessRunnerRejectsInvalidProgrammaticCredentialBeforeLaunch(t *testing.T) {
+	for _, test := range []struct {
+		name      string
+		reference configkit.Secret
+	}{
+		{name: "unresolved", reference: configkit.SecretEnv("NARRATIO_NOTARIUS_UNRESOLVED_KEY")},
+		{name: "empty", reference: configkit.SecretEnv("")},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.LLMAPIKey = &test.reference
+			started := false
+			runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				started = true
+				return sharedsubprocess.RunResult{}, nil
+			}}
+			_, err := runner.Run(context.Background(), req)
+			if err == nil || !strings.Contains(err.Error(), "resolved and nonempty") {
+				t.Fatalf("Run() error = %v, want credential validation failure", err)
+			}
+			if started {
+				t.Fatal("subprocess started with an invalid credential")
+			}
+		})
+	}
+}
+
 func TestSubprocessRunnerBuildsOrderedReferenceArguments(t *testing.T) {
 	req := validRunRequest(t)
 	referenceRoot := t.TempDir()
@@ -172,16 +341,23 @@ func TestSubprocessRunnerRejectsInvalidReferencesBeforeLaunch(t *testing.T) {
 }
 
 func TestSubprocessRunnerUsesMinimalEnvironmentAndSeparatesStreams(t *testing.T) {
+	const envName = "NARRATIO_NOTARIUS_CUSTOM_CAPTURE_KEY"
+	const canary = "notarius-captured-stream-canary"
+	const unrelated = "NARRATIO_NOTARIUS_UNRELATED_PARENT_KEY"
 	req := validRunRequest(t)
+	req.LLMAPIKey = loadTestCredential(t, envName, canary)
 	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)
+	var receipt map[string]any
+	if err := json.Unmarshal(data, &receipt); err != nil {
+		t.Fatalf("decode receipt fixture: %v", err)
 	}
+	receipt["credential_echo"] = canary
+	writeJSONFile(t, receiptFixture, receipt)
 	if err := os.Remove(req.ReceiptPath); err != nil {
 		t.Fatalf("Remove(receipt) error = %v", err)
 	}
@@ -193,27 +369,55 @@ func TestSubprocessRunnerUsesMinimalEnvironmentAndSeparatesStreams(t *testing.T)
 	script := writeShellScript(t, `#!/bin/sh
 pwd > "$NOTARIUS_CAPTURE_DIR/working-directory"
 printf '%s' "$NOTARIUS_INHERITED_VALUE" > "$NOTARIUS_CAPTURE_DIR/environment"
-printf 'diagnostic stream\n' >&2
+if [ "$`+envName+`" != '`+canary+`' ]; then
+  printf 'selected credential mismatch\n' >&2
+  exit 8
+fi
+if [ -n "$`+unrelated+`" ]; then
+  printf 'unrelated credential inherited\n' >&2
+  exit 9
+fi
+printf '%s\n' "$`+envName+`" >&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)
+	t.Setenv(unrelated, "unrelated-parent-credential")
 
 	if _, err := NewSubprocessRunner().Run(context.Background(), req); err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
 	assertTextFile(t, filepath.Join(captureDir, "working-directory"), req.WorkingDirectory+"\n")
 	assertTextFile(t, filepath.Join(captureDir, "environment"), "")
-	assertTextFile(t, req.LogPath, "diagnostic stream\n")
+	assertRedactedCredentialFile(t, req.LogPath, canary)
 	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")
+	assertRedactedCredentialBytes(t, receiptBytes, canary)
+}
+
+func TestSubprocessRunnerRedactsCredentialOnProcessFailure(t *testing.T) {
+	const envName = "NARRATIO_NOTARIUS_CUSTOM_FAILURE_KEY"
+	const canary = "notarius-failure-stream-canary"
+	req := validRunRequest(t)
+	req.LLMAPIKey = loadTestCredential(t, envName, canary)
+	req.Binary = writeShellScript(t, `#!/bin/sh
+printf '%s\n' "$`+envName+`"
+printf '%s\n' "$`+envName+`" >&2
+exit 7
+`)
+	_, err := NewSubprocessRunner().Run(context.Background(), req)
+	if err == nil {
+		t.Fatal("Run() error = nil, want process failure")
 	}
+	if strings.Contains(err.Error(), canary) || !strings.Contains(err.Error(), "<redacted>") {
+		t.Fatal("process failure did not return redacted diagnostics")
+	}
+	assertRedactedCredentialFile(t, req.ReceiptPath, canary)
+	assertRedactedCredentialFile(t, req.LogPath, canary)
 }
 
 func TestSubprocessRunnerReturnsProcessFailuresWithoutParsingStdout(t *testing.T) {
@@ -231,6 +435,10 @@ func TestSubprocessRunnerReturnsProcessFailuresWithoutParsingStdout(t *testing.T
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
 			req := validRunRequest(t)
+			if test.name == "nonzero" {
+				profile := "missing-profile"
+				req.ConfigProfile = &profile
+			}
 			req.Binary = writeShellScript(t, "#!/bin/sh\n"+test.scriptBody)
 			req.Timeout = test.timeout
 			ctx := context.Background()
@@ -527,7 +735,8 @@ func TestLoadDiagnosticSummariesValidateBoundsAndTolerateUnknownFields(t *testin
 }
 
 func TestFakeRunnerCapturesRequestsAndHonorsContextAndError(t *testing.T) {
-	req := RunRequest{PipelineID: "pipeline", References: []ReferenceBinding{{Selector: "party", Path: "/references/party.json"}}}
+	profile := "testing"
+	req := RunRequest{PipelineID: "pipeline", ConfigProfile: &profile, References: []ReferenceBinding{{Selector: "party", Path: "/references/party.json"}}}
 	want := RunResult{BundleRoot: "/bundle"}
 	fake := &FakeRunner{Result: want}
 	got, err := fake.Run(context.Background(), req)
@@ -538,6 +747,10 @@ func TestFakeRunnerCapturesRequestsAndHonorsContextAndError(t *testing.T) {
 	if fake.Requests[0].References[0].Path != "/references/party.json" {
 		t.Fatalf("fake retained aliased request references: %#v", fake.Requests[0].References)
 	}
+	profile = "changed"
+	if *fake.Requests[0].ConfigProfile != "testing" {
+		t.Fatalf("fake retained aliased config profile: %q", *fake.Requests[0].ConfigProfile)
+	}
 
 	wantErr := errors.New("configured failure")
 	fake.Err = wantErr
@@ -579,6 +792,56 @@ func validRunRequest(t *testing.T) RunRequest {
 	}
 }
 
+func loadTestCredential(t *testing.T, envName, value string) *configkit.Secret {
+	t.Helper()
+	t.Setenv(envName, value)
+	type testConfig struct {
+		Credential *configkit.Secret `yaml:"credential"`
+	}
+	result, err := configkit.Load(configkit.Bytes("credential.yml", []byte("credential: "+envName+"\n")), configkit.Options[testConfig]{})
+	if err != nil {
+		t.Fatalf("load test credential: %v", err)
+	}
+	return result.Config.Credential
+}
+
+func assertRedactedCredentialFile(t *testing.T, path, canary string) {
+	t.Helper()
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("ReadFile(%q) error = %v", path, err)
+	}
+	assertRedactedCredentialBytes(t, data, canary)
+}
+
+func assertRedactedCredentialBytes(t *testing.T, data []byte, canary string) {
+	t.Helper()
+	text := string(data)
+	if strings.Contains(text, canary) || !strings.Contains(text, "<redacted>") {
+		t.Fatal("captured stream did not contain the required redaction marker")
+	}
+}
+
+func TestValidateValidationSummariesAcceptsReusedProducerWithoutAttempts(t *testing.T) {
+	zero := 0
+	summaries, err := validateValidationSummaries([]validationSummaryDocument{{
+		Stage: "chunk", Status: "complete", ProducerAttemptCount: &zero, TerminalAction: "accepted",
+	}})
+	if err != nil {
+		t.Fatalf("validateValidationSummaries() error = %v", err)
+	}
+	if len(summaries) != 1 || summaries[0].ProducerAttemptCount != 0 {
+		t.Fatalf("summaries = %#v, want one reused producer with zero attempts", summaries)
+	}
+
+	negative := -1
+	if _, err := validateValidationSummaries([]validationSummaryDocument{{
+		Stage: "chunk", Status: "complete", ProducerAttemptCount: &negative, TerminalAction: "accepted",
+	}}); err == nil {
+		t.Fatal("validateValidationSummaries() negative producer attempts error = nil")
+	}
+}
+
 func writeValidBundleAndReceipt(t *testing.T, req RunRequest, includeUnknown bool) {
 	t.Helper()
 	bundle := filepath.Join(req.OutputRoot, "notarius-run-1")

internal/adapters/scriptorium/subprocess_test.go

@@ -6,16 +6,12 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
 func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "run_success")
@@ -95,9 +91,6 @@ func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *tes
 }
 
 func TestSubprocessRunnerRunExitCodeOneFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "fail_exit1")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -118,9 +111,6 @@ func TestSubprocessRunnerRunExitCodeOneFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRunExitCodeTwoReturnsValidationFailureAndPreservesOutput(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "fail_exit2_with_output")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -144,9 +134,6 @@ func TestSubprocessRunnerRunExitCodeTwoReturnsValidationFailureAndPreservesOutpu
 }
 
 func TestSubprocessRunnerRunMissingOutputOnSuccessFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "run_success_missing_output")
 	t.Setenv("SCRIPTORIUM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -166,9 +153,6 @@ func TestSubprocessRunnerRunMissingOutputOnSuccessFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderSuccess(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SCRIPTORIUM_HELPER", "1")
 	t.Setenv("SCRIPTORIUM_HELPER_MODE", "render_success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")

internal/adapters/seriatim/subprocess_test.go

@@ -6,16 +6,12 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
 func TestSubprocessRunnerSuccessWithReportArgsAndEnv(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "success")
@@ -131,9 +127,6 @@ func TestSubprocessRunnerSuccessWithReportArgsAndEnv(t *testing.T) {
 }
 
 func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "fail")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -153,9 +146,6 @@ func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
 }
 
 func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "missing_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -172,9 +162,6 @@ func TestSubprocessRunnerMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "invalid_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -191,9 +178,6 @@ func TestSubprocessRunnerInvalidOutputJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "invalid_report")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -210,9 +194,6 @@ func TestSubprocessRunnerInvalidReportJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerTrimSuccessInvocationAndProvenance(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "trim_success")
@@ -297,9 +278,6 @@ func TestSubprocessRunnerTrimSuccessInvocationAndProvenance(t *testing.T) {
 }
 
 func TestSubprocessRunnerTrimSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "fail")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -319,9 +297,6 @@ func TestSubprocessRunnerTrimSubprocessFailure(t *testing.T) {
 }
 
 func TestSubprocessRunnerTrimMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "missing_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -338,9 +313,6 @@ func TestSubprocessRunnerTrimMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerTrimInvalidOutputJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "invalid_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -357,9 +329,6 @@ func TestSubprocessRunnerTrimInvalidOutputJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerTrimOutputMissingSegmentsFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "trim_missing_segments")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -376,9 +345,6 @@ func TestSubprocessRunnerTrimOutputMissingSegmentsFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeSuccessInvocationAndProvenance(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_success")
@@ -453,9 +419,6 @@ func TestSubprocessRunnerNormalizeSuccessInvocationAndProvenance(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "fail")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -475,9 +438,6 @@ func TestSubprocessRunnerNormalizeSubprocessFailure(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_missing_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -494,9 +454,6 @@ func TestSubprocessRunnerNormalizeMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeInvalidOutputJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_invalid_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -513,9 +470,6 @@ func TestSubprocessRunnerNormalizeInvalidOutputJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeOutputMissingSegmentsFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_missing_segments")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -532,9 +486,6 @@ func TestSubprocessRunnerNormalizeOutputMissingSegmentsFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeReportMissingFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_report_missing")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -551,9 +502,6 @@ func TestSubprocessRunnerNormalizeReportMissingFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerNormalizeInvalidReportJSONFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "normalize_invalid_report")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -570,9 +518,6 @@ func TestSubprocessRunnerNormalizeInvalidReportJSONFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderSuccessInvocationAndProvenance(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "render_success")
@@ -639,9 +584,6 @@ func TestSubprocessRunnerRenderSuccessInvocationAndProvenance(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderWithoutTitleOmitsTitleArg(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "render_success")
 	recordPath := filepath.Join(t.TempDir(), "record.json")
@@ -663,9 +605,6 @@ func TestSubprocessRunnerRenderWithoutTitleOmitsTitleArg(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderSubprocessFailure(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "fail")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -682,9 +621,6 @@ func TestSubprocessRunnerRenderSubprocessFailure(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderMissingOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "missing_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
@@ -701,9 +637,6 @@ func TestSubprocessRunnerRenderMissingOutputFails(t *testing.T) {
 }
 
 func TestSubprocessRunnerRenderEmptyOutputFails(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("helper wrapper script uses /bin/sh")
-	}
 	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
 	t.Setenv("SERIATIM_HELPER_MODE", "render_empty_output")
 	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))

internal/adapters/storage/s3_backend.go

@@ -37,8 +37,12 @@ type s3ClientOptions struct {
 	Region         string
 	Endpoint       string
 	ForcePathStyle bool
-	AccessKeyID    string
-	SecretKey      string
+	Credentials    *s3StaticCredentials
+}
+
+type s3StaticCredentials struct {
+	AccessKeyID     string
+	SecretAccessKey string
 }
 
 var newS3Client = func(ctx context.Context, opts s3ClientOptions) (s3API, error) {
@@ -46,11 +50,11 @@ var newS3Client = func(ctx context.Context, opts s3ClientOptions) (s3API, error)
 	if strings.TrimSpace(opts.Region) != "" {
 		loadOpts = append(loadOpts, awsconfig.WithRegion(strings.TrimSpace(opts.Region)))
 	}
-	if strings.TrimSpace(opts.AccessKeyID) != "" && strings.TrimSpace(opts.SecretKey) != "" {
+	if opts.Credentials != nil {
 		loadOpts = append(loadOpts, awsconfig.WithCredentialsProvider(
 			credentials.NewStaticCredentialsProvider(
-				strings.TrimSpace(opts.AccessKeyID),
-				strings.TrimSpace(opts.SecretKey),
+				opts.Credentials.AccessKeyID,
+				opts.Credentials.SecretAccessKey,
 				"",
 			),
 		))
@@ -75,14 +79,24 @@ func NewS3BackendFromConfig(ctx context.Context, cfg config.StorageS3Config) (*S
 	if bucket == "" {
 		return nil, fmt.Errorf("storage.s3.bucket is required")
 	}
+	if (cfg.AccessKeyID == nil) != (cfg.SecretAccessKey == nil) {
+		return nil, fmt.Errorf("storage.s3 access key ID and secret access key references must be configured together")
+	}
+	if cfg.AccessKeyID != nil && (!cfg.AccessKeyID.IsResolved() || !cfg.SecretAccessKey.IsResolved()) {
+		return nil, fmt.Errorf("storage.s3 configured credential references must be resolved")
+	}
 
-	client, err := newS3Client(ctx, s3ClientOptions{
+	clientOptions := s3ClientOptions{
 		Region:         cfg.Region,
 		Endpoint:       cfg.Endpoint,
 		ForcePathStyle: cfg.ForcePathStyle,
-		AccessKeyID:    s3CredentialFromEnv(orDefaultEnvName(cfg.AccessKeyIDEnv, config.DefaultS3AccessKeyIDEnv)),
-		SecretKey:      s3CredentialFromEnv(orDefaultEnvName(cfg.SecretKeyEnv, config.DefaultS3SecretAccessKeyEnv)),
-	})
+	}
+	if cfg.AccessKeyID != nil && cfg.SecretAccessKey != nil {
+		clientOptions.Credentials = &s3StaticCredentials{
+			AccessKeyID: cfg.AccessKeyID.Value(), SecretAccessKey: cfg.SecretAccessKey.Value(),
+		}
+	}
+	client, err := newS3Client(ctx, clientOptions)
 	if err != nil {
 		return nil, fmt.Errorf("build s3 client: %w", err)
 	}
@@ -93,26 +107,6 @@ func NewS3BackendFromConfig(ctx context.Context, cfg config.StorageS3Config) (*S
 	}, nil
 }
 
-func s3CredentialFromEnv(envVarName string) string {
-	name := strings.TrimSpace(envVarName)
-	if name == "" {
-		return ""
-	}
-	value, ok := os.LookupEnv(name)
-	if !ok {
-		return ""
-	}
-	return strings.TrimSpace(value)
-}
-
-func orDefaultEnvName(name, fallback string) string {
-	trimmed := strings.TrimSpace(name)
-	if trimmed == "" {
-		return fallback
-	}
-	return trimmed
-}
-
 // List returns objects under prefix.
 func (b *S3Backend) List(ctx context.Context, prefix string) ([]ObjectInfo, error) {
 	normalizedPrefix := normalizeObjectKey(prefix)

internal/adapters/storage/s3_backend_test.go

@@ -14,6 +14,7 @@ import (
 	"github.com/aws/aws-sdk-go-v2/service/s3/types"
 	"github.com/aws/smithy-go"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
@@ -319,8 +320,10 @@ func TestS3BackendDeleteHonorsCancellation(t *testing.T) {
 func TestNewS3BackendFromConfigUsesClientOptions(t *testing.T) {
 	original := newS3Client
 	t.Cleanup(func() { newS3Client = original })
-	t.Setenv("OBJECT_STORAGE_KEY_ID", "id-123")
-	t.Setenv("OBJECT_STORAGE_KEY", "secret-abc")
+	accessKey := resolvedTestSecret(t, "NARRATIO_S3_TEST_ACCESS", "id-123")
+	secretKey := resolvedTestSecret(t, "NARRATIO_S3_TEST_SECRET", "secret-abc")
+	t.Setenv("NARRATIO_S3_TEST_ACCESS", "changed-id")
+	t.Setenv("NARRATIO_S3_TEST_SECRET", "changed-secret")
 
 	var got s3ClientOptions
 	newS3Client = func(_ context.Context, opts s3ClientOptions) (s3API, error) {
@@ -329,10 +332,12 @@ func TestNewS3BackendFromConfigUsesClientOptions(t *testing.T) {
 	}
 
 	backend, err := NewS3BackendFromConfig(context.Background(), config.StorageS3Config{
-		Bucket:         "my-archive",
-		Region:         "us-east-1",
-		Endpoint:       "http://localhost:9000",
-		ForcePathStyle: true,
+		Bucket:          "my-archive",
+		Region:          "us-east-1",
+		Endpoint:        "http://localhost:9000",
+		ForcePathStyle:  true,
+		AccessKeyID:     accessKey,
+		SecretAccessKey: secretKey,
 	})
 	if err != nil {
 		t.Fatalf("NewS3BackendFromConfig() error = %v", err)
@@ -343,7 +348,7 @@ func TestNewS3BackendFromConfigUsesClientOptions(t *testing.T) {
 	if got.Region != "us-east-1" || got.Endpoint != "http://localhost:9000" || !got.ForcePathStyle {
 		t.Fatalf("client options = %#v, want region/endpoint/path-style values", got)
 	}
-	if got.AccessKeyID != "id-123" || got.SecretKey != "secret-abc" {
+	if got.Credentials == nil || got.Credentials.AccessKeyID != "id-123" || got.Credentials.SecretAccessKey != "secret-abc" {
 		t.Fatalf("client options credentials = %#v, want env-resolved static credentials", got)
 	}
 }
@@ -355,7 +360,21 @@ func TestNewS3BackendFromConfigRequiresBucket(t *testing.T) {
 	}
 }
 
-func TestNewS3BackendFromConfigFallsBackWhenCredentialEnvMissing(t *testing.T) {
+func TestNewS3BackendFromConfigRejectsIncompleteOrUnresolvedCredentials(t *testing.T) {
+	unresolved := configkit.SecretEnv("NARRATIO_UNRESOLVED_S3_KEY")
+	for name, cfg := range map[string]config.StorageS3Config{
+		"partial":    {Bucket: "my-archive", AccessKeyID: &unresolved},
+		"unresolved": {Bucket: "my-archive", AccessKeyID: &unresolved, SecretAccessKey: &unresolved},
+	} {
+		t.Run(name, func(t *testing.T) {
+			if _, err := NewS3BackendFromConfig(context.Background(), cfg); err == nil {
+				t.Fatal("NewS3BackendFromConfig() succeeded, want credential error")
+			}
+		})
+	}
+}
+
+func TestNewS3BackendFromConfigUsesDefaultProviderWithoutCredentialReferences(t *testing.T) {
 	original := newS3Client
 	t.Cleanup(func() { newS3Client = original })
 
@@ -366,19 +385,30 @@ func TestNewS3BackendFromConfigFallsBackWhenCredentialEnvMissing(t *testing.T) {
 	}
 
 	_, err := NewS3BackendFromConfig(context.Background(), config.StorageS3Config{
-		Bucket:         "my-archive",
-		Region:         "us-east-1",
-		AccessKeyIDEnv: "MISSING_ACCESS_KEY_ID",
-		SecretKeyEnv:   "MISSING_SECRET_KEY",
+		Bucket: "my-archive",
+		Region: "us-east-1",
 	})
 	if err != nil {
 		t.Fatalf("NewS3BackendFromConfig() error = %v", err)
 	}
-	if got.AccessKeyID != "" || got.SecretKey != "" {
-		t.Fatalf("client options credentials = %#v, want empty fallback values", got)
+	if got.Credentials != nil {
+		t.Fatalf("client options credentials = %#v, want default provider chain", got.Credentials)
 	}
 }
 
+func resolvedTestSecret(t *testing.T, name, value string) *configkit.Secret {
+	t.Helper()
+	t.Setenv(name, value)
+	type document struct {
+		Secret *configkit.Secret `yaml:"secret"`
+	}
+	result, err := configkit.Load(configkit.Bytes("test.yml", []byte("secret: "+name+"\n")), configkit.Options[document]{})
+	if err != nil {
+		t.Fatalf("resolve test secret: %v", err)
+	}
+	return result.Config.Secret
+}
+
 func strPtr(v string) *string { return &v }
 func int64Ptr(v int64) *int64 { return &v }
 func boolPtr(v bool) *bool    { return &v }

internal/adapters/subprocess/process_tree_supported_test.go

@@ -1,4 +1,4 @@
-//go:build linux || darwin || windows
+//go:build linux || darwin
 
 package subprocess
 

internal/adapters/subprocess/process_tree_unsupported.go

@@ -1,10 +1,9 @@
-//go:build !linux && !darwin && !windows
+//go:build !linux && !darwin
 
 package subprocess
 
 import (
 	"fmt"
-	"runtime"
 )
 
 func newOwnedProcessTree() (ownedProcessTree, error) {

internal/adapters/subprocess/process_tree_windows.go

@@ -1,145 +0,0 @@
-//go:build windows
-
-package subprocess
-
-import (
-	"errors"
-	"fmt"
-	"os/exec"
-	"syscall"
-	"unsafe"
-
-	"golang.org/x/sys/windows"
-)
-
-type windowsProcessTree struct {
-	job windows.Handle
-}
-
-func newOwnedProcessTree() (ownedProcessTree, error) {
-	return &windowsProcessTree{}, nil
-}
-
-func (tree *windowsProcessTree) Start(cmd *exec.Cmd) error {
-	job, err := windows.CreateJobObject(nil, nil)
-	if err != nil {
-		return fmt.Errorf("create job object: %w", err)
-	}
-
-	limits := windows.JOBOBJECT_EXTENDED_LIMIT_INFORMATION{}
-	limits.BasicLimitInformation.LimitFlags = windows.JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE
-	if _, err := windows.SetInformationJobObject(job, windows.JobObjectExtendedLimitInformation, uintptr(unsafe.Pointer(&limits)), uint32(unsafe.Sizeof(limits))); err != nil {
-		_ = windows.CloseHandle(job)
-		return fmt.Errorf("configure job object: %w", err)
-	}
-
-	cmd.SysProcAttr = &syscall.SysProcAttr{CreationFlags: windows.CREATE_SUSPENDED}
-	if err := cmd.Start(); err != nil {
-		_ = windows.CloseHandle(job)
-		return err
-	}
-
-	var assignErr error
-	withHandleErr := cmd.Process.WithHandle(func(handle uintptr) {
-		assignErr = windows.AssignProcessToJobObject(job, windows.Handle(handle))
-	})
-	if withHandleErr != nil {
-		return failWindowsProcessStart(cmd, job, fmt.Errorf("acquire stable process handle for job assignment: %w", withHandleErr))
-	}
-	if assignErr != nil {
-		return failWindowsProcessStart(cmd, job, fmt.Errorf("assign process to job object: %w", assignErr))
-	}
-	if err := resumeInitialThread(uint32(cmd.Process.Pid)); err != nil {
-		return failWindowsProcessStart(cmd, job, fmt.Errorf("resume initial process thread: %w", err))
-	}
-
-	tree.job = job
-	return nil
-}
-
-func failWindowsProcessStart(cmd *exec.Cmd, job windows.Handle, setupErr error) error {
-	killErr := cmd.Process.Kill()
-	waitErr := cmd.Wait()
-	var exitErr *exec.ExitError
-	if killErr == nil && errors.As(waitErr, &exitErr) {
-		// A killed suspended child normally reports a non-zero exit from Wait;
-		// successful reaping is not an additional setup failure.
-		waitErr = nil
-	}
-	return joinErrors(
-		setupErr,
-		wrapWindowsProcessCleanupError("kill suspended child", killErr),
-		wrapWindowsProcessCleanupError("reap suspended child", waitErr),
-		wrapWindowsProcessCleanupError("close unowned job object", windows.CloseHandle(job)),
-	)
-}
-
-func wrapWindowsProcessCleanupError(operation string, err error) error {
-	if err == nil {
-		return nil
-	}
-	return fmt.Errorf("%s: %w", operation, err)
-}
-
-func resumeInitialThread(processID uint32) error {
-	snapshot, err := windows.CreateToolhelp32Snapshot(windows.TH32CS_SNAPTHREAD, 0)
-	if err != nil {
-		return fmt.Errorf("snapshot initial thread: %w", err)
-	}
-	defer func() { _ = windows.CloseHandle(snapshot) }()
-
-	entry := windows.ThreadEntry32{Size: uint32(unsafe.Sizeof(windows.ThreadEntry32{}))}
-	if err := windows.Thread32First(snapshot, &entry); err != nil {
-		return fmt.Errorf("find initial thread: %w", err)
-	}
-	for {
-		if entry.OwnerProcessID != processID {
-			// Keep enumerating until the suspended process's only initial thread
-			// is found.
-		} else {
-			thread, openErr := windows.OpenThread(windows.THREAD_SUSPEND_RESUME, false, entry.ThreadID)
-			if openErr != nil {
-				return fmt.Errorf("open initial thread: %w", openErr)
-			}
-			defer func() { _ = windows.CloseHandle(thread) }()
-			if _, resumeErr := windows.ResumeThread(thread); resumeErr != nil {
-				return fmt.Errorf("resume initial thread: %w", resumeErr)
-			}
-			return nil
-		}
-
-		if err := windows.Thread32Next(snapshot, &entry); err != nil {
-			if errors.Is(err, windows.ERROR_NO_MORE_FILES) {
-				break
-			}
-			return fmt.Errorf("find initial thread: %w", err)
-		}
-	}
-	return fmt.Errorf("find initial thread: no thread found for process %d", processID)
-}
-
-func (tree *windowsProcessTree) TerminateGracefully() error {
-	// Windows jobs have no portable graceful signal. Terminating the owned job
-	// is the safe fallback and prevents a descendant from escaping cleanup.
-	return tree.terminate()
-}
-
-func (tree *windowsProcessTree) TerminateForcefully() error {
-	return tree.terminate()
-}
-
-func (tree *windowsProcessTree) Dispose() error {
-	if tree.job == 0 {
-		return nil
-	}
-	err := windows.CloseHandle(tree.job)
-	tree.job = 0
-	return err
-}
-
-func (tree *windowsProcessTree) terminate() error {
-	if tree.job == 0 {
-		return nil
-	}
-	return windows.TerminateJobObject(tree.job, 1)
-}

internal/adapters/subprocess/run_test.go

@@ -8,7 +8,6 @@ import (
 	"os/exec"
 	"os/signal"
 	"path/filepath"
-	"runtime"
 	"strconv"
 	"strings"
 	"syscall"
@@ -137,9 +136,6 @@ func TestRunFailureRedactsSensitiveTail(t *testing.T) {
 }
 
 func TestRunRejectsSymlinkDiagnosticWithoutTruncatingTarget(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("creating symlinks requires privileges that are not available on every Windows runner")
-	}
 	exe, err := os.Executable()
 	if err != nil {
 		t.Fatalf("os.Executable() error = %v", err)

internal/adapters/whisperx/http.go

@@ -6,14 +6,11 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"mime/multipart"
 	"net"
 	"net/http"
 	"net/url"
 	"os"
-	"path/filepath"
 	"strings"
-	"sync"
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
@@ -21,7 +18,6 @@ import (
 
 const (
 	defaultMaxWhisperXResponseBytes int64 = 10 * 1024 * 1024
-	whisperXUploadBufferSize              = 32 * 1024
 )
 
 // HTTPClientConfig contains parsed, deterministic WhisperX HTTP client settings.
@@ -192,21 +188,32 @@ func (c *HTTPClient) Transcribe(ctx context.Context, req TranscribeRequest) (Tra
 }
 
 func (c *HTTPClient) doTranscribeAttempt(ctx context.Context, audioPath string) (int, []byte, error) {
-	upload := newMultipartUpload(ctx, audioPath, c.language, c.openAudio)
-	defer upload.Close()
+	upload, err := newMultipartBodyFactory(ctx, audioPath, c.language, c.openAudio)
+	if err != nil {
+		return 0, nil, fmt.Errorf("prepare whisperx request body: %w", err)
+	}
+	defer func() { _ = upload.Close() }()
+	body, err := upload.NewBody()
+	if err != nil {
+		return 0, nil, fmt.Errorf("open whisperx request body: %w", err)
+	}
 
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, c.url.String(), upload)
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, c.url.String(), body)
 	if err != nil {
 		return 0, nil, fmt.Errorf("build whisperx request: %w", err)
 	}
 	req.Header.Set("Content-Type", upload.contentType)
 	req.Header.Set("Expect", "100-continue")
+	req.GetBody = upload.NewBody
 
 	resp, err := c.httpClient.Do(req)
 	if err != nil {
 		_ = upload.Close()
-		if producerErr := upload.Wait(); producerErr != nil {
-			return 0, nil, fmt.Errorf("stream whisperx request body: %w", producerErr)
+		if sourceErr := upload.SourceError(true); sourceErr != nil {
+			return 0, nil, fmt.Errorf("stream whisperx request body: %w", sourceErr)
+		}
+		if ctxErr := ctx.Err(); ctxErr != nil {
+			return 0, nil, fmt.Errorf("perform whisperx request: %w", ctxErr)
 		}
 		return 0, nil, fmt.Errorf("perform whisperx request: %w", err)
 	}
@@ -214,8 +221,8 @@ func (c *HTTPClient) doTranscribeAttempt(ctx context.Context, audioPath string)
 
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
 		_ = upload.Close()
-		if producerErr := upload.Wait(); producerErr != nil {
-			return resp.StatusCode, nil, fmt.Errorf("stream whisperx request body: %w", producerErr)
+		if sourceErr := upload.SourceError(false); sourceErr != nil {
+			return resp.StatusCode, nil, fmt.Errorf("stream whisperx request body: %w", sourceErr)
 		}
 		if _, err := readWhisperXResponse(resp.Body, c.maxResponseBytes); err != nil {
 			return resp.StatusCode, nil, fmt.Errorf("read whisperx response body: %w", err)
@@ -223,13 +230,19 @@ func (c *HTTPClient) doTranscribeAttempt(ctx context.Context, audioPath string)
 		return resp.StatusCode, nil, fmt.Errorf("whisperx returned status %d", resp.StatusCode)
 	}
 
-	if err := upload.Wait(); err != nil {
-		return resp.StatusCode, nil, fmt.Errorf("stream whisperx request body: %w", err)
+	data, responseErr := readWhisperXResponse(resp.Body, c.maxResponseBytes)
+	_ = upload.Close()
+	if sourceErr := upload.SourceError(false); sourceErr != nil {
+		if responseErr != nil {
+			return resp.StatusCode, nil, fmt.Errorf("stream whisperx request body: %w", errors.Join(sourceErr, responseErr))
+		}
+		return resp.StatusCode, nil, fmt.Errorf("stream whisperx request body: %w", sourceErr)
 	}
-
-	data, err := readWhisperXResponse(resp.Body, c.maxResponseBytes)
-	if err != nil {
-		return resp.StatusCode, nil, fmt.Errorf("read whisperx response body: %w", err)
+	if err := ctx.Err(); err != nil {
+		return resp.StatusCode, nil, fmt.Errorf("complete whisperx request: %w", err)
+	}
+	if responseErr != nil {
+		return resp.StatusCode, nil, fmt.Errorf("read whisperx response body: %w", responseErr)
 	}
 	return resp.StatusCode, data, nil
 }
@@ -286,161 +299,6 @@ func isHTTPURLScheme(scheme string) bool {
 	}
 }
 
-type multipartUpload struct {
-	reader      *io.PipeReader
-	writer      *io.PipeWriter
-	contentType string
-	done        chan struct{}
-
-	mu      sync.Mutex
-	audio   io.Closer
-	err     error
-	aborted bool
-}
-
-func newMultipartUpload(ctx context.Context, audioPath, language string, openAudio func(string) (io.ReadCloser, error)) *multipartUpload {
-	reader, writer := io.Pipe()
-	multipartWriter := multipart.NewWriter(writer)
-	upload := &multipartUpload{
-		reader:      reader,
-		writer:      writer,
-		contentType: multipartWriter.FormDataContentType(),
-		done:        make(chan struct{}),
-	}
-
-	go func() {
-		err := upload.write(ctx, multipartWriter, audioPath, language, openAudio)
-		if err != nil {
-			_ = writer.CloseWithError(err)
-		} else {
-			_ = writer.Close()
-		}
-		upload.mu.Lock()
-		upload.err = err
-		upload.audio = nil
-		upload.mu.Unlock()
-		close(upload.done)
-	}()
-
-	go func() {
-		select {
-		case <-ctx.Done():
-			upload.abort()
-		case <-upload.done:
-		}
-	}()
-
-	return upload
-}
-
-func (u *multipartUpload) Read(p []byte) (int, error) {
-	return u.reader.Read(p)
-}
-
-func (u *multipartUpload) Close() error {
-	u.abort()
-	return nil
-}
-
-func (u *multipartUpload) Wait() error {
-	<-u.done
-	u.mu.Lock()
-	defer u.mu.Unlock()
-	return u.err
-}
-
-func (u *multipartUpload) write(ctx context.Context, writer *multipart.Writer, audioPath, language string, openAudio func(string) (io.ReadCloser, error)) error {
-	fileWriter, err := writer.CreateFormFile("file", filepath.Base(audioPath))
-	if err != nil {
-		return u.producerError(ctx, fmt.Errorf("create multipart file field: %w", err))
-	}
-
-	audioFile, err := openAudio(audioPath)
-	if err != nil {
-		return u.producerError(ctx, fmt.Errorf("open audio file %q: %w", audioPath, err))
-	}
-	u.setAudio(audioFile)
-
-	_, copyErr := io.CopyBuffer(fileWriter, &contextReader{ctx: ctx, reader: audioFile}, make([]byte, whisperXUploadBufferSize))
-	closeErr := audioFile.Close()
-	u.clearAudio(audioFile)
-	if copyErr != nil {
-		return u.producerError(ctx, fmt.Errorf("copy audio file %q: %w", audioPath, copyErr))
-	}
-	if closeErr != nil {
-		return u.producerError(ctx, fmt.Errorf("close audio file %q: %w", audioPath, closeErr))
-	}
-
-	if err := writer.WriteField("language", language); err != nil {
-		return u.producerError(ctx, fmt.Errorf("write language form field: %w", err))
-	}
-	if err := writer.Close(); err != nil {
-		return u.producerError(ctx, fmt.Errorf("close multipart writer: %w", err))
-	}
-	return nil
-}
-
-func (u *multipartUpload) producerError(ctx context.Context, err error) error {
-	if ctx.Err() != nil {
-		return ctx.Err()
-	}
-	u.mu.Lock()
-	aborted := u.aborted
-	u.mu.Unlock()
-	if aborted {
-		return nil
-	}
-	return err
-}
-
-func (u *multipartUpload) setAudio(audio io.Closer) {
-	u.mu.Lock()
-	u.audio = audio
-	aborted := u.aborted
-	u.mu.Unlock()
-	if aborted {
-		_ = audio.Close()
-	}
-}
-
-func (u *multipartUpload) clearAudio(audio io.Closer) {
-	u.mu.Lock()
-	if u.audio == audio {
-		u.audio = nil
-	}
-	u.mu.Unlock()
-}
-
-func (u *multipartUpload) abort() {
-	u.mu.Lock()
-	if u.aborted {
-		u.mu.Unlock()
-		return
-	}
-	u.aborted = true
-	audio := u.audio
-	u.mu.Unlock()
-
-	_ = u.reader.Close()
-	if audio != nil {
-		_ = audio.Close()
-	}
-}
-
-type contextReader struct {
-	ctx    context.Context
-	reader io.Reader
-}
-
-func (r *contextReader) Read(p []byte) (int, error) {
-	select {
-	case <-r.ctx.Done():
-		return 0, r.ctx.Err()
-	default:
-		return r.reader.Read(p)
-	}
-}
-
 func writeFileAtomic(path string, data []byte, perm os.FileMode) error {
 	if strings.TrimSpace(path) == "" {
 		return fmt.Errorf("path is required")

internal/adapters/whisperx/http_replay_test.go

@@ -0,0 +1,261 @@
+package whisperx
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestHTTPClientTransportReplaysAbandonedMultipartBody(t *testing.T) {
+	var opens atomic.Int32
+	var closes atomic.Int32
+	transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
+		partial := make([]byte, 17)
+		if _, err := io.ReadFull(request.Body, partial); err != nil {
+			return nil, err
+		}
+		if err := request.Body.Close(); err != nil {
+			return nil, err
+		}
+		if request.GetBody == nil {
+			return nil, errors.New("request GetBody is nil")
+		}
+		replay, err := request.GetBody()
+		if err != nil {
+			return nil, err
+		}
+		replayedBytes, err := io.ReadAll(replay)
+		if err != nil {
+			_ = replay.Close()
+			return nil, err
+		}
+		if err := replay.Close(); err != nil {
+			return nil, err
+		}
+		parts := parseMultipartParts(t, request.Header.Get("Content-Type"), replayedBytes)
+		if len(parts) != 2 || parts[0].name != "file" || parts[0].filename != "audio.flac" || string(parts[0].data) != "audio-data" {
+			t.Fatalf("replayed file part = %#v, want exact audio.flac payload", parts)
+		}
+		if parts[1].name != "language" || string(parts[1].data) != "en" {
+			t.Fatalf("replayed language part = %#v, want language=en", parts[1])
+		}
+		return whisperXJSONResponse(request), nil
+	})
+	client := newReplayTestHTTPClient(t, transport)
+	client.openAudio = func(string) (io.ReadCloser, error) {
+		opens.Add(1)
+		return &countedReadCloser{Reader: strings.NewReader("audio-data"), closes: &closes}, nil
+	}
+	outPath := filepath.Join(t.TempDir(), "raw.json")
+
+	result, err := client.Transcribe(context.Background(), TranscribeRequest{
+		AudioPath:               "audio.flac",
+		OutputRawTranscriptPath: outPath,
+	})
+	if err != nil {
+		t.Fatalf("Transcribe() error = %v", err)
+	}
+	if result.Attempts != 1 {
+		t.Fatalf("application attempts = %d, want 1", result.Attempts)
+	}
+	verifyJSONFile(t, outPath)
+	if got := opens.Load(); got != 2 {
+		t.Fatalf("audio opens = %d, want original and replay", got)
+	}
+	if got := closes.Load(); got != 2 {
+		t.Fatalf("audio closes = %d, want all body handles disposed", got)
+	}
+}
+
+func TestHTTPClientTransportReturnsReplayOpenFailure(t *testing.T) {
+	replayErr := errors.New("replay open failed")
+	var opens atomic.Int32
+	var closes atomic.Int32
+	transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
+		if _, err := io.ReadFull(request.Body, make([]byte, 1)); err != nil {
+			return nil, err
+		}
+		_ = request.Body.Close()
+		_, err := request.GetBody()
+		return nil, err
+	})
+	client := newReplayTestHTTPClient(t, transport)
+	client.openAudio = func(string) (io.ReadCloser, error) {
+		if opens.Add(1) == 2 {
+			return nil, replayErr
+		}
+		return &countedReadCloser{Reader: strings.NewReader("audio-data"), closes: &closes}, nil
+	}
+	outPath := filepath.Join(t.TempDir(), "raw.json")
+
+	result, err := client.Transcribe(context.Background(), TranscribeRequest{
+		AudioPath:               "audio.flac",
+		OutputRawTranscriptPath: outPath,
+	})
+	if !errors.Is(err, replayErr) {
+		t.Fatalf("Transcribe() error = %v, want replay open failure", err)
+	}
+	if result.Attempts != 1 {
+		t.Fatalf("application attempts = %d, want 1", result.Attempts)
+	}
+	if got := closes.Load(); got != 1 {
+		t.Fatalf("audio closes = %d, want original disposed", got)
+	}
+	if _, statErr := os.Stat(outPath); !errors.Is(statErr, os.ErrNotExist) {
+		t.Fatalf("output should not exist, stat error = %v", statErr)
+	}
+}
+
+func TestHTTPClientTransportCannotSwallowReplayReadFailure(t *testing.T) {
+	replayErr := errors.New("replay read failed")
+	var opens atomic.Int32
+	transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
+		if _, err := io.ReadFull(request.Body, make([]byte, 1)); err != nil {
+			return nil, err
+		}
+		_ = request.Body.Close()
+		replay, err := request.GetBody()
+		if err != nil {
+			return nil, err
+		}
+		_, _ = io.ReadAll(replay)
+		_ = replay.Close()
+		return whisperXJSONResponse(request), nil
+	})
+	client := newReplayTestHTTPClient(t, transport)
+	client.openAudio = func(string) (io.ReadCloser, error) {
+		if opens.Add(1) == 2 {
+			return &failingReadCloser{first: []byte("partial"), err: replayErr}, nil
+		}
+		return io.NopCloser(strings.NewReader("audio-data")), nil
+	}
+	outPath := filepath.Join(t.TempDir(), "raw.json")
+
+	_, err := client.Transcribe(context.Background(), TranscribeRequest{
+		AudioPath:               "audio.flac",
+		OutputRawTranscriptPath: outPath,
+	})
+	if !errors.Is(err, replayErr) {
+		t.Fatalf("Transcribe() error = %v, want replay read failure", err)
+	}
+	if _, statErr := os.Stat(outPath); !errors.Is(statErr, os.ErrNotExist) {
+		t.Fatalf("output should not exist, stat error = %v", statErr)
+	}
+}
+
+func TestHTTPClientTransportEarlyResponseDisposesOriginalBody(t *testing.T) {
+	var closes atomic.Int32
+	client := newReplayTestHTTPClient(t, roundTripFunc(func(request *http.Request) (*http.Response, error) {
+		return whisperXJSONResponse(request), nil
+	}))
+	client.openAudio = func(string) (io.ReadCloser, error) {
+		return &countedReadCloser{Reader: strings.NewReader("audio-data"), closes: &closes}, nil
+	}
+
+	_, err := client.Transcribe(context.Background(), TranscribeRequest{
+		AudioPath:               "audio.flac",
+		OutputRawTranscriptPath: filepath.Join(t.TempDir(), "raw.json"),
+	})
+	if err != nil {
+		t.Fatalf("Transcribe() error = %v", err)
+	}
+	if got := closes.Load(); got != 1 {
+		t.Fatalf("audio closes = %d, want early-response body disposed", got)
+	}
+}
+
+func TestHTTPClientParentCancellationDuringReplayClosesEveryBody(t *testing.T) {
+	var opens atomic.Int32
+	var originalCloses atomic.Int32
+	replaySource := newSignalingBlockedReadCloser()
+	transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
+		if _, err := io.ReadFull(request.Body, make([]byte, 1)); err != nil {
+			return nil, err
+		}
+		_ = request.Body.Close()
+		replay, err := request.GetBody()
+		if err != nil {
+			return nil, err
+		}
+		_, _ = io.ReadAll(replay)
+		_ = replay.Close()
+		return nil, request.Context().Err()
+	})
+	client := newReplayTestHTTPClient(t, transport)
+	client.openAudio = func(string) (io.ReadCloser, error) {
+		if opens.Add(1) == 1 {
+			return &countedReadCloser{Reader: strings.NewReader("audio-data"), closes: &originalCloses}, nil
+		}
+		return replaySource, nil
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan error, 1)
+	go func() {
+		_, err := client.Transcribe(ctx, TranscribeRequest{
+			AudioPath:               "audio.flac",
+			OutputRawTranscriptPath: filepath.Join(t.TempDir(), "raw.json"),
+		})
+		done <- err
+	}()
+
+	select {
+	case <-replaySource.blocked:
+		cancel()
+	case <-time.After(time.Second):
+		cancel()
+		t.Fatal("replay source did not block")
+	}
+	select {
+	case err := <-done:
+		if !errors.Is(err, context.Canceled) {
+			t.Fatalf("Transcribe() error = %v, want context cancellation", err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("Transcribe() did not return after replay cancellation")
+	}
+	if got := originalCloses.Load(); got != 1 {
+		t.Fatalf("original audio closes = %d, want 1", got)
+	}
+	select {
+	case <-replaySource.closed:
+	case <-time.After(time.Second):
+		t.Fatal("replay audio source was not closed")
+	}
+}
+
+type roundTripFunc func(*http.Request) (*http.Response, error)
+
+func (roundTrip roundTripFunc) RoundTrip(request *http.Request) (*http.Response, error) {
+	return roundTrip(request)
+}
+
+func newReplayTestHTTPClient(t *testing.T, transport http.RoundTripper) *HTTPClient {
+	t.Helper()
+	client, err := NewHTTPClient(HTTPClientConfig{
+		TranscribeURL: "http://whisperx.test/transcribe",
+		Language:      "en",
+		Timeout:       time.Second,
+		HTTPClient:    &http.Client{Transport: transport},
+	})
+	if err != nil {
+		t.Fatalf("NewHTTPClient() error = %v", err)
+	}
+	return client
+}
+
+func whisperXJSONResponse(request *http.Request) *http.Response {
+	return &http.Response{
+		StatusCode: http.StatusOK,
+		Status:     "200 OK",
+		Header:     make(http.Header),
+		Body:       io.NopCloser(strings.NewReader(`{"source":"transport-replay","segments":[]}`)),
+		Request:    request,
+	}
+}

internal/adapters/whisperx/http_test.go

@@ -365,7 +365,7 @@ func TestHTTPClientEarlyServerResponseReturns(t *testing.T) {
 	}
 }
 
-func TestHTTPClientCancellationReleasesBlockedProducer(t *testing.T) {
+func TestHTTPClientCancellationReleasesBlockedSource(t *testing.T) {
 	release := make(chan struct{})
 	source := newGatedReadCloser([]byte("audio-data"), release)
 	firstByteReceived := make(chan struct{})

internal/adapters/whisperx/multipart_body.go

@@ -0,0 +1,263 @@
+package whisperx
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"mime/multipart"
+	"path/filepath"
+	"sync"
+)
+
+type multipartBodyFactory struct {
+	ctx         context.Context
+	audioPath   string
+	openAudio   func(string) (io.ReadCloser, error)
+	prefix      []byte
+	suffix      []byte
+	contentType string
+
+	mu     sync.Mutex
+	closed bool
+	bodies []*multipartBody
+}
+
+func newMultipartBodyFactory(
+	ctx context.Context,
+	audioPath string,
+	language string,
+	openAudio func(string) (io.ReadCloser, error),
+) (*multipartBodyFactory, error) {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	if openAudio == nil {
+		return nil, fmt.Errorf("audio opener is required")
+	}
+	var framing bytes.Buffer
+	writer := multipart.NewWriter(&framing)
+	if _, err := writer.CreateFormFile("file", filepath.Base(audioPath)); err != nil {
+		return nil, fmt.Errorf("create multipart file field: %w", err)
+	}
+	prefixLength := framing.Len()
+	if err := writer.WriteField("language", language); err != nil {
+		return nil, fmt.Errorf("write language form field: %w", err)
+	}
+	if err := writer.Close(); err != nil {
+		return nil, fmt.Errorf("close multipart framing: %w", err)
+	}
+	framingBytes := framing.Bytes()
+	return &multipartBodyFactory{
+		ctx:         ctx,
+		audioPath:   audioPath,
+		openAudio:   openAudio,
+		prefix:      append([]byte(nil), framingBytes[:prefixLength]...),
+		suffix:      append([]byte(nil), framingBytes[prefixLength:]...),
+		contentType: writer.FormDataContentType(),
+	}, nil
+}
+
+func (factory *multipartBodyFactory) NewBody() (io.ReadCloser, error) {
+	factory.mu.Lock()
+	if factory.closed {
+		factory.mu.Unlock()
+		return nil, fmt.Errorf("multipart body factory is closed")
+	}
+	if err := factory.ctx.Err(); err != nil {
+		factory.mu.Unlock()
+		return nil, err
+	}
+	audio, err := factory.openAudio(factory.audioPath)
+	if err != nil {
+		factory.mu.Unlock()
+		return nil, fmt.Errorf("open audio file %q: %w", factory.audioPath, err)
+	}
+	if err := factory.ctx.Err(); err != nil {
+		factory.mu.Unlock()
+		_ = audio.Close()
+		return nil, err
+	}
+	source := &multipartAudioSource{audio: audio, closeDone: make(chan struct{})}
+	body := &multipartBody{
+		ctx:    factory.ctx,
+		source: source,
+		reader: io.MultiReader(bytes.NewReader(factory.prefix), source, bytes.NewReader(factory.suffix)),
+	}
+	factory.bodies = append(factory.bodies, body)
+	factory.mu.Unlock()
+	stopCancel := context.AfterFunc(factory.ctx, func() {
+		_ = body.Close()
+	})
+	body.mu.Lock()
+	body.stopCancel = stopCancel
+	closed := body.closed
+	body.mu.Unlock()
+	if closed {
+		stopCancel()
+	}
+	if err := factory.ctx.Err(); err != nil {
+		_ = body.Close()
+		return nil, err
+	}
+	return body, nil
+}
+
+func (factory *multipartBodyFactory) Close() error {
+	factory.mu.Lock()
+	if factory.closed {
+		bodies := append([]*multipartBody(nil), factory.bodies...)
+		factory.mu.Unlock()
+		return multipartBodyCloseErrors(bodies)
+	}
+	factory.closed = true
+	bodies := append([]*multipartBody(nil), factory.bodies...)
+	factory.mu.Unlock()
+	for _, body := range bodies {
+		_ = body.Close()
+	}
+	return multipartBodyCloseErrors(bodies)
+}
+
+func (factory *multipartBodyFactory) SourceError(includeAbandoned bool) error {
+	factory.mu.Lock()
+	bodies := append([]*multipartBody(nil), factory.bodies...)
+	factory.mu.Unlock()
+	var result error
+	for _, body := range bodies {
+		if err := body.sourceError(includeAbandoned); err != nil {
+			result = errors.Join(result, err)
+		}
+	}
+	return result
+}
+
+func multipartBodyCloseErrors(bodies []*multipartBody) error {
+	var result error
+	for _, body := range bodies {
+		if err := body.sourceError(false); err != nil {
+			result = errors.Join(result, err)
+		}
+	}
+	return result
+}
+
+type multipartBody struct {
+	ctx        context.Context
+	reader     io.Reader
+	source     *multipartAudioSource
+	stopCancel func() bool
+
+	mu        sync.Mutex
+	closed    bool
+	abandoned bool
+}
+
+func (body *multipartBody) Read(buffer []byte) (int, error) {
+	if err := body.ctx.Err(); err != nil {
+		return 0, err
+	}
+	body.mu.Lock()
+	closed := body.closed
+	body.mu.Unlock()
+	if closed {
+		return 0, io.ErrClosedPipe
+	}
+	return body.reader.Read(buffer)
+}
+
+func (body *multipartBody) Close() error {
+	body.mu.Lock()
+	if body.closed {
+		body.mu.Unlock()
+		return body.source.Close()
+	}
+	body.closed = true
+	body.abandoned = !body.source.reachedEOF()
+	stopCancel := body.stopCancel
+	body.mu.Unlock()
+	if stopCancel != nil {
+		stopCancel()
+	}
+	return body.source.Close()
+}
+
+func (body *multipartBody) sourceError(includeAbandoned bool) error {
+	body.mu.Lock()
+	abandoned := body.abandoned
+	body.mu.Unlock()
+	err := body.source.Err()
+	if err != nil {
+		return err
+	}
+	if abandoned && !includeAbandoned {
+		return nil
+	}
+	return nil
+}
+
+type multipartAudioSource struct {
+	audio io.ReadCloser
+
+	mu          sync.Mutex
+	closing     bool
+	eof         bool
+	readErr     error
+	closeErr    error
+	closeDone   chan struct{}
+	closeResult error
+	closed      bool
+}
+
+func (source *multipartAudioSource) Read(buffer []byte) (int, error) {
+	n, err := source.audio.Read(buffer)
+	source.mu.Lock()
+	switch {
+	case errors.Is(err, io.EOF):
+		source.eof = true
+	case err != nil && !source.closing:
+		source.readErr = errors.Join(source.readErr, err)
+	}
+	source.mu.Unlock()
+	return n, err
+}
+
+func (source *multipartAudioSource) Close() error {
+	source.mu.Lock()
+	if source.closed {
+		done := source.closeDone
+		source.mu.Unlock()
+		<-done
+		source.mu.Lock()
+		err := source.closeResult
+		source.mu.Unlock()
+		return err
+	}
+	source.closed = true
+	source.closing = true
+	reachedEOF := source.eof
+	source.mu.Unlock()
+
+	err := source.audio.Close()
+	source.mu.Lock()
+	source.closeResult = err
+	if err != nil && reachedEOF {
+		source.closeErr = err
+	}
+	close(source.closeDone)
+	source.mu.Unlock()
+	return err
+}
+
+func (source *multipartAudioSource) reachedEOF() bool {
+	source.mu.Lock()
+	defer source.mu.Unlock()
+	return source.eof
+}
+
+func (source *multipartAudioSource) Err() error {
+	source.mu.Lock()
+	defer source.mu.Unlock()
+	return errors.Join(source.readErr, source.closeErr)
+}

internal/adapters/whisperx/multipart_body_test.go

@@ -0,0 +1,327 @@
+package whisperx
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"mime"
+	"mime/multipart"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestMultipartBodyReplayUsesIdenticalFraming(t *testing.T) {
+	factory := newTestMultipartBodyFactory(t, context.Background(), "audio-data")
+	defer factory.Close()
+
+	original, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() original error = %v", err)
+	}
+	replay, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() replay error = %v", err)
+	}
+	originalBytes, err := io.ReadAll(original)
+	if err != nil {
+		t.Fatalf("ReadAll(original) error = %v", err)
+	}
+	replayBytes, err := io.ReadAll(replay)
+	if err != nil {
+		t.Fatalf("ReadAll(replay) error = %v", err)
+	}
+	if !bytes.Equal(originalBytes, replayBytes) {
+		t.Fatal("original and replay multipart bytes differ")
+	}
+
+	parts := parseMultipartParts(t, factory.contentType, replayBytes)
+	if len(parts) != 2 {
+		t.Fatalf("part count = %d, want 2", len(parts))
+	}
+	if parts[0].name != "file" || parts[0].filename != "audio.flac" || string(parts[0].data) != "audio-data" {
+		t.Fatalf("file part = %#v, want file/audio.flac with exact audio", parts[0])
+	}
+	if parts[1].name != "language" || string(parts[1].data) != "en" {
+		t.Fatalf("language part = %#v, want language=en", parts[1])
+	}
+}
+
+func TestMultipartBodyReplayRestartsAfterPartialOriginal(t *testing.T) {
+	factory := newTestMultipartBodyFactory(t, context.Background(), "audio-data")
+	defer factory.Close()
+
+	original, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() original error = %v", err)
+	}
+	partial := make([]byte, len(factory.prefix)+3)
+	if _, err := io.ReadFull(original, partial); err != nil {
+		t.Fatalf("ReadFull(original) error = %v", err)
+	}
+	if err := original.Close(); err != nil {
+		t.Fatalf("Close(original) error = %v", err)
+	}
+
+	replay, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() replay error = %v", err)
+	}
+	replayBytes, err := io.ReadAll(replay)
+	if err != nil {
+		t.Fatalf("ReadAll(replay) error = %v", err)
+	}
+	parts := parseMultipartParts(t, factory.contentType, replayBytes)
+	if got := string(parts[0].data); got != "audio-data" {
+		t.Fatalf("replayed audio = %q, want complete payload", got)
+	}
+	if err := factory.SourceError(false); err != nil {
+		t.Fatalf("SourceError(false) = %v, want abandoned original ignored", err)
+	}
+}
+
+func TestMultipartBodySimultaneousBodiesUseIndependentSources(t *testing.T) {
+	var opens atomic.Int32
+	var closes atomic.Int32
+	factory, err := newMultipartBodyFactory(context.Background(), "audio.flac", "en", func(string) (io.ReadCloser, error) {
+		opens.Add(1)
+		return &countedReadCloser{Reader: strings.NewReader("audio-data"), closes: &closes}, nil
+	})
+	if err != nil {
+		t.Fatalf("newMultipartBodyFactory() error = %v", err)
+	}
+
+	first, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() first error = %v", err)
+	}
+	second, err := factory.NewBody()
+	if err != nil {
+		t.Fatalf("NewBody() second error = %v", err)
+	}
+	var firstBytes, secondBytes []byte
+	var firstErr, secondErr error
+	var group sync.WaitGroup
+	group.Add(2)
+	go func() {
+		defer group.Done()
+		firstBytes, firstErr = io.ReadAll(first)
+	}()
+	go func() {
+		defer group.Done()
+		secondBytes, secondErr = io.ReadAll(second)
+	}()
+	group.Wait()
+	if firstErr != nil || secondErr != nil {
+		t.Fatalf("concurrent reads returned (%v, %v)", firstErr, secondErr)
+	}
+	if !bytes.Equal(firstBytes, secondBytes) {
+		t.Fatal("independent body bytes differ")
+	}
+	if got := opens.Load(); got != 2 {
+		t.Fatalf("audio opens = %d, want 2", got)
+	}
+	if err := factory.Close(); err != nil {
+		t.Fatalf("Close() error = %v", err)
+	}
+	if err := factory.Close(); err != nil {
+		t.Fatalf("second Close() error = %v", err)
+	}
+	if got := closes.Load(); got != 2 {
+		t.Fatalf("audio closes = %d, want 2", got)
+	}
+}
+
+func TestMultipartBodyReportsSourceFailures(t *testing.T) {
+	t.Run("read", func(t *testing.T) {
+		sourceErr := errors.New("read source")
+		factory, err := newMultipartBodyFactory(context.Background(), "audio.flac", "en", func(string) (io.ReadCloser, error) {
+			return &failingReadCloser{first: []byte("partial"), err: sourceErr}, nil
+		})
+		if err != nil {
+			t.Fatalf("newMultipartBodyFactory() error = %v", err)
+		}
+		body, err := factory.NewBody()
+		if err != nil {
+			t.Fatalf("NewBody() error = %v", err)
+		}
+		if _, err := io.ReadAll(body); !errors.Is(err, sourceErr) {
+			t.Fatalf("ReadAll() error = %v, want source failure", err)
+		}
+		_ = body.Close()
+		if err := factory.SourceError(false); !errors.Is(err, sourceErr) {
+			t.Fatalf("SourceError(false) = %v, want source failure", err)
+		}
+	})
+
+	t.Run("close after complete read", func(t *testing.T) {
+		closeErr := errors.New("close source")
+		factory, err := newMultipartBodyFactory(context.Background(), "audio.flac", "en", func(string) (io.ReadCloser, error) {
+			return &errorCloseReadCloser{Reader: strings.NewReader("audio-data"), err: closeErr}, nil
+		})
+		if err != nil {
+			t.Fatalf("newMultipartBodyFactory() error = %v", err)
+		}
+		body, err := factory.NewBody()
+		if err != nil {
+			t.Fatalf("NewBody() error = %v", err)
+		}
+		if _, err := io.ReadAll(body); err != nil {
+			t.Fatalf("ReadAll() error = %v", err)
+		}
+		if err := body.Close(); !errors.Is(err, closeErr) {
+			t.Fatalf("Close() error = %v, want source close failure", err)
+		}
+		if err := factory.SourceError(false); !errors.Is(err, closeErr) {
+			t.Fatalf("SourceError(false) = %v, want source close failure", err)
+		}
+	})
+}
+
+func TestMultipartBodyCloseAndCancelReleaseBlockedSource(t *testing.T) {
+	for _, test := range []struct {
+		name    string
+		release func(context.CancelFunc, io.ReadCloser)
+	}{
+		{name: "close", release: func(_ context.CancelFunc, body io.ReadCloser) { _ = body.Close() }},
+		{name: "cancel", release: func(cancel context.CancelFunc, _ io.ReadCloser) { cancel() }},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+			source := newSignalingBlockedReadCloser()
+			factory, err := newMultipartBodyFactory(ctx, "audio.flac", "en", func(string) (io.ReadCloser, error) {
+				return source, nil
+			})
+			if err != nil {
+				t.Fatalf("newMultipartBodyFactory() error = %v", err)
+			}
+			defer factory.Close()
+			body, err := factory.NewBody()
+			if err != nil {
+				t.Fatalf("NewBody() error = %v", err)
+			}
+			readDone := make(chan error, 1)
+			go func() {
+				_, err := io.ReadAll(body)
+				readDone <- err
+			}()
+			select {
+			case <-source.blocked:
+			case <-time.After(time.Second):
+				t.Fatal("audio source did not block")
+			}
+			test.release(cancel, body)
+			select {
+			case <-source.closed:
+			case <-time.After(time.Second):
+				t.Fatal("audio source was not closed")
+			}
+			select {
+			case <-readDone:
+			case <-time.After(time.Second):
+				t.Fatal("blocked body read did not return")
+			}
+			if test.name == "cancel" {
+				if _, err := factory.NewBody(); !errors.Is(err, context.Canceled) {
+					t.Fatalf("NewBody() after cancellation error = %v, want context.Canceled", err)
+				}
+			}
+		})
+	}
+}
+
+func newTestMultipartBodyFactory(t *testing.T, ctx context.Context, audio string) *multipartBodyFactory {
+	t.Helper()
+	factory, err := newMultipartBodyFactory(ctx, "audio.flac", "en", func(string) (io.ReadCloser, error) {
+		return io.NopCloser(strings.NewReader(audio)), nil
+	})
+	if err != nil {
+		t.Fatalf("newMultipartBodyFactory() error = %v", err)
+	}
+	return factory
+}
+
+type multipartTestPart struct {
+	name     string
+	filename string
+	data     []byte
+}
+
+func parseMultipartParts(t *testing.T, contentType string, body []byte) []multipartTestPart {
+	t.Helper()
+	mediaType, params, err := mime.ParseMediaType(contentType)
+	if err != nil {
+		t.Fatalf("ParseMediaType() error = %v", err)
+	}
+	if mediaType != "multipart/form-data" {
+		t.Fatalf("media type = %q, want multipart/form-data", mediaType)
+	}
+	reader := multipart.NewReader(bytes.NewReader(body), params["boundary"])
+	var parts []multipartTestPart
+	for {
+		part, err := reader.NextPart()
+		if errors.Is(err, io.EOF) {
+			return parts
+		}
+		if err != nil {
+			t.Fatalf("NextPart() error = %v", err)
+		}
+		data, err := io.ReadAll(part)
+		if err != nil {
+			t.Fatalf("ReadAll(part %q) error = %v", part.FormName(), err)
+		}
+		parts = append(parts, multipartTestPart{name: part.FormName(), filename: part.FileName(), data: data})
+	}
+}
+
+type countedReadCloser struct {
+	io.Reader
+	closes *atomic.Int32
+	once   sync.Once
+}
+
+func (reader *countedReadCloser) Close() error {
+	reader.once.Do(func() { reader.closes.Add(1) })
+	return nil
+}
+
+type errorCloseReadCloser struct {
+	io.Reader
+	err error
+}
+
+func (reader *errorCloseReadCloser) Close() error { return reader.err }
+
+type signalingBlockedReadCloser struct {
+	blocked chan struct{}
+	closed  chan struct{}
+	once    sync.Once
+	read    bool
+}
+
+func newSignalingBlockedReadCloser() *signalingBlockedReadCloser {
+	return &signalingBlockedReadCloser{blocked: make(chan struct{}), closed: make(chan struct{})}
+}
+
+func (reader *signalingBlockedReadCloser) Read(buffer []byte) (int, error) {
+	if !reader.read {
+		reader.read = true
+		return copy(buffer, "audio-data"), nil
+	}
+	reader.once.Do(func() { close(reader.blocked) })
+	<-reader.closed
+	return 0, errors.New("audio source closed")
+}
+
+func (reader *signalingBlockedReadCloser) Close() error {
+	select {
+	case <-reader.closed:
+	default:
+		close(reader.closed)
+	}
+	return nil
+}

internal/app/analyze_artifacts_commands_test.go

@@ -272,13 +272,18 @@ func TestExecuteAnalyzeRejectsPositionalArgsAndForceFlag(t *testing.T) {
 }
 
 func TestExecuteAnalyzeMissingConfigUsesRunStageLoadingPath(t *testing.T) {
+	deps, missingPath, executor := guardedMissingPipelineDiscovery(t)
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
-	code := Execute([]string{"analyze", "2026-05-03"}, &stdout, &stderr)
+	code := executeWithDependencies([]string{"analyze", "2026-05-03"}, &stdout, &stderr, deps, executor)
 	if code == 0 {
 		t.Fatal("exit code = 0, want non-zero")
 	}
-	if !strings.Contains(stderr.String(), "analyze: no pipeline config path provided and no default pipeline config found; searched:") {
+	if stdout.Len() != 0 {
+		t.Fatalf("stdout = %q, want empty", stdout.String())
+	}
+	if got := stderr.String(); !strings.Contains(got, "analyze: no pipeline config path provided") ||
+		!strings.Contains(got, "no default pipeline config found") || !strings.Contains(got, missingPath) {
 		t.Fatalf("stderr = %q, want pipeline discovery error", stderr.String())
 	}
 }
@@ -371,13 +376,18 @@ func TestExecutePublishUnknownArtifactFailsValidation(t *testing.T) {
 }
 
 func TestExecutePublishMissingConfigUsesRunStageLoadingPath(t *testing.T) {
+	deps, missingPath, executor := guardedMissingPipelineDiscovery(t)
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
-	code := Execute([]string{"publish", "2026-05-03"}, &stdout, &stderr)
+	code := executeWithDependencies([]string{"publish", "2026-05-03"}, &stdout, &stderr, deps, executor)
 	if code == 0 {
 		t.Fatal("exit code = 0, want non-zero")
 	}
-	if !strings.Contains(stderr.String(), "publish: no pipeline config path provided and no default pipeline config found; searched:") {
+	if stdout.Len() != 0 {
+		t.Fatalf("stdout = %q, want empty", stdout.String())
+	}
+	if got := stderr.String(); !strings.Contains(got, "publish: no pipeline config path provided") ||
+		!strings.Contains(got, "no default pipeline config found") || !strings.Contains(got, missingPath) {
 		t.Fatalf("stderr = %q, want pipeline discovery error", stderr.String())
 	}
 }

internal/app/assembled_workflow_test.go

@@ -254,10 +254,10 @@ func TestAssembledSplitBundleCommandsUseOneEffectiveFamilyConfiguration(t *testi
 
 	var validated bytes.Buffer
 	if err := ConfigValidate(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath}, &validated); err != nil {
-		t.Fatalf("validate production default: %v", err)
+		t.Fatalf("validate production base: %v", err)
 	}
-	if !strings.Contains(validated.String(), "profile=production") {
-		t.Fatalf("validate output = %q, want production profile", validated.String())
+	if !strings.Contains(validated.String(), "profile=none") {
+		t.Fatalf("validate output = %q, want unprofiled production base", validated.String())
 	}
 	for _, command := range []func(context.Context, []string, io.Writer) error{ConfigShow, ConfigSources} {
 		if err := command(context.Background(), []string{
@@ -268,7 +268,7 @@ func TestAssembledSplitBundleCommandsUseOneEffectiveFamilyConfiguration(t *testi
 	}
 	var diff bytes.Buffer
 	if err := ConfigDiff(context.Background(), []string{
-		"production", "testing", "--config", pipelinePath, "--campaign-file", campaignPath,
+		"default", "testing", "--config", pipelinePath, "--campaign-file", campaignPath,
 	}, &diff); err != nil {
 		t.Fatalf("compare profiles: %v", err)
 	}
@@ -365,7 +365,7 @@ func TestAssembledSplitBundleProfileSwitchRecordsProvenanceAndLimitsReuse(t *tes
 	if !reflect.DeepEqual(productionRuns, make([]int, len(names))) {
 		t.Fatalf("production runs = %v, want complete reuse", productionRuns)
 	}
-	assertAssembledProvenance(t, store, productionSummary, "production", config.EffectivePipelineDigest(production.Pipeline))
+	assertAssembledProvenance(t, store, productionSummary, "", config.EffectivePipelineDigest(production.Pipeline))
 
 	testingRuns := make([]int, len(names))
 	testingStages := assembledSemanticStages(names, providers, testingRuns)
@@ -415,7 +415,9 @@ func assertAssembledProvenance(t *testing.T, store *manifest.LocalStore, summary
 		t.Fatal(err)
 	}
 	value := session.EffectiveConfig
-	if value == nil || value.SelectedProfile == nil || value.SelectedProfile.Name != profile || value.EffectiveConfigDigest != digest {
+	profileMatches := value != nil && ((profile == "" && value.SelectedProfile == nil) ||
+		(profile != "" && value.SelectedProfile != nil && value.SelectedProfile.Name == profile))
+	if !profileMatches || value.EffectiveConfigDigest != digest {
 		t.Fatalf("effective configuration provenance = %#v, want profile=%q digest=%q", value, profile, digest)
 	}
 }

internal/app/clean.go

@@ -45,7 +45,7 @@ func cleanSessionWithDependencies(ctx context.Context, flags commonConfigFlags,
 	if strings.TrimSpace(flags.sessionID) == "" {
 		return fmt.Errorf("clean: session_id is required unless --all is set")
 	}
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("clean: %w", err)
 	}
@@ -93,7 +93,7 @@ func cleanAllLocalWithDependencies(flags commonConfigFlags, dryRun, clearCache b
 		strings.TrimSpace(flags.previousSessionID) != "" {
 		return fmt.Errorf("clean: --all cannot be combined with --campaign, --campaign-file, --session, a session_id, or --previous-session-id")
 	}
-	_, pipelineCfg, err := loadPipelineConfigWithDependencies(flags.pipelinePath, config.PipelineLoadOptions{Profile: flags.profile.pointer()}, deps)
+	_, pipelineCfg, err := loadPipelineConfigWithDependencies(flags.pipelinePath, config.PipelineLoadOptions{Profile: flags.profile.pointer()}, flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("clean: %w", err)
 	}

internal/app/command_dependencies.go

@@ -2,16 +2,20 @@ package app
 
 import (
 	"context"
-	"os"
+	"io"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
 type commandDependencies struct {
+	diagnostics                io.Writer
+	resolvePipelineConfigPath  func(string) (string, error)
 	loadPipelineConfig         func(string, config.PipelineLoadOptions) (*config.PipelineConfig, error)
 	downloadObjectToTemp       func(context.Context, storage.ObjectStore, string, string) (string, error)
-	readSecretsDirectory       func(string) ([]os.DirEntry, error)
+	loadSecretsDir             func(string) (*configkit.SecretsReport, error)
+	secretsBootstrap           *secretsBootstrap
 	newObjectStore             func(context.Context, *config.Config) (storage.ObjectStore, error)
 	discoverRemoteCurrentState func(context.Context, *config.Config, storage.ObjectStore) (*RemoteCurrentState, error)
 	buildRestorePlan           func(context.Context, *config.Config, *RemoteCurrentState, storage.ObjectStore, RestorePlanOptions) (*RestorePlan, error)
@@ -19,14 +23,23 @@ type commandDependencies struct {
 }
 
 func normalizeCommandDependencies(deps commandDependencies) commandDependencies {
+	if deps.diagnostics == nil {
+		deps.diagnostics = io.Discard
+	}
+	if deps.resolvePipelineConfigPath == nil {
+		deps.resolvePipelineConfigPath = resolvePipelineConfigPath
+	}
 	if deps.loadPipelineConfig == nil {
 		deps.loadPipelineConfig = config.LoadPipelineWithOptions
 	}
 	if deps.downloadObjectToTemp == nil {
 		deps.downloadObjectToTemp = storage.DownloadObjectToTemp
 	}
-	if deps.readSecretsDirectory == nil {
-		deps.readSecretsDirectory = os.ReadDir
+	if deps.loadSecretsDir == nil {
+		deps.loadSecretsDir = configkit.LoadSecretsDir
+	}
+	if deps.secretsBootstrap == nil {
+		deps.secretsBootstrap = &secretsBootstrap{}
 	}
 	if deps.newObjectStore == nil {
 		deps.newObjectStore = storage.NewObjectStoreFromConfig

internal/app/commands.go

@@ -19,6 +19,9 @@ func executeWithPipelineExecutor(args []string, stdout, stderr io.Writer, execut
 }
 
 func executeWithDependencies(args []string, stdout, stderr io.Writer, deps commandDependencies, executor pipelineExecutor) int {
+	if deps.diagnostics == nil {
+		deps.diagnostics = stderr
+	}
 	deps = normalizeCommandDependencies(deps)
 	if len(args) == 0 {
 		printUsage(stderr)

internal/app/commands_test.go

@@ -59,9 +59,10 @@ func TestExecuteValidCommands(t *testing.T) {
 
 func TestExecuteMissingRequiredFlags(t *testing.T) {
 	cases := []struct {
-		name string
-		args []string
-		want string
+		name                string
+		args                []string
+		want                string
+		controlledDiscovery bool
 	}{
 		{name: "run missing session", args: []string{"run"}, want: "run: session_id is required"},
 		{name: "plan old top-level removed", args: []string{"plan"}, want: `unknown command: "plan"`},
@@ -69,7 +70,7 @@ func TestExecuteMissingRequiredFlags(t *testing.T) {
 		{name: "resume removed", args: []string{"resume"}, want: `unknown command: "resume"`},
 		{name: "run-stage missing name", args: []string{"run-stage", "--config", "a", "--session", "b"}, want: "run-stage: expected stage name and session_id"},
 		{name: "run-stage missing session", args: []string{"run-stage", "polish"}, want: "run-stage: expected stage name and session_id"},
-		{name: "run missing config uses defaults", args: []string{"run", "2026-05-03", "--session", "session.yml"}, want: "run: no pipeline config path provided and no default pipeline config found; searched:"},
+		{name: "run missing config uses defaults", args: []string{"run", "2026-05-03", "--session", "session.yml"}, want: "run: no pipeline config path provided", controlledDiscovery: true},
 	}
 
 	for _, tc := range cases {
@@ -77,7 +78,13 @@ func TestExecuteMissingRequiredFlags(t *testing.T) {
 			var stdout bytes.Buffer
 			var stderr bytes.Buffer
 
-			code := Execute(tc.args, &stdout, &stderr)
+			deps := commandDependencies{}
+			executor := pipelineExecutor(executePlan)
+			missingPath := ""
+			if tc.controlledDiscovery {
+				deps, missingPath, executor = guardedMissingPipelineDiscovery(t)
+			}
+			code := executeWithDependencies(tc.args, &stdout, &stderr, deps, executor)
 			if code == 0 {
 				t.Fatalf("exit code = 0, want non-zero")
 			}
@@ -87,6 +94,11 @@ func TestExecuteMissingRequiredFlags(t *testing.T) {
 			if !strings.Contains(stderr.String(), tc.want) {
 				t.Fatalf("stderr = %q, want to contain %q", stderr.String(), tc.want)
 			}
+			if tc.controlledDiscovery {
+				if got := stderr.String(); !strings.Contains(got, "no default pipeline config found") || !strings.Contains(got, missingPath) {
+					t.Fatalf("stderr = %q, want controlled searched path %q", got, missingPath)
+				}
+			}
 		})
 	}
 }
@@ -192,7 +204,7 @@ func TestExecuteRunStagePolishLoadsCredentialFromSecretsDir(t *testing.T) {
 	configDir := t.TempDir()
 	sessionID := "2026-05-03"
 	secretsDir := filepath.Join(configDir, "secrets")
-	if err := os.MkdirAll(secretsDir, secretDirectoryPrivateMode); err != nil {
+	if err := os.MkdirAll(secretsDir, 0o755); err != nil {
 		t.Fatalf("MkdirAll(%q): %v", secretsDir, err)
 	}
 	if err := os.WriteFile(filepath.Join(secretsDir, "OPENROUTER_API_KEY"), []byte("from-secret-file\n"), 0o600); err != nil {
@@ -211,15 +223,13 @@ func TestExecuteRunStagePolishLoadsCredentialFromSecretsDir(t *testing.T) {
   root: ` + workspaceRoot + `
 storage:
   backend: local
-secrets:
-  env_dir: ./secrets
 whisperx:
   transcribe_url: https://example.com/transcribe
 seriatim:
   binary: ` + seriatimBinary + `
 audita:
   binary: ` + auditaBinary + `
-  llm_api_key_env: OPENROUTER_API_KEY
+  llm_api_key: OPENROUTER_API_KEY
 notification:
   mode: noop
 `
@@ -264,9 +274,31 @@ inputs:
 	if !strings.Contains(stdout.String(), "stage=polish executed=1 skipped=0") {
 		t.Fatalf("stdout = %q, want polish execution", stdout.String())
 	}
+	foundReference := false
+	if err := filepath.WalkDir(workRoot, func(path string, entry os.DirEntry, walkErr error) error {
+		if walkErr != nil || entry.IsDir() {
+			return walkErr
+		}
+		data, err := os.ReadFile(path)
+		if err != nil {
+			return err
+		}
+		if bytes.Contains(data, []byte("from-secret-file")) {
+			t.Fatalf("workspace output %q exposed credential value", path)
+		}
+		if bytes.Contains(data, []byte("OPENROUTER_API_KEY")) {
+			foundReference = true
+		}
+		return nil
+	}); err != nil {
+		t.Fatalf("inspect polish outputs: %v", err)
+	}
+	if !foundReference {
+		t.Fatal("polish outputs omitted safe credential reference")
+	}
 }
 
-func TestExecuteRunFailsWhenConfiguredSecretsDirMissing(t *testing.T) {
+func TestExecuteRunFailsWhenExplicitSecretsDirMissing(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	configDir := t.TempDir()
 	pipelinePath := filepath.Join(configDir, "pipeline.yml")
@@ -277,8 +309,6 @@ func TestExecuteRunFailsWhenConfiguredSecretsDirMissing(t *testing.T) {
   root: ` + workspaceRoot + `
 storage:
   backend: local
-secrets:
-  env_dir: ./missing-secrets
 whisperx:
   transcribe_url: https://example.com/transcribe
 seriatim:
@@ -305,11 +335,11 @@ inputs:
 
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
-	code := Execute([]string{"run", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, &stdout, &stderr)
+	code := Execute([]string{"run", "2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath, "--secrets-dir", filepath.Join(configDir, "missing-secrets")}, &stdout, &stderr)
 	if code == 0 {
 		t.Fatal("exit code = 0, want non-zero")
 	}
-	if !strings.Contains(stderr.String(), "validate secrets env_dir") {
+	if !strings.Contains(stderr.String(), "bootstrap secrets directory") {
 		t.Fatalf("stderr = %q, want secrets validation error context", stderr.String())
 	}
 }

internal/app/config_artifacts_test.go

@@ -0,0 +1,211 @@
+package app
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+)
+
+const artifactCatalogHeader = "key\tsource\tdefinitions\tenabled\toutput\tdependencies\tfamily\tcharacter\tfamily_source\tparty_source\n"
+
+func TestConfigArtifactsReportsOrdinaryCatalogWithoutCampaign(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	mustWriteTestFile(t, pipelinePath, `workspace:
+  root: `+filepath.Join(dir, "workspace")+`
+whisperx:
+  transcribe_url: https://transcription.example.com/transcribe
+scriptorium:
+  binary: scriptorium
+  artifacts:
+    zeta:
+      enabled: false
+      output_path: artifacts/zeta.md
+    alpha:
+      enabled: false
+      output_path: artifacts/alpha.md
+      inputs:
+        upstream: {source: narratio.artifact.zeta, required: true}
+`)
+
+	var out bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath}, &out); err != nil {
+		t.Fatalf("ConfigArtifacts() error = %v", err)
+	}
+	lines := strings.Split(strings.TrimSuffix(out.String(), "\n"), "\n")
+	if len(lines) != 3 || !strings.HasPrefix(lines[1], "alpha\t") || !strings.HasPrefix(lines[2], "zeta\t") {
+		t.Fatalf("catalog order/output = %q", out.String())
+	}
+	alpha := strings.Split(lines[1], "\t")
+	if len(alpha) != 10 {
+		t.Fatalf("alpha fields = %#v", alpha)
+	}
+	if alpha[1] != "narratio.artifact.alpha" || alpha[2] != pipelinePath || alpha[3] != "false" || alpha[4] != "artifacts/alpha.md" || alpha[5] != "zeta" {
+		t.Fatalf("alpha record = %#v", alpha)
+	}
+}
+
+func TestConfigArtifactsEmptyCatalogWritesOnlyHeader(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	mustWriteTestFile(t, pipelinePath, "workspace:\n  root: "+filepath.Join(dir, "workspace")+"\nwhisperx:\n  transcribe_url: https://transcription.example.com/transcribe\n")
+	var out bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath}, &out); err != nil {
+		t.Fatal(err)
+	}
+	if out.String() != artifactCatalogHeader {
+		t.Fatalf("empty catalog = %q, want header only", out.String())
+	}
+}
+
+func TestConfigArtifactsReportsGeneratedFamilyProvenance(t *testing.T) {
+	pipelinePath, campaignPath := writeInspectionFamilyConfig(t)
+	var out bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath}, &out); err != nil {
+		t.Fatal(err)
+	}
+	records := catalogRecordsByKey(t, out.String())
+	meta := records["character_meta_arannis"]
+	note := records["character_note_arannis"]
+	partyPath := filepath.Join(filepath.Dir(campaignPath), "party.yml")
+	if len(meta) != 10 || meta[6] != "character_meta" || meta[7] != "arannis" || meta[8] != pipelinePath || meta[9] != partyPath {
+		t.Fatalf("generated meta record = %#v", meta)
+	}
+	if len(note) != 10 || note[5] != "character_meta_arannis" || note[6] != "character_note" {
+		t.Fatalf("generated note record = %#v", note)
+	}
+}
+
+func TestConfigArtifactsReportsProfiledFamilyContributors(t *testing.T) {
+	pipelinePath, campaignPath := writeInspectionFamilyConfig(t)
+	profilePath := filepath.Join(filepath.Dir(pipelinePath), "profiles", "testing.yml")
+	mustWriteTestFile(t, profilePath, `scriptorium:
+  artifact_families:
+    character_note:
+      prompt_id: dnd.character_note.testing
+`)
+
+	args := []string{"--config", pipelinePath, "--campaign-file", campaignPath, "--profile", "testing"}
+	var catalog bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), args, &catalog); err != nil {
+		t.Fatal(err)
+	}
+	records := catalogRecordsByKey(t, catalog.String())
+	if got := strings.Split(records["character_note_arannis"][2], ","); !reflect.DeepEqual(got, []string{pipelinePath, profilePath}) {
+		t.Fatalf("generated family definitions = %#v", got)
+	}
+
+	var sources bytes.Buffer
+	if err := ConfigSources(context.Background(), args, &sources); err != nil {
+		t.Fatal(err)
+	}
+	want := "scriptorium.artifacts.character_note_arannis.prompt_id\tfamily\t" + profilePath
+	if !strings.Contains(sources.String(), want) {
+		t.Fatalf("sources output missing profiled family contributor %q:\n%s", want, sources.String())
+	}
+}
+
+func TestConfigArtifactsReportsBaseAndProfileContributors(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	profilePath := filepath.Join(dir, "profiles", "testing.yml")
+	mustWriteTestFile(t, pipelinePath, `workspace:
+  root: /srv/narratio
+whisperx:
+  transcribe_url: https://transcription.example.com/transcribe
+scriptorium:
+  binary: scriptorium
+  artifacts:
+    notes:
+      enabled: false
+      output_path: artifacts/notes.md
+`)
+	mustWriteTestFile(t, profilePath, `scriptorium:
+  artifacts:
+    notes:
+      enabled: true
+      prompt_id: notes
+    profile_only:
+      enabled: false
+      output_path: artifacts/profile-only.md
+`)
+
+	var out bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath, "--profile", "testing"}, &out); err != nil {
+		t.Fatal(err)
+	}
+	records := catalogRecordsByKey(t, out.String())
+	if got := strings.Split(records["notes"][2], ","); !reflect.DeepEqual(got, []string{pipelinePath, profilePath}) {
+		t.Fatalf("notes definitions = %#v", got)
+	}
+	if got := records["profile_only"][2]; got != profilePath {
+		t.Fatalf("profile-only definition = %q, want %q", got, profilePath)
+	}
+}
+
+func TestConfigArtifactsIdentityDoesNotDependOnDefinitionFilename(t *testing.T) {
+	inlineDir := t.TempDir()
+	inlinePath := filepath.Join(inlineDir, "pipeline.yml")
+	mustWriteTestFile(t, inlinePath, artifactIdentityPipeline("", "recap"))
+	importDir := t.TempDir()
+	importPath := filepath.Join(importDir, "pipeline.yml")
+	mustWriteTestFile(t, importPath, artifactIdentityPipeline("artifacts.yml", ""))
+	mustWriteTestFile(t, filepath.Join(importDir, "artifacts.yml"), "scriptorium:\n  artifacts:\n    recap:\n      enabled: false\n      output_path: artifacts/recap.md\n")
+
+	for _, path := range []string{inlinePath, importPath} {
+		var out bytes.Buffer
+		if err := ConfigArtifacts(context.Background(), []string{"--config", path}, &out); err != nil {
+			t.Fatal(err)
+		}
+		if got := catalogRecordsByKey(t, out.String())["recap"][1]; got != "narratio.artifact.recap" {
+			t.Fatalf("source identity for %s = %q", path, got)
+		}
+	}
+}
+
+func TestConfigArtifactsPropagatesWriterErrorsAndEscapesFields(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	mustWriteTestFile(t, pipelinePath, "workspace:\n  root: /srv/narratio\nwhisperx:\n  transcribe_url: https://transcription.example.com/transcribe\n")
+	want := errors.New("write rejected")
+	err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath}, rejectingCatalogWriter{err: want})
+	if !errors.Is(err, want) {
+		t.Fatalf("ConfigArtifacts() error = %v, want writer error", err)
+	}
+	if got := escapeCatalogField("a\\b\tc\nd\r"); got != `a\\b\tc\nd\r` {
+		t.Fatalf("escaped field = %q", got)
+	}
+}
+
+func catalogRecordsByKey(t *testing.T, output string) map[string][]string {
+	t.Helper()
+	lines := strings.Split(strings.TrimSuffix(output, "\n"), "\n")
+	if len(lines) == 0 || lines[0]+"\n" != artifactCatalogHeader {
+		t.Fatalf("catalog header = %q", output)
+	}
+	records := make(map[string][]string, len(lines)-1)
+	for _, line := range lines[1:] {
+		fields := strings.Split(line, "\t")
+		records[fields[0]] = fields
+	}
+	return records
+}
+
+func artifactIdentityPipeline(importPath, inlineKey string) string {
+	document := "workspace:\n  root: /srv/narratio\nwhisperx:\n  transcribe_url: https://transcription.example.com/transcribe\nscriptorium:\n  binary: scriptorium\n"
+	if importPath != "" {
+		document = "configkit:\n  imports:\n    - path: " + importPath + "\n" + document
+	}
+	if inlineKey != "" {
+		document += "  artifacts:\n    " + inlineKey + ":\n      enabled: false\n      output_path: artifacts/recap.md\n"
+	}
+	return document
+}
+
+type rejectingCatalogWriter struct{ err error }
+
+func (w rejectingCatalogWriter) Write([]byte) (int, error) { return 0, w.err }

internal/app/config_commands.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"io"
 	"sort"
+	"strconv"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
@@ -17,6 +18,7 @@ type inspectionFlags struct {
 	campaignPath     string
 	campaignFilePath string
 	profile          singletonStringFlag
+	secretsDir       singletonStringFlag
 }
 
 type inspectionConfig struct {
@@ -35,7 +37,7 @@ func Config(ctx context.Context, args []string, out io.Writer) error {
 func configWithDependencies(ctx context.Context, args []string, out io.Writer, deps commandDependencies) error {
 	deps = normalizeCommandDependencies(deps)
 	if len(args) == 0 {
-		return fmt.Errorf("config: expected subcommand: validate|show|sources|diff")
+		return fmt.Errorf("config: expected subcommand: validate|show|sources|artifacts|diff")
 	}
 	if args[0] == "--help" || args[0] == "-h" {
 		writeConfigCommandUsage(out)
@@ -48,8 +50,10 @@ func configWithDependencies(ctx context.Context, args []string, out io.Writer, d
 		return configShowWithDependencies(ctx, args[1:], out, deps)
 	case "sources":
 		return configSourcesWithDependencies(ctx, args[1:], out, deps)
+	case "artifacts":
+		return configArtifactsWithDependencies(ctx, args[1:], out, deps)
 	case "diff":
-		return ConfigDiff(ctx, args[1:], out)
+		return configDiffWithDependencies(ctx, args[1:], out, deps)
 	default:
 		return fmt.Errorf("config: unknown subcommand %q", args[0])
 	}
@@ -139,6 +143,70 @@ func configSourcesWithDependencies(_ context.Context, args []string, out io.Writ
 	return nil
 }
 
+// ConfigArtifacts validates and reports the effective concrete artifact catalog
+// without loading session state or runtime collaborators.
+func ConfigArtifacts(ctx context.Context, args []string, out io.Writer) error {
+	return configArtifactsWithDependencies(ctx, args, out, commandDependencies{})
+}
+
+func configArtifactsWithDependencies(_ context.Context, args []string, out io.Writer, deps commandDependencies) error {
+	flags, err := parseInspectionFlags("config artifacts", args, out)
+	if err != nil {
+		if errors.Is(err, flag.ErrHelp) {
+			return nil
+		}
+		return err
+	}
+	resolved, err := resolveInspectionConfigWithDependencies(flags, deps)
+	if err != nil {
+		return fmt.Errorf("config artifacts: %w", err)
+	}
+	records, err := config.EffectiveArtifactCatalog(resolved.Pipeline, resolved.Party)
+	if err != nil {
+		return fmt.Errorf("config artifacts: %w", err)
+	}
+	if _, err := fmt.Fprintln(out, "key\tsource\tdefinitions\tenabled\toutput\tdependencies\tfamily\tcharacter\tfamily_source\tparty_source"); err != nil {
+		return err
+	}
+	for _, record := range records {
+		fields := []string{
+			record.Key,
+			record.SourceID,
+			strings.Join(normalizedCatalogPaths(record.DefinitionSources), ","),
+			strconv.FormatBool(record.Enabled),
+			record.OutputPath,
+			strings.Join(record.Dependencies, ","),
+			record.FamilyKey,
+			record.CharacterID,
+			config.NormalizedConfigurationPath(record.FamilySource),
+			config.NormalizedConfigurationPath(record.PartySource),
+		}
+		for index := range fields {
+			fields[index] = escapeCatalogField(fields[index])
+		}
+		if _, err := fmt.Fprintln(out, strings.Join(fields, "\t")); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func normalizedCatalogPaths(paths []string) []string {
+	result := make([]string, len(paths))
+	for index, path := range paths {
+		result[index] = config.NormalizedConfigurationPath(path)
+	}
+	sort.Strings(result)
+	return result
+}
+
+func escapeCatalogField(value string) string {
+	value = strings.ReplaceAll(value, "\\", "\\\\")
+	value = strings.ReplaceAll(value, "\t", "\\t")
+	value = strings.ReplaceAll(value, "\r", "\\r")
+	return strings.ReplaceAll(value, "\n", "\\n")
+}
+
 func parseInspectionFlags(command string, args []string, out io.Writer) (inspectionFlags, error) {
 	fs := flag.NewFlagSet(command, flag.ContinueOnError)
 	fs.SetOutput(out)
@@ -148,8 +216,10 @@ func parseInspectionFlags(command string, args []string, out io.Writer) (inspect
 	fs.StringVar(&flags.campaignFilePath, "campaign-file", "", "path to campaign.yml")
 	flags.profile.name = "profile"
 	fs.Var(&flags.profile, "profile", "named pipeline profile")
+	flags.secretsDir.name = "secrets-dir"
+	fs.Var(&flags.secretsDir, "secrets-dir", "secrets directory (optional; defaults beside pipeline.yml)")
 	fs.Usage = func() {
-		fmt.Fprintf(out, "Usage: narratio %s [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>]\n\n", command)
+		fmt.Fprintf(out, "Usage: narratio %s [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--profile <name>] [--secrets-dir <path>]\n\n", command)
 		fmt.Fprintln(out, "Resolves configuration without a session, workspace, manifest, adapters, or external services.")
 		fmt.Fprintln(out)
 		fmt.Fprintln(out, "Flags:")
@@ -168,7 +238,7 @@ func parseInspectionFlags(command string, args []string, out io.Writer) (inspect
 }
 
 func resolveInspectionConfigWithDependencies(flags inspectionFlags, deps commandDependencies) (*inspectionConfig, error) {
-	pipelinePath, pipeline, err := loadPipelineConfigWithDependencies(flags.pipelinePath, config.PipelineLoadOptions{Profile: flags.profile.pointer()}, deps)
+	pipelinePath, pipeline, err := loadPipelineConfigWithDependencies(flags.pipelinePath, config.PipelineLoadOptions{Profile: flags.profile.pointer()}, flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return nil, err
 	}
@@ -253,9 +323,9 @@ func inspectionProfile(pipeline *config.PipelineConfig) string {
 }
 
 func writeConfigCommandUsage(out io.Writer) {
-	fmt.Fprintln(out, "Usage: narratio config <validate|show|sources|diff>")
+	fmt.Fprintln(out, "Usage: narratio config <validate|show|sources|artifacts|diff>")
 	fmt.Fprintln(out)
-	fmt.Fprintln(out, "Use config validate to check an effective pipeline, config show to print normalized YAML, config sources to report ownership, or config diff to compare two profiles.")
+	fmt.Fprintln(out, "Use config validate to check an effective pipeline, config show to print normalized YAML, config sources to report ownership, config artifacts to list concrete artifacts, or config diff to compare two profiles.")
 }
 
 func writeInspectionSourceHeader(out io.Writer, resolved *inspectionConfig) error {

internal/app/config_commands_test.go

@@ -11,6 +11,7 @@ import (
 	"strings"
 	"testing"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
@@ -21,13 +22,14 @@ func TestConfigCommandHelpAndDispatch(t *testing.T) {
 		want string
 		code int
 	}{
-		{name: "top level help", args: []string{"config", "--help"}, want: "Usage: narratio config <validate|show|sources|diff>"},
+		{name: "top level help", args: []string{"config", "--help"}, want: "Usage: narratio config <validate|show|sources|artifacts|diff>"},
 		{name: "validate help", args: []string{"config", "validate", "--help"}, want: "Usage: narratio config validate"},
 		{name: "show help", args: []string{"config", "show", "--help"}, want: "Usage: narratio config show"},
 		{name: "sources help", args: []string{"config", "sources", "--help"}, want: "Usage: narratio config sources"},
+		{name: "artifacts help", args: []string{"config", "artifacts", "--help"}, want: "Usage: narratio config artifacts"},
 		{name: "diff help", args: []string{"config", "diff", "--help"}, want: "Usage: narratio config diff"},
 		{name: "unknown", args: []string{"config", "unknown"}, want: `config: unknown subcommand "unknown"`, code: 1},
-		{name: "missing", args: []string{"config"}, want: "config: expected subcommand: validate|show|sources|diff", code: 1},
+		{name: "missing", args: []string{"config"}, want: "config: expected subcommand: validate|show|sources|artifacts|diff", code: 1},
 		{name: "session flag", args: []string{"config", "validate", "--session", "session.yml"}, want: "flag provided but not defined", code: 1},
 		{name: "stage flag", args: []string{"config", "show", "--from", "prepare"}, want: "flag provided but not defined", code: 1},
 		{name: "force flag", args: []string{"config", "show", "--force"}, want: "flag provided but not defined", code: 1},
@@ -52,7 +54,7 @@ func TestExecuteConfigUsesInvocationPipelineLoader(t *testing.T) {
 	pipelinePath := filepath.Join(dir, "pipeline.yml")
 	mustWriteTestFile(t, pipelinePath, "workspace:\n  root: "+filepath.Join(dir, "workspace")+"\nwhisperx:\n  transcribe_url: https://transcription.example.com/transcribe\n")
 
-	for _, subcommand := range []string{"validate", "show", "sources"} {
+	for _, subcommand := range []string{"validate", "show", "sources", "artifacts"} {
 		t.Run(subcommand, func(t *testing.T) {
 			calls := 0
 			deps := commandDependencies{
@@ -91,8 +93,6 @@ func TestConfigValidateAndShowPipelineOnlyAreSideEffectFree(t *testing.T) {
 	defer server.Close()
 	mustWriteTestFile(t, pipelinePath, `workspace:
   root: `+workspaceRoot+`
-secrets:
-  env_dir: `+secretsDir+`
 whisperx:
   transcribe_url: `+server.URL+`
 `)
@@ -207,6 +207,56 @@ scriptorium:
 	}
 }
 
+func TestConfigShowPreservesNullAndEmptyCollections(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	mustWriteTestFile(t, pipelinePath, `workspace:
+  root: /tmp/narratio-work
+whisperx:
+  transcribe_url: https://transcription.example.test/transcribe
+audita:
+  modules: null
+scriptorium:
+  artifacts: null
+`)
+	mustWriteTestFile(t, filepath.Join(dir, "profiles", "empty.yml"), `audita:
+  modules: []
+scriptorium:
+  artifacts: {}
+`)
+
+	var base, empty bytes.Buffer
+	if err := ConfigShow(context.Background(), []string{"--config", pipelinePath}, &base); err != nil {
+		t.Fatal(err)
+	}
+	if err := ConfigShow(context.Background(), []string{"--config", pipelinePath, "--profile", "empty"}, &empty); err != nil {
+		t.Fatal(err)
+	}
+	for _, want := range []string{"modules: null\n", "artifacts: null\n"} {
+		if !strings.Contains(base.String(), want) {
+			t.Fatalf("base show output missing %q:\n%s", want, base.String())
+		}
+	}
+	for _, want := range []string{"modules: []\n", "artifacts: {}\n"} {
+		if !strings.Contains(empty.String(), want) {
+			t.Fatalf("empty-profile show output missing %q:\n%s", want, empty.String())
+		}
+	}
+
+	var diff bytes.Buffer
+	if err := ConfigDiff(context.Background(), []string{"default", "empty", "--config", pipelinePath}, &diff); err != nil {
+		t.Fatal(err)
+	}
+	for _, want := range []string{
+		"changed\taudita.modules\tnull\t[]\n",
+		"changed\tscriptorium.artifacts\tnull\t{}\n",
+	} {
+		if !strings.Contains(diff.String(), want) {
+			t.Fatalf("diff output missing %q:\n%s", want, diff.String())
+		}
+	}
+}
+
 func TestConfigDiffComparesExpandedFamiliesAndPublishRules(t *testing.T) {
 	pipelinePath, campaignPath := writeInspectionFamilyConfig(t)
 	writeInspectionProfiles(t, pipelinePath, `scriptorium:
@@ -276,6 +326,137 @@ func TestConfigDiffReportsEqualityAndRejectsInvalidInput(t *testing.T) {
 	}
 }
 
+func TestConfigDiffBootstrapsSecretsOnceForBothProfiles(t *testing.T) {
+	workspaceRoot := t.TempDir()
+	pipelinePath, _, _ := writeValidConfigFiles(t, workspaceRoot)
+	writeInspectionProfiles(t, pipelinePath, "whisperx:\n  language: en\n", "whisperx:\n  language: fr\n")
+	secretsDir := t.TempDir()
+	const secretName = "NARRATIO_DIFF_BOOTSTRAP_REPORT"
+	restoreEnvAfterTest(t, secretName)
+	if err := os.Unsetenv(secretName); err != nil {
+		t.Fatal(err)
+	}
+	mustWriteSecretFile(t, filepath.Join(secretsDir, secretName), "diff-secret-must-not-appear\n")
+	bootstrapCalls := 0
+	var diagnostics bytes.Buffer
+	deps := commandDependencies{loadSecretsDir: func(got string) (*configkit.SecretsReport, error) {
+		bootstrapCalls++
+		if got != secretsDir {
+			t.Fatalf("secrets directory = %q, want %q", got, secretsDir)
+		}
+		return configkit.LoadSecretsDir(got)
+	}, diagnostics: &diagnostics}
+
+	if err := configDiffWithDependencies(context.Background(), []string{
+		"production", "testing", "--config", pipelinePath, "--secrets-dir", secretsDir,
+	}, io.Discard, deps); err != nil {
+		t.Fatalf("config diff: %v", err)
+	}
+	if bootstrapCalls != 1 {
+		t.Fatalf("bootstrap calls = %d, want 1", bootstrapCalls)
+	}
+	if got := diagnostics.String(); strings.Count(got, `name="`+secretName+`"`) != 1 {
+		t.Fatalf("bootstrap report count = %d, want 1; diagnostics = %q", strings.Count(got, `name="`+secretName+`"`), got)
+	}
+	if strings.Contains(diagnostics.String(), "diff-secret-must-not-appear") {
+		t.Fatalf("bootstrap diagnostics exposed secret value: %q", diagnostics.String())
+	}
+}
+
+func TestConfigInspectionOutputsNeverExposeResolvedCredentialValues(t *testing.T) {
+	const firstCanary = "inspection-credential-canary-one"
+	const secondCanary = "inspection-credential-canary-two"
+	for name, value := range map[string]string{
+		"NARRATIO_INSPECT_ACCESS":       firstCanary,
+		"NARRATIO_INSPECT_SECRET":       firstCanary,
+		"NARRATIO_INSPECT_LLM_ONE":      firstCanary,
+		"NARRATIO_INSPECT_LLM_TWO":      secondCanary,
+		"NARRATIO_INSPECT_NOTARIUS_ONE": firstCanary,
+		"NARRATIO_INSPECT_NOTARIUS_TWO": secondCanary,
+	} {
+		t.Setenv(name, value)
+	}
+	workspaceRoot := t.TempDir()
+	pipelinePath, _, _ := writeValidConfigFiles(t, workspaceRoot)
+	pipelineData, err := os.ReadFile(pipelinePath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	pipelineYAML := strings.Replace(string(pipelineData), "    bucket: test-bucket\n", `    bucket: test-bucket
+    access_key_id: NARRATIO_INSPECT_ACCESS
+    secret_access_key: NARRATIO_INSPECT_SECRET
+`, 1)
+	pipelineYAML += `notarius:
+  enabled: false
+  llm_api_key: NARRATIO_INSPECT_NOTARIUS_ONE
+`
+	if err := os.WriteFile(pipelinePath, []byte(pipelineYAML), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	writeInspectionProfiles(t, pipelinePath,
+		"audita:\n  model: production\n  llm_api_key: NARRATIO_INSPECT_LLM_ONE\n",
+		"audita:\n  model: testing\n  llm_api_key: NARRATIO_INSPECT_LLM_TWO\nnotarius:\n  llm_api_key: NARRATIO_INSPECT_NOTARIUS_TWO\n",
+	)
+	mustWriteTestFile(t, filepath.Join(filepath.Dir(pipelinePath), "profiles", "without-notarius-key.yml"), "notarius:\n  llm_api_key: null\n")
+
+	var validate bytes.Buffer
+	if err := ConfigValidate(context.Background(), []string{"--config", pipelinePath, "--profile", "production"}, &validate); err != nil {
+		t.Fatalf("config validate: %v", err)
+	}
+
+	var show bytes.Buffer
+	if err := ConfigShow(context.Background(), []string{"--config", pipelinePath, "--profile", "production"}, &show); err != nil {
+		t.Fatalf("config show: %v", err)
+	}
+	var sources bytes.Buffer
+	if err := ConfigSources(context.Background(), []string{"--config", pipelinePath, "--profile", "production"}, &sources); err != nil {
+		t.Fatalf("config sources: %v", err)
+	}
+	var catalog bytes.Buffer
+	if err := ConfigArtifacts(context.Background(), []string{"--config", pipelinePath, "--profile", "production"}, &catalog); err != nil {
+		t.Fatalf("config artifacts: %v", err)
+	}
+	var diff bytes.Buffer
+	if err := ConfigDiff(context.Background(), []string{"production", "testing", "--config", pipelinePath}, &diff); err != nil {
+		t.Fatalf("config diff: %v", err)
+	}
+	var removed bytes.Buffer
+	if err := ConfigShow(context.Background(), []string{"--config", pipelinePath, "--profile", "without-notarius-key"}, &removed); err != nil {
+		t.Fatalf("config show with removed Notarius credential: %v", err)
+	}
+	for label, output := range map[string]string{"show": show.String(), "sources": sources.String(), "artifacts": catalog.String(), "diff": diff.String()} {
+		if strings.Contains(output, firstCanary) || strings.Contains(output, secondCanary) {
+			t.Fatalf("%s output exposed credential value: %q", label, output)
+		}
+	}
+	for _, reference := range []string{"NARRATIO_INSPECT_ACCESS", "NARRATIO_INSPECT_SECRET", "NARRATIO_INSPECT_LLM_ONE", "NARRATIO_INSPECT_NOTARIUS_ONE"} {
+		if !strings.Contains(show.String(), reference) {
+			t.Fatalf("show omitted safe credential reference %q: %q", reference, show.String())
+		}
+	}
+	if !strings.Contains(diff.String(), "NARRATIO_INSPECT_LLM_ONE") || !strings.Contains(diff.String(), "NARRATIO_INSPECT_LLM_TWO") ||
+		!strings.Contains(diff.String(), "NARRATIO_INSPECT_NOTARIUS_ONE") || !strings.Contains(diff.String(), "NARRATIO_INSPECT_NOTARIUS_TWO") {
+		t.Fatalf("diff omitted safe credential references: %q", diff.String())
+	}
+	if !strings.Contains(sources.String(), "notarius.llm_api_key") {
+		t.Fatalf("sources omitted Notarius credential ownership: %q", sources.String())
+	}
+	if strings.Contains(removed.String(), "llm_api_key") || strings.Contains(removed.String(), "NARRATIO_INSPECT_NOTARIUS_ONE") {
+		t.Fatalf("profile null did not remove Notarius credential: %q", removed.String())
+	}
+}
+
+func TestConfigValidateRejectsUnavailableNotariusCredential(t *testing.T) {
+	const envName = "NARRATIO_INSPECT_NOTARIUS_MISSING"
+	restoreEnvAfterTest(t, envName)
+	path := filepath.Join(t.TempDir(), "pipeline.yml")
+	mustWriteTestFile(t, path, "whisperx:\n  transcribe_url: https://transcription.example.test/transcribe\nnotarius:\n  enabled: false\n  llm_api_key: "+envName+"\n")
+	err := ConfigValidate(context.Background(), []string{"--config", path}, io.Discard)
+	if err == nil || !strings.Contains(err.Error(), "secret.unset") {
+		t.Fatalf("ConfigValidate() error = %v, want unavailable credential failure", err)
+	}
+}
+
 func TestConfigCommandsInspectMaintainedSplitBundleWithoutRuntimeState(t *testing.T) {
 	examplesDir := filepath.Join("..", "..", "examples")
 	pipelinePath := filepath.Join(examplesDir, "production-testing", "pipeline.yml")
@@ -285,13 +466,13 @@ func TestConfigCommandsInspectMaintainedSplitBundleWithoutRuntimeState(t *testin
 		t.Fatalf("example workspace stat = %v, want absent", err)
 	}
 
-	for _, command := range []func(context.Context, []string, io.Writer) error{ConfigValidate, ConfigShow, ConfigSources} {
-		if err := command(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath, "--profile", "production"}, io.Discard); err != nil {
+	for _, command := range []func(context.Context, []string, io.Writer) error{ConfigValidate, ConfigShow, ConfigSources, ConfigArtifacts} {
+		if err := command(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath}, io.Discard); err != nil {
 			t.Fatalf("inspection command %T error = %v", command, err)
 		}
 	}
 	var diff bytes.Buffer
-	if err := ConfigDiff(context.Background(), []string{"production", "testing", "--config", pipelinePath, "--campaign-file", campaignPath}, &diff); err != nil {
+	if err := ConfigDiff(context.Background(), []string{"default", "testing", "--config", pipelinePath, "--campaign-file", campaignPath}, &diff); err != nil {
 		t.Fatalf("ConfigDiff() error = %v", err)
 	}
 	if !strings.Contains(diff.String(), "audita.model") {
@@ -304,17 +485,9 @@ func TestConfigCommandsInspectMaintainedSplitBundleWithoutRuntimeState(t *testin
 
 func writeInspectionProfiles(t *testing.T, pipelinePath, production, testing string) {
 	t.Helper()
-	data, err := os.ReadFile(pipelinePath)
-	if err != nil {
-		t.Fatal(err)
-	}
-	composition := "composition:\n  default_profile: production\n  profiles:\n    production:\n      overlay: production.yml\n    testing:\n      overlay: testing.yml\n"
-	if err := os.WriteFile(pipelinePath, append([]byte(composition), data...), 0o644); err != nil {
-		t.Fatal(err)
-	}
-	dir := filepath.Dir(pipelinePath)
-	mustWriteTestFile(t, filepath.Join(dir, "production.yml"), production)
-	mustWriteTestFile(t, filepath.Join(dir, "testing.yml"), testing)
+	profilesDir := filepath.Join(filepath.Dir(pipelinePath), "profiles")
+	mustWriteTestFile(t, filepath.Join(profilesDir, "production.yml"), production)
+	mustWriteTestFile(t, filepath.Join(profilesDir, "testing.yml"), testing)
 }
 
 func TestConfigCommandsSelectProfilesAndCampaigns(t *testing.T) {
@@ -326,7 +499,7 @@ func TestConfigCommandsSelectProfilesAndCampaigns(t *testing.T) {
 	if err := ConfigValidate(context.Background(), []string{"--config", pipelinePath}, &defaultOut); err != nil {
 		t.Fatalf("default ConfigValidate() error = %v", err)
 	}
-	if !strings.Contains(defaultOut.String(), "profile=production") {
+	if !strings.Contains(defaultOut.String(), "profile=none") {
 		t.Fatalf("default output = %q", defaultOut.String())
 	}
 
@@ -352,7 +525,7 @@ func TestConfigCommandsSelectProfilesAndCampaigns(t *testing.T) {
 func TestConfigCommandsExpandFamiliesAndRequireCampaign(t *testing.T) {
 	pipelinePath, campaignPath := writeInspectionFamilyConfig(t)
 
-	for _, command := range []func(context.Context, []string, io.Writer) error{ConfigValidate, ConfigShow} {
+	for _, command := range []func(context.Context, []string, io.Writer) error{ConfigValidate, ConfigShow, ConfigArtifacts} {
 		var out bytes.Buffer
 		err := command(context.Background(), []string{"--config", pipelinePath}, &out)
 		if err == nil || !strings.Contains(err.Error(), "artifact_families requires a campaign") {
@@ -398,8 +571,9 @@ func TestConfigValidateReportsCanonicalPartyAndImportErrors(t *testing.T) {
 
 	dir := t.TempDir()
 	rootPath := filepath.Join(dir, "pipeline.yml")
-	mustWriteTestFile(t, rootPath, `composition:
-  imports: [conf.yml]
+	mustWriteTestFile(t, rootPath, `configkit:
+  imports:
+    - path: conf.yml
 workspace:
   root: /one
 whisperx:
@@ -407,7 +581,7 @@ whisperx:
 `)
 	mustWriteTestFile(t, filepath.Join(dir, "conf.yml"), "workspace:\n  root: /two\n")
 	err := ConfigValidate(context.Background(), []string{"--config", rootPath}, io.Discard)
-	if err == nil || !strings.Contains(err.Error(), "workspace.root") || !strings.Contains(err.Error(), "pipeline.yml") || !strings.Contains(err.Error(), "conf.yml") {
+	if err == nil || !strings.Contains(err.Error(), "/workspace/root") || !strings.Contains(err.Error(), "pipeline.yml") || !strings.Contains(err.Error(), "conf.yml") {
 		t.Fatalf("import diagnostic = %v", err)
 	}
 }
@@ -432,7 +606,7 @@ func TestConfigSourcesReportsStableOwnership(t *testing.T) {
 	for _, want := range []string{
 		"root: ",
 		"imports: none",
-		"profile: name=testing selection=cli overlay=",
+		"profile: name=testing selection=explicit overlay=",
 		"campaign: id=sample-campaign source=",
 		"party: source=",
 		"digest: ",
@@ -475,15 +649,7 @@ func TestConfigSourcesReportsCanonicalFamilyAndPublishOrigins(t *testing.T) {
 
 func TestConfigSourcesPreservesProfileOwnershipThroughFamilyExpansion(t *testing.T) {
 	pipelinePath, campaignPath := writeInspectionFamilyConfig(t)
-	data, err := os.ReadFile(pipelinePath)
-	if err != nil {
-		t.Fatal(err)
-	}
-	profilePath := filepath.Join(filepath.Dir(pipelinePath), "testing.yml")
-	composition := "composition:\n  default_profile: testing\n  profiles:\n    testing:\n      overlay: testing.yml\n"
-	if err := os.WriteFile(pipelinePath, append([]byte(composition), data...), 0o644); err != nil {
-		t.Fatal(err)
-	}
+	profilePath := filepath.Join(filepath.Dir(pipelinePath), "profiles", "testing.yml")
 	mustWriteTestFile(t, profilePath, `scriptorium:
   artifact_families:
     character_note:
@@ -492,7 +658,7 @@ func TestConfigSourcesPreservesProfileOwnershipThroughFamilyExpansion(t *testing
 `)
 
 	var out bytes.Buffer
-	if err := ConfigSources(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath}, &out); err != nil {
+	if err := ConfigSources(context.Background(), []string{"--config", pipelinePath, "--campaign-file", campaignPath, "--profile", "testing"}, &out); err != nil {
 		t.Fatalf("ConfigSources() error = %v", err)
 	}
 	want := "scriptorium.artifacts.character_note_arannis.enabled\tfamily\t" + profilePath
@@ -501,17 +667,74 @@ func TestConfigSourcesPreservesProfileOwnershipThroughFamilyExpansion(t *testing
 	}
 }
 
+func TestConfigSourcesAttributesInferredArtifactDependencies(t *testing.T) {
+	dir := t.TempDir()
+	pipelinePath := filepath.Join(dir, "pipeline.yml")
+	inputPath := filepath.Join(dir, "input-only.yml")
+	profilePath := filepath.Join(dir, "profiles", "testing.yml")
+	mustWriteTestFile(t, pipelinePath, `configkit:
+  imports:
+    - path: input-only.yml
+workspace:
+  root: /srv/narratio
+whisperx:
+  transcribe_url: https://transcription.example.test/transcribe
+scriptorium:
+  artifacts:
+    first:
+      enabled: false
+      output_path: artifacts/first.md
+    second:
+      enabled: false
+      output_path: artifacts/second.md
+    mixed:
+      enabled: false
+      output_path: artifacts/mixed.md
+      depends_on: [first]
+`)
+	mustWriteTestFile(t, inputPath, `scriptorium:
+  artifacts:
+    input_only:
+      enabled: false
+      output_path: artifacts/input-only.md
+      inputs:
+        prior:
+          source: narratio.artifact.first
+`)
+	mustWriteTestFile(t, profilePath, `scriptorium:
+  artifacts:
+    mixed:
+      inputs:
+        prior:
+          source: narratio.artifact.second
+`)
+
+	var out bytes.Buffer
+	if err := ConfigSources(context.Background(), []string{"--config", pipelinePath, "--profile", "testing"}, &out); err != nil {
+		t.Fatal(err)
+	}
+	for _, want := range []string{
+		"scriptorium.artifacts.input_only.depends_on\timport\t" + inputPath,
+		"scriptorium.artifacts.mixed.depends_on\troot\t" + pipelinePath,
+		"scriptorium.artifacts.mixed.depends_on\tprofile\t" + profilePath,
+	} {
+		if !strings.Contains(out.String(), want) {
+			t.Fatalf("sources output missing dependency contributor %q:\n%s", want, out.String())
+		}
+	}
+	if strings.Contains(out.String(), "scriptorium.artifacts.input_only.depends_on\tdefault\tdefault") {
+		t.Fatalf("inferred dependency retained default attribution:\n%s", out.String())
+	}
+}
+
 func TestConfigSourcesReportsRootImportProfileAndDefaultOwnership(t *testing.T) {
 	dir := t.TempDir()
 	rootPath := filepath.Join(dir, "pipeline.yml")
 	importPath := filepath.Join(dir, "base.yml")
-	profilePath := filepath.Join(dir, "testing.yml")
-	mustWriteTestFile(t, rootPath, `composition:
-  imports: [base.yml]
-  default_profile: testing
-  profiles:
-    testing:
-      overlay: testing.yml
+	profilePath := filepath.Join(dir, "profiles", "testing.yml")
+	mustWriteTestFile(t, rootPath, `configkit:
+  imports:
+    - path: base.yml
 workspace:
   root: /srv/narratio
 whisperx:
@@ -521,13 +744,13 @@ whisperx:
 	mustWriteTestFile(t, profilePath, "whisperx:\n  language: fr\n")
 
 	var out bytes.Buffer
-	if err := ConfigSources(context.Background(), []string{"--config", rootPath}, &out); err != nil {
+	if err := ConfigSources(context.Background(), []string{"--config", rootPath, "--profile", "testing"}, &out); err != nil {
 		t.Fatalf("ConfigSources() error = %v", err)
 	}
 	output := out.String()
 	for _, want := range []string{
 		"import: " + importPath,
-		"profile: name=testing selection=default overlay=" + profilePath,
+		"profile: name=testing selection=explicit overlay=" + profilePath,
 		"workspace.root\troot\t" + rootPath,
 		"storage.backend\timport\t" + importPath,
 		"whisperx.language\tprofile\t" + profilePath,

internal/app/config_diff.go

@@ -16,6 +16,7 @@ type configDiffFlags struct {
 	pipelinePath     singletonStringFlag
 	campaignPath     singletonStringFlag
 	campaignFilePath singletonStringFlag
+	secretsDir       singletonStringFlag
 }
 
 type configDiffRequest struct {
@@ -40,7 +41,11 @@ type configDiffRecord struct {
 // read-only configuration resolution boundary used by config validate, show,
 // and sources. Differences describe normalized effective values rather than
 // the layout of root, import, or overlay files.
-func ConfigDiff(_ context.Context, args []string, out io.Writer) error {
+func ConfigDiff(ctx context.Context, args []string, out io.Writer) error {
+	return configDiffWithDependencies(ctx, args, out, commandDependencies{})
+}
+
+func configDiffWithDependencies(_ context.Context, args []string, out io.Writer, deps commandDependencies) error {
 	request, err := parseConfigDiffRequest(args, out)
 	if err != nil {
 		if errors.Is(err, flag.ErrHelp) {
@@ -48,7 +53,7 @@ func ConfigDiff(_ context.Context, args []string, out io.Writer) error {
 		}
 		return err
 	}
-	resolved, err := resolveConfigDiff(request)
+	resolved, err := resolveConfigDiff(request, deps)
 	if err != nil {
 		return fmt.Errorf("config diff: %w", err)
 	}
@@ -102,11 +107,13 @@ func parseConfigDiffRequest(args []string, out io.Writer) (configDiffRequest, er
 	request.flags.pipelinePath.name = "config"
 	request.flags.campaignPath.name = "campaign"
 	request.flags.campaignFilePath.name = "campaign-file"
+	request.flags.secretsDir.name = "secrets-dir"
 	fs.Var(&request.flags.pipelinePath, "config", "path to pipeline.yml (optional; defaults searched)")
 	fs.Var(&request.flags.campaignPath, "campaign", "campaign ID")
 	fs.Var(&request.flags.campaignFilePath, "campaign-file", "path to campaign.yml")
+	fs.Var(&request.flags.secretsDir, "secrets-dir", "secrets directory (optional; defaults beside pipeline.yml)")
 	fs.Usage = func() {
-		fmt.Fprintln(out, "Usage: narratio config diff <left-profile> <right-profile> [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>]")
+		fmt.Fprintln(out, "Usage: narratio config diff <left-profile> <right-profile> [--config <pipeline.yml>] [--campaign <id> | --campaign-file <campaign.yml>] [--secrets-dir <path>]")
 		fmt.Fprintln(out)
 		fmt.Fprintln(out, "Compares fully resolved profiles without a session, workspace, manifest, adapters, or external services.")
 		fmt.Fprintln(out)
@@ -120,6 +127,9 @@ func parseConfigDiffRequest(args []string, out io.Writer) (configDiffRequest, er
 	if len(profiles) != 2 || strings.TrimSpace(profiles[0]) == "" || strings.TrimSpace(profiles[1]) == "" {
 		return configDiffRequest{}, fmt.Errorf("config diff: exactly two non-empty profile names are required")
 	}
+	if profiles[0] == profiles[1] {
+		return configDiffRequest{}, fmt.Errorf("config diff: profile names must differ")
+	}
 	if request.flags.campaignPath.set && request.flags.campaignFilePath.set {
 		return configDiffRequest{}, fmt.Errorf("config diff: --campaign and --campaign-file are mutually exclusive")
 	}
@@ -156,21 +166,22 @@ func splitConfigDiffArguments(args []string) (profiles, flagArgs []string) {
 
 func configDiffStringFlag(name string) bool {
 	switch name {
-	case "config", "campaign", "campaign-file", "profile":
+	case "config", "campaign", "campaign-file", "profile", "secrets-dir":
 		return true
 	default:
 		return false
 	}
 }
 
-func resolveConfigDiff(request configDiffRequest) (*configDiffResolved, error) {
-	pipelinePath, err := resolvePipelineConfigPath(request.flags.pipelinePath.value)
+func resolveConfigDiff(request configDiffRequest, deps commandDependencies) (*configDiffResolved, error) {
+	deps = normalizeCommandDependencies(deps)
+	pipelinePath, leftPipeline, err := loadPipelineConfigWithDependencies(request.flags.pipelinePath.value, config.PipelineLoadOptions{Profile: &request.leftProfile}, request.flags.secretsDir.pointer(), deps)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("load left pipeline profile: %w", err)
 	}
-	leftPipeline, rightPipeline, err := config.LoadPipelineProfilePair(pipelinePath, request.leftProfile, request.rightProfile)
+	_, rightPipeline, err := loadPipelineConfigWithDependencies(request.flags.pipelinePath.value, config.PipelineLoadOptions{Profile: &request.rightProfile}, request.flags.secretsDir.pointer(), deps)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("load right pipeline profile: %w", err)
 	}
 	needsCampaign := pipelineHasArtifactFamilies(leftPipeline) || pipelineHasArtifactFamilies(rightPipeline)
 	hasCampaignSelection := request.flags.campaignPath.set || request.flags.campaignFilePath.set

internal/app/config_loader.go

@@ -34,7 +34,7 @@ func retainedCommandConfig(cfg *config.Config) *commandConfig {
 	return &commandConfig{Config: cfg}
 }
 
-func loadCommandConfigWithDependencies(ctx context.Context, pipelineFlag, campaignFlag, campaignFileFlag, sessionFlag string, sessionOpts config.SessionLoadOptions, deps commandDependencies) (loaded *commandConfig, err error) {
+func loadCommandConfigWithDependencies(ctx context.Context, pipelineFlag, campaignFlag, campaignFileFlag, sessionFlag string, sessionOpts config.SessionLoadOptions, secretsDir *string, deps commandDependencies) (loaded *commandConfig, err error) {
 	deps = normalizeCommandDependencies(deps)
 	var cleanup func() error
 	defer func() {
@@ -46,7 +46,7 @@ func loadCommandConfigWithDependencies(ctx context.Context, pipelineFlag, campai
 		}
 	}()
 
-	base, err := loadPipelineCampaignConfigWithDependencies(pipelineFlag, campaignFlag, campaignFileFlag, config.PipelineLoadOptions{Profile: sessionOpts.Profile}, deps)
+	base, err := loadPipelineCampaignConfigWithDependencies(pipelineFlag, campaignFlag, campaignFileFlag, config.PipelineLoadOptions{Profile: sessionOpts.Profile}, secretsDir, deps)
 	if err != nil {
 		return nil, err
 	}
@@ -135,9 +135,9 @@ func loadCommandConfigWithDependencies(ctx context.Context, pipelineFlag, campai
 	return loaded, nil
 }
 
-func loadPipelineCampaignConfigWithDependencies(pipelineFlag, campaignFlag, campaignFileFlag string, pipelineOpts config.PipelineLoadOptions, deps commandDependencies) (*pipelineCampaignConfig, error) {
+func loadPipelineCampaignConfigWithDependencies(pipelineFlag, campaignFlag, campaignFileFlag string, pipelineOpts config.PipelineLoadOptions, secretsDir *string, deps commandDependencies) (*pipelineCampaignConfig, error) {
 	deps = normalizeCommandDependencies(deps)
-	loadedPipelinePath, pipelineCfg, err := loadPipelineConfigWithDependencies(pipelineFlag, pipelineOpts, deps)
+	loadedPipelinePath, pipelineCfg, err := loadPipelineConfigWithDependencies(pipelineFlag, pipelineOpts, secretsDir, deps)
 	if err != nil {
 		return nil, err
 	}
@@ -161,12 +161,19 @@ func loadPipelineCampaignConfigWithDependencies(pipelineFlag, campaignFlag, camp
 	return &loaded, nil
 }
 
-func loadPipelineConfigWithDependencies(pipelineFlag string, opts config.PipelineLoadOptions, deps commandDependencies) (string, *config.PipelineConfig, error) {
+func loadPipelineConfigWithDependencies(pipelineFlag string, opts config.PipelineLoadOptions, secretsDir *string, deps commandDependencies) (string, *config.PipelineConfig, error) {
 	deps = normalizeCommandDependencies(deps)
-	resolvedPipelinePath, err := resolvePipelineConfigPath(pipelineFlag)
+	resolvedPipelinePath, err := deps.resolvePipelineConfigPath(pipelineFlag)
 	if err != nil {
 		return "", nil, err
 	}
+	if !deps.secretsBootstrap.selectionSet {
+		deps.secretsBootstrap.selection = secretsDir
+		deps.secretsBootstrap.selectionSet = true
+	}
+	if err := deps.secretsBootstrap.load(resolvedPipelinePath, deps.loadSecretsDir, deps.diagnostics); err != nil {
+		return "", nil, err
+	}
 	pipelineCfg, err := deps.loadPipelineConfig(resolvedPipelinePath, opts)
 	if err != nil {
 		return "", nil, err

internal/app/config_loader_test.go

@@ -35,7 +35,7 @@ func TestLoadCommandConfigRetainsInitiallyLoadedPipeline(t *testing.T) {
 		},
 	}
 
-	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, sessionPath, config.SessionLoadOptions{SessionID: "2026-05-03"}, deps)
+	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, sessionPath, config.SessionLoadOptions{SessionID: "2026-05-03"}, nil, deps)
 	if err != nil {
 		t.Fatalf("loadCommandConfig() error = %v", err)
 	}

internal/app/extract_lifecycle_test.go

@@ -7,11 +7,13 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"reflect"
 	"sort"
 	"strings"
 	"testing"
 	"time"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/notarius"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactmodel"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifactpolicy"
@@ -181,12 +183,16 @@ func TestExtractLifecyclePrepareBindsVerifiedReferenceSnapshots(t *testing.T) {
 	for index, expected := range want {
 		binding := request.References[index]
 		canonical := filepath.Join(paths.InputsDir, expected.filename)
-		snapshot := filepath.Join(
-			artifacts.SessionRunNotariusReferencesDirForCampaign(
-				cfg.Pipeline.Workspace.Root, cfg.Session.Campaign, cfg.Session.SessionID, loaded.RunID,
-			),
-			expected.filename,
+		descriptor, ok := artifactpolicy.DescribePreparedInputSource(expected.sourceID)
+		if !ok {
+			t.Fatalf("DescribePreparedInputSource(%q) ok = false", expected.sourceID)
+		}
+		snapshot, err := artifacts.SessionNotariusReferenceSnapshotPath(
+			paths, lifecycleInputChecksum(t, loaded, descriptor.ManifestKind), expected.filename,
 		)
+		if err != nil {
+			t.Fatalf("SessionNotariusReferenceSnapshotPath(%q) error = %v", expected.sourceID, err)
+		}
 		if binding.Selector != expected.selector || binding.Path != snapshot || binding.Path == canonical || binding.Path == originalPaths[expected.sourceID] {
 			t.Fatalf("reference[%d] = %#v, want selector %q snapshot %q and not prepared/source paths", index, binding, expected.selector, snapshot)
 		}
@@ -393,6 +399,140 @@ func TestExtractLifecycleFailureInvalidatesAndOrdinaryRetryRerunsDownstream(t *t
 	}
 }
 
+func TestExtractLifecycleRetryReusesStableSnapshotsAndChangedReferenceSelectsNewPath(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	configureLifecycleReferences(t, cfg)
+	runner.failuresRemaining = 1
+	plan := prepareExtractLifecyclePlan(t)
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err == nil {
+		t.Fatal("first executeStages() error = nil, want extraction failure")
+	}
+	if len(runner.requests) != 1 {
+		t.Fatalf("failed requests = %d, want 1", len(runner.requests))
+	}
+	firstPaths := lifecycleReferencePaths(runner.requests[0])
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("retry executeStages() error = %v", err)
+	}
+	if len(runner.requests) != 2 {
+		t.Fatalf("retry requests = %d, want 2", len(runner.requests))
+	}
+	if secondPaths := lifecycleReferencePaths(runner.requests[1]); !reflect.DeepEqual(secondPaths, firstPaths) {
+		t.Fatalf("snapshot paths changed on retry:\nfirst %#v\nsecond %#v", firstPaths, secondPaths)
+	}
+
+	partyDocument, err := config.ParseParty([]byte(`schema_version: narratio.party.v1
+characters:
+  changed:
+    player: {name: Eric}
+    character:
+      name: Changed
+      classes: [{name: wizard}]
+`))
+	if err != nil {
+		t.Fatalf("ParseParty(changed party) error = %v", err)
+	}
+	cfg.Party.Canonical = partyDocument
+	persisted := loadLifecycleManifest(t, cfg)
+	prepare, err := stage.Select("prepare")
+	if err != nil {
+		t.Fatalf("stage.Select(prepare) error = %v", err)
+	}
+	if _, err := prepare.Run(context.Background(), env, persisted); err != nil {
+		t.Fatalf("prepare.Run() error = %v", err)
+	}
+	if err := (&manifest.LocalStore{}).Save(context.Background(), manifestPathFor(cfg), persisted); err != nil {
+		t.Fatalf("Save(reprepared manifest) error = %v", err)
+	}
+	extractPlan, _ := BuildSingleStagePlan("extract")
+	if _, err := executeStages(context.Background(), cfg, extractPlan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("changed-reference executeStages() error = %v", err)
+	}
+	if len(runner.requests) != 3 {
+		t.Fatalf("changed-reference requests = %d, want 3", len(runner.requests))
+	}
+	thirdPaths := lifecycleReferencePaths(runner.requests[2])
+	if thirdPaths["party"] == firstPaths["party"] {
+		t.Fatalf("changed party retained snapshot path %q", thirdPaths["party"])
+	}
+	if thirdPaths["players"] == firstPaths["players"] {
+		t.Fatalf("changed party retained derived players snapshot path %q", thirdPaths["players"])
+	}
+	for selector, firstPath := range firstPaths {
+		if selector != "party" && selector != "players" && thirdPaths[selector] != firstPath {
+			t.Fatalf("unchanged %s snapshot path = %q, want %q", selector, thirdPaths[selector], firstPath)
+		}
+		if _, err := os.Stat(firstPath); err != nil {
+			t.Fatalf("old %s snapshot %q was not retained: %v", selector, firstPath, err)
+		}
+	}
+}
+
+func TestExtractLifecycleCacheControlsAndSnapshotLossDoNotInvalidateCompleteResult(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	configureLifecycleReferences(t, cfg)
+	if _, err := executeStages(context.Background(), cfg, prepareExtractLifecyclePlan(t), RunOptions{Env: env}); err != nil {
+		t.Fatalf("initial executeStages() error = %v", err)
+	}
+	paths := artifacts.NewLocalStore(cfg.Pipeline.Workspace.Root).SessionPathsFor(cfg.Session.Campaign, cfg.Session.SessionID)
+	cacheRoot := artifacts.SessionNotariusReferenceCacheDir(paths)
+	if err := os.RemoveAll(cacheRoot); err != nil {
+		t.Fatalf("RemoveAll(snapshot cache) error = %v", err)
+	}
+	cfg.Pipeline.Notarius.Resume = true
+	cfg.Pipeline.Notarius.ChunkCache = "bypass"
+	extractPlan, _ := BuildSingleStagePlan("extract")
+	summary, err := executeStages(context.Background(), cfg, extractPlan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("resume executeStages() error = %v", err)
+	}
+	if len(summary.Executed) != 0 || !reflect.DeepEqual(summary.Skipped, []string{"extract"}) || len(runner.requests) != 1 {
+		t.Fatalf("resume summary = %#v requests=%d", summary, len(runner.requests))
+	}
+	if _, err := os.Stat(cacheRoot); !errors.Is(err, os.ErrNotExist) {
+		t.Fatalf("skipped extraction recreated snapshot cache: %v", err)
+	}
+}
+
+func TestAnalyzeOnlyLifecyclePreservesExtractionWithoutInvokingNotarius(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	extractPlan, _ := BuildSingleStagePlan("extract")
+	if _, err := executeStages(context.Background(), cfg, extractPlan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("extract executeStages() error = %v", err)
+	}
+	before := loadLifecycleManifest(t, cfg).Stages["extract"]
+	beforeJSON, _ := json.Marshal(before)
+	analyzeRuns := 0
+	if _, err := executeStages(context.Background(), cfg, []stage.Stage{countingStage{name: "analyze", runs: &analyzeRuns}}, RunOptions{Env: env, Force: true}); err != nil {
+		t.Fatalf("analyze-only executeStages() error = %v", err)
+	}
+	after := loadLifecycleManifest(t, cfg).Stages["extract"]
+	afterJSON, _ := json.Marshal(after)
+	if analyzeRuns != 1 || len(runner.requests) != 1 || string(afterJSON) != string(beforeJSON) {
+		t.Fatalf("analyze-only lifecycle changed extraction: analyze=%d requests=%d\nbefore=%s\nafter=%s", analyzeRuns, len(runner.requests), beforeJSON, afterJSON)
+	}
+	for _, output := range after.Outputs {
+		if _, err := os.Stat(output.LocalPath); err != nil {
+			t.Fatalf("analyze-only lifecycle removed extract output %q: %v", output.LocalPath, err)
+		}
+	}
+}
+
+func TestForcedArtifactRegenerationUsesOrdinaryExtractAndAnalyzePath(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	cfg.Pipeline.Notarius.Resume = true
+	cfg.Pipeline.Notarius.ChunkCache = "bypass"
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env, Force: true, SelectedArtifacts: []string{"session_recap"}}); err != nil {
+		t.Fatalf("forced artifact regeneration error = %v", err)
+	}
+	if len(runner.requests) != 1 || runner.requests[0].Resume || runner.requests[0].ChunkCache != "refresh" || analyzeRuns != 1 {
+		t.Fatalf("forced path request=%#v analyze=%d", runner.requests, analyzeRuns)
+	}
+}
+
 func TestExtractLifecycleForcedSelfSkipInvalidatesDownstream(t *testing.T) {
 	cfg, env, _ := extractionLifecycleFixture(t, false)
 	markLifecycleStageSucceeded(t, cfg, "analyze")
@@ -485,6 +625,273 @@ func TestExtractLifecycleSkipsCurrentResumableResult(t *testing.T) {
 	}
 }
 
+func TestExtractLifecycleCredentialChangesRemainResumable(t *testing.T) {
+	for _, test := range []struct {
+		name   string
+		before func(*testing.T) *configkit.Secret
+		after  func(*testing.T) *configkit.Secret
+	}{
+		{
+			name: "added",
+			after: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_ADDED_KEY", "added-value")
+			},
+		},
+		{
+			name: "removed",
+			before: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_REMOVED_KEY", "removed-value")
+			},
+		},
+		{
+			name: "renamed",
+			before: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_FIRST_KEY", "first-value")
+			},
+			after: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_SECOND_KEY", "second-value")
+			},
+		},
+		{
+			name: "value rotated",
+			before: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_ROTATED_KEY", "first-value")
+			},
+			after: func(t *testing.T) *configkit.Secret {
+				return loadLifecycleNotariusCredential(t, "NARRATIO_LIFECYCLE_ROTATED_KEY", "second-value")
+			},
+		},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			cfg, env, runner := extractionLifecycleFixture(t, true)
+			if test.before != nil {
+				cfg.Pipeline.Notarius.LLMAPIKey = test.before(t)
+			}
+			analyzeRuns := 0
+			plan := extractionLifecyclePlan(t, &analyzeRuns)
+			if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+				t.Fatalf("initial executeStages() error = %v", err)
+			}
+			cfg.Pipeline.Notarius.LLMAPIKey = nil
+			if test.after != nil {
+				cfg.Pipeline.Notarius.LLMAPIKey = test.after(t)
+			}
+			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("credential-only change reexecuted work: summary=%#v requests=%d analyze=%d", resumed, len(runner.requests), analyzeRuns)
+			}
+			loaded := loadLifecycleManifest(t, cfg)
+			if loaded.Stages["extract"].Status != manifest.StatusSucceeded || loaded.Stages["analyze"].Status != manifest.StatusSucceeded {
+				t.Fatalf("credential-only change invalidated stages: extract=%q analyze=%q", loaded.Stages["extract"].Status, loaded.Stages["analyze"].Status)
+			}
+		})
+	}
+}
+
+func TestExtractLifecycleConfigProfileChangesRerunExtractionAndDependents(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	selected := "inexpensive"
+	cfg.Pipeline.Notarius.ConfigProfile = &selected
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("initial executeStages() error = %v", err)
+	}
+	if len(runner.requests) != 1 || runner.requests[0].ConfigProfile == nil || *runner.requests[0].ConfigProfile != selected || analyzeRuns != 1 {
+		t.Fatalf("initial requests=%#v analyze=%d, want selected extraction and analysis", runner.requests, analyzeRuns)
+	}
+	reused, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("reuse executeStages() error = %v", err)
+	}
+	if len(reused.Executed) != 0 || len(runner.requests) != 1 || analyzeRuns != 1 {
+		t.Fatalf("same-profile reuse summary=%#v requests=%d analyze=%d", reused, len(runner.requests), analyzeRuns)
+	}
+
+	markLifecycleStageSucceeded(t, cfg, "render")
+	markLifecycleStageSucceeded(t, cfg, "publish")
+	before := loadLifecycleManifest(t, cfg)
+	trimBefore, err := json.Marshal(before.Stages["trim"])
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	changed := "experimental"
+	cfg.Pipeline.Notarius.ConfigProfile = &changed
+	rerun, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("changed-profile executeStages() error = %v", err)
+	}
+	if len(rerun.Executed) != 2 || len(runner.requests) != 2 || runner.requests[1].ConfigProfile == nil || *runner.requests[1].ConfigProfile != changed || analyzeRuns != 2 {
+		t.Fatalf("changed-profile summary=%#v requests=%#v analyze=%d", rerun, runner.requests, analyzeRuns)
+	}
+	after := loadLifecycleManifest(t, cfg)
+	trimAfter, err := json.Marshal(after.Stages["trim"])
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(trimAfter) != string(trimBefore) {
+		t.Fatalf("profile change altered upstream trim record:\nbefore=%s\nafter=%s", trimBefore, trimAfter)
+	}
+	if after.Stages["render"].Status != manifest.StatusSucceeded || after.Stages["analyze"].Status != manifest.StatusSucceeded || after.Stages["publish"].Status != manifest.StatusStale {
+		t.Fatalf("changed-profile lifecycle render=%#v analyze=%#v publish=%#v", after.Stages["render"], after.Stages["analyze"], after.Stages["publish"])
+	}
+
+	cfg.Pipeline.Notarius.ConfigProfile = nil
+	returned, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("return-to-base executeStages() error = %v", err)
+	}
+	if len(returned.Executed) != 2 || len(runner.requests) != 3 || runner.requests[2].ConfigProfile != nil || analyzeRuns != 3 {
+		t.Fatalf("return-to-base summary=%#v requests=%#v analyze=%d", returned, runner.requests, analyzeRuns)
+	}
+	final := loadLifecycleManifest(t, cfg)
+	if final.Stages["trim"].Status != manifest.StatusSucceeded || final.Stages["render"].Status != manifest.StatusSucceeded {
+		t.Fatalf("return-to-base invalidated upstream siblings: trim=%#v render=%#v", final.Stages["trim"], final.Stages["render"])
+	}
+}
+
+func TestExtractLifecycleConfigProfileFailureClearsCurrentResultAndDependents(t *testing.T) {
+	cfg, env, runner := extractionLifecycleFixture(t, true)
+	selected := "inexpensive"
+	cfg.Pipeline.Notarius.ConfigProfile = &selected
+	analyzeRuns := 0
+	plan := extractionLifecyclePlan(t, &analyzeRuns)
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("initial executeStages() error = %v", err)
+	}
+	markLifecycleStageSucceeded(t, cfg, "render")
+	markLifecycleStageSucceeded(t, cfg, "publish")
+
+	changed := "missing-profile"
+	cfg.Pipeline.Notarius.ConfigProfile = &changed
+	runner.failuresRemaining = 1
+	if _, err := executeStages(context.Background(), cfg, plan, RunOptions{Env: env}); err == nil {
+		t.Fatal("changed-profile executeStages() error = nil, want downstream failure")
+	}
+	failed := loadLifecycleManifest(t, cfg)
+	extract := failed.Stages["extract"]
+	if extract.Status != manifest.StatusFailed || len(extract.Outputs) != 0 || len(extract.Logs) != 0 || len(extract.GeneratedConfigs) != 0 || len(extract.Metadata) != 0 {
+		t.Fatalf("failed changed-profile extraction advertised prior success: %#v", extract)
+	}
+	if failed.Stages["trim"].Status != manifest.StatusSucceeded || failed.Stages["render"].Status != manifest.StatusSucceeded ||
+		failed.Stages["analyze"].Status != manifest.StatusStale || failed.Stages["publish"].Status != manifest.StatusStale || analyzeRuns != 1 {
+		t.Fatalf("failed changed-profile lifecycle trim=%#v render=%#v analyze=%#v publish=%#v analyzeRuns=%d",
+			failed.Stages["trim"], failed.Stages["render"], failed.Stages["analyze"], failed.Stages["publish"], analyzeRuns)
+	}
+	if len(runner.requests) != 2 || runner.requests[1].ConfigProfile == nil || *runner.requests[1].ConfigProfile != changed {
+		t.Fatalf("failed request = %#v, want changed selector", runner.requests)
+	}
+}
+
+func TestExtractLifecycleRealConfigProfilesSelectNotariusExplicitly(t *testing.T) {
+	seed, env, runner := extractionLifecycleFixture(t, true)
+	workspace := seed.Pipeline.Workspace.Root
+	pipelinePath, campaignPath, sessionPath := writeValidConfigFiles(t, workspace)
+	pipelineFile, err := os.OpenFile(pipelinePath, os.O_APPEND|os.O_WRONLY, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	_, writeErr := pipelineFile.WriteString(`notarius:
+  enabled: true
+  binary: ` + filepath.Join(workspace, "notarius") + `
+  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
+`)
+	closeErr := pipelineFile.Close()
+	if writeErr != nil || closeErr != nil {
+		t.Fatalf("write Notarius configuration: %v / %v", writeErr, closeErr)
+	}
+	if err := os.WriteFile(filepath.Join(filepath.Dir(pipelinePath), "notarius.yml"), []byte("pipelines: {}\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	profilesDir := filepath.Join(filepath.Dir(pipelinePath), "profiles")
+	if err := os.MkdirAll(profilesDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	for name, contents := range map[string]string{
+		"testing":           "whisperx:\n  language: fr\nnotarius:\n  config_profile: downstream-economy\n",
+		"narratio-only":     "whisperx:\n  language: de\n",
+		"narratio-only-two": "whisperx:\n  language: es\n",
+	} {
+		if err := os.WriteFile(filepath.Join(profilesDir, name+".yml"), []byte(contents), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	load := func(profile string) *config.Config {
+		t.Helper()
+		loaded, err := config.LoadWithSessionOptions(pipelinePath, campaignPath, sessionPath, config.SessionLoadOptions{Profile: &profile})
+		if err != nil {
+			t.Fatalf("load profile %q: %v", profile, err)
+		}
+		return loaded
+	}
+	plan, err := BuildSingleStagePlan("extract")
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	testingConfig := load("testing")
+	runner.cfg = testingConfig.Pipeline.Notarius
+	if _, err := executeStages(context.Background(), testingConfig, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("execute testing profile: %v", err)
+	}
+	if len(runner.requests) != 1 || runner.requests[0].ConfigProfile == nil || *runner.requests[0].ConfigProfile != "downstream-economy" {
+		t.Fatalf("testing profile request = %#v, want explicit differently named downstream selector", runner.requests)
+	}
+
+	narratioOnly := load("narratio-only")
+	runner.cfg = narratioOnly.Pipeline.Notarius
+	if _, err := executeStages(context.Background(), narratioOnly, plan, RunOptions{Env: env}); err != nil {
+		t.Fatalf("execute Narratio-only profile: %v", err)
+	}
+	if len(runner.requests) != 2 || runner.requests[1].ConfigProfile != nil {
+		t.Fatalf("Narratio-only profile request = %#v, want no automatic downstream selector", runner.requests)
+	}
+
+	narratioOnlyTwo := load("narratio-only-two")
+	runner.cfg = narratioOnlyTwo.Pipeline.Notarius
+	reused, err := executeStages(context.Background(), narratioOnlyTwo, plan, RunOptions{Env: env})
+	if err != nil {
+		t.Fatalf("execute second Narratio-only profile: %v", err)
+	}
+	if len(reused.Executed) != 0 || len(runner.requests) != 2 {
+		t.Fatalf("Narratio profile label/effective digest broadened invalidation: summary=%#v requests=%#v", reused, runner.requests)
+	}
+}
+
+func TestPreparedPipelineAndManifestExcludeNotariusCredentialValue(t *testing.T) {
+	const name = "NARRATIO_LIFECYCLE_PERSISTED_KEY"
+	const canary = "notarius-persisted-credential-canary"
+	cfg, env, _ := extractionLifecycleFixture(t, true)
+	cfg.Pipeline.Notarius.LLMAPIKey = loadLifecycleNotariusCredential(t, name, canary)
+	if _, err := executeStages(context.Background(), cfg, prepareExtractLifecyclePlan(t), RunOptions{Env: env}); err != nil {
+		t.Fatalf("executeStages() error = %v", err)
+	}
+	paths := artifacts.NewLocalStore(cfg.Pipeline.Workspace.Root).SessionPathsFor(cfg.Session.Campaign, cfg.Session.SessionID)
+	for _, path := range []string{paths.ManifestPath, filepath.Join(paths.InputsDir, "pipeline.resolved.yml")} {
+		data, err := os.ReadFile(path)
+		if err != nil {
+			t.Fatalf("ReadFile(%q) error = %v", path, err)
+		}
+		if strings.Contains(string(data), canary) {
+			t.Fatalf("persisted Narratio file %q contains the credential value", path)
+		}
+	}
+}
+
 func TestExtractLifecycleRerunsAfterDirectTranscriptChange(t *testing.T) {
 	cfg, env, runner := extractionLifecycleFixture(t, true)
 	analyzeRuns := 0
@@ -696,6 +1103,14 @@ func lifecycleInputChecksum(t *testing.T, m *manifest.Manifest, kind string) str
 	return ""
 }
 
+func lifecycleReferencePaths(request notarius.RunRequest) map[string]string {
+	paths := make(map[string]string, len(request.References))
+	for _, reference := range request.References {
+		paths[reference.Selector] = reference.Path
+	}
+	return paths
+}
+
 func lifecycleDndOutputs() map[string]config.NotariusOutputConfig {
 	return map[string]config.NotariusOutputConfig{
 		"item_registry":        {LaneID: "item-registry", MediaType: "application/json", SchemaID: "notarius.dnd.item_registry", SchemaVersion: "v1", ModuleKey: "dnd/item-registry"},
@@ -761,3 +1176,17 @@ func extractionLifecycleFixture(t *testing.T, enabled bool) (*config.Config, *st
 	runner := &materializingNotariusRunner{cfg: cfg.Pipeline.Notarius}
 	return cfg, &stage.Env{Notarius: runner}, runner
 }
+
+func loadLifecycleNotariusCredential(t *testing.T, name, value string) *configkit.Secret {
+	t.Helper()
+	t.Setenv(name, value)
+	path := filepath.Join(t.TempDir(), "pipeline.yml")
+	if err := os.WriteFile(path, []byte("whisperx:\n  transcribe_url: https://transcription.example.test/transcribe\nnotarius:\n  enabled: false\n  llm_api_key: "+name+"\n"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	pipeline, err := config.LoadPipeline(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return pipeline.Notarius.LLMAPIKey
+}

internal/app/object_store.go

@@ -11,9 +11,6 @@ import (
 
 func newCommandObjectStoreWithDependencies(ctx context.Context, cfg *config.Config, logger *slog.Logger, deps commandDependencies) (storage.ObjectStore, error) {
 	deps = normalizeCommandDependencies(deps)
-	if _, err := loadSecretsFromConfigWithDependencies(cfg, logger, deps); err != nil {
-		return nil, fmt.Errorf("load secrets from files: %w", err)
-	}
 	return initializeObjectStoreBackend(ctx, cfg, deps.newObjectStore)
 }
 

internal/app/object_store_test.go

@@ -3,8 +3,6 @@ package app
 import (
 	"context"
 	"errors"
-	"os"
-	"path/filepath"
 	"strings"
 	"testing"
 
@@ -12,89 +10,8 @@ import (
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
-func TestNewCommandObjectStoreLoadsSecretsBeforeFactory(t *testing.T) {
-	accessKeyEnv := "NARRATIO_TEST_COMMAND_STORE_KEY_ID"
-	secretKeyEnv := "NARRATIO_TEST_COMMAND_STORE_SECRET"
-	restoreEnvAfterTest(t, accessKeyEnv, secretKeyEnv)
-
-	secretsDir := t.TempDir()
-	mustWriteSecretFile(t, filepath.Join(secretsDir, accessKeyEnv), "loaded-key-id\n")
-	mustWriteSecretFile(t, filepath.Join(secretsDir, secretKeyEnv), "loaded-secret\n")
-
-	cfg := commandObjectStoreTestConfig(secretsDir)
-	fake := &storage.FakeBackend{}
-	called := false
-	deps := commandDependencies{newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
-		called = true
-		if got := os.Getenv(accessKeyEnv); got != "loaded-key-id" {
-			return nil, errors.New("access key was not loaded before object store init")
-		}
-		if got := os.Getenv(secretKeyEnv); got != "loaded-secret" {
-			return nil, errors.New("secret key was not loaded before object store init")
-		}
-		return fake, nil
-	}}
-
-	store, err := newCommandObjectStoreWithDependencies(context.Background(), cfg, nil, deps)
-	if err != nil {
-		t.Fatalf("newCommandObjectStore() error = %v", err)
-	}
-	if store != fake {
-		t.Fatalf("store = %#v, want fake backend", store)
-	}
-	if !called {
-		t.Fatal("object store factory was not called")
-	}
-}
-
-func TestNewCommandObjectStorePreservesExistingEnv(t *testing.T) {
-	accessKeyEnv := "NARRATIO_TEST_COMMAND_STORE_EXISTING_KEY_ID"
-	secretKeyEnv := "NARRATIO_TEST_COMMAND_STORE_EXISTING_SECRET"
-	t.Setenv(accessKeyEnv, "existing-key-id")
-	t.Setenv(secretKeyEnv, "existing-secret")
-
-	secretsDir := t.TempDir()
-	mustWriteSecretFile(t, filepath.Join(secretsDir, accessKeyEnv), "file-key-id\n")
-	mustWriteSecretFile(t, filepath.Join(secretsDir, secretKeyEnv), "file-secret\n")
-
-	cfg := commandObjectStoreTestConfig(secretsDir)
-	deps := commandDependencies{newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
-		if got := os.Getenv(accessKeyEnv); got != "existing-key-id" {
-			return nil, errors.New("existing access key was overwritten")
-		}
-		if got := os.Getenv(secretKeyEnv); got != "existing-secret" {
-			return nil, errors.New("existing secret key was overwritten")
-		}
-		return &storage.FakeBackend{}, nil
-	}}
-
-	if _, err := newCommandObjectStoreWithDependencies(context.Background(), cfg, nil, deps); err != nil {
-		t.Fatalf("newCommandObjectStore() error = %v", err)
-	}
-}
-
-func TestNewCommandObjectStoreSecretErrorStopsFactory(t *testing.T) {
-	cfg := commandObjectStoreTestConfig(filepath.Join(t.TempDir(), "missing"))
-	called := false
-	deps := commandDependencies{newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
-		called = true
-		return &storage.FakeBackend{}, nil
-	}}
-
-	_, err := newCommandObjectStoreWithDependencies(context.Background(), cfg, nil, deps)
-	if err == nil {
-		t.Fatal("expected error, got nil")
-	}
-	if called {
-		t.Fatal("object store factory was called after secret load failure")
-	}
-	if !strings.Contains(err.Error(), "load secrets from files") {
-		t.Fatalf("error = %q, want secret loading context", err.Error())
-	}
-}
-
 func TestNewCommandObjectStoreFactoryErrorIsContextual(t *testing.T) {
-	cfg := commandObjectStoreTestConfig("")
+	cfg := commandObjectStoreTestConfig()
 	deps := commandDependencies{newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
 		return nil, errors.New("factory boom")
 	}}
@@ -108,21 +25,17 @@ func TestNewCommandObjectStoreFactoryErrorIsContextual(t *testing.T) {
 	}
 }
 
-func commandObjectStoreTestConfig(secretsDir string) *config.Config {
-	cfg := &config.Config{
+func commandObjectStoreTestConfig() *config.Config {
+	return &config.Config{
 		Pipeline: &config.PipelineConfig{
 			Storage: config.StorageConfig{
 				Backend: "s3",
 				S3: &config.StorageS3Config{
-					Bucket:         "test-bucket",
-					AccessKeyIDEnv: "NARRATIO_TEST_COMMAND_STORE_KEY_ID",
-					SecretKeyEnv:   "NARRATIO_TEST_COMMAND_STORE_SECRET",
+					Bucket:          "test-bucket",
+					AccessKeyID:     testSecretReference("NARRATIO_TEST_COMMAND_STORE_KEY_ID"),
+					SecretAccessKey: testSecretReference("NARRATIO_TEST_COMMAND_STORE_SECRET"),
 				},
 			},
 		},
 	}
-	if strings.TrimSpace(secretsDir) != "" {
-		cfg.Pipeline.Secrets = &config.SecretsConfig{EnvDir: secretsDir}
-	}
-	return cfg
 }

internal/app/operator_helpers.go

@@ -24,6 +24,7 @@ type commonConfigFlags struct {
 	sessionID         string
 	previousSessionID string
 	profile           singletonStringFlag
+	secretsDir        singletonStringFlag
 }
 
 func addCommonConfigFlags(fs *flag.FlagSet, flags *commonConfigFlags) {
@@ -35,6 +36,8 @@ func addCommonConfigFlags(fs *flag.FlagSet, flags *commonConfigFlags) {
 	fs.StringVar(&flags.previousSessionID, "previous-session-id", "", "expected previous session identifier")
 	flags.profile.name = "profile"
 	fs.Var(&flags.profile, "profile", "named pipeline profile")
+	flags.secretsDir.name = "secrets-dir"
+	fs.Var(&flags.secretsDir, "secrets-dir", "secrets directory (optional; defaults beside pipeline.yml)")
 }
 
 func (f commonConfigFlags) sessionOptions() config.SessionLoadOptions {
@@ -112,7 +115,7 @@ func artifactsWithDependencies(ctx context.Context, args []string, out io.Writer
 
 func loadHelperContextWithDependencies(ctx context.Context, flags commonConfigFlags, needStore bool, deps commandDependencies) (*config.Config, storage.ObjectStore, []artifacts.RemoteLock, *manifest.Manifest, func(), error) {
 	deps = normalizeCommandDependencies(deps)
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return nil, nil, nil, nil, nil, err
 	}

internal/app/operator_helpers_test.go

@@ -164,19 +164,19 @@ func TestExecuteSessionInitRequiresSessionID(t *testing.T) {
 }
 
 func TestExecuteSessionInitMissingDefaultConfigReportsSearchedPaths(t *testing.T) {
-	origPipelineDefaults := append([]string(nil), config.DefaultPipelineConfigSearchPaths...)
-	config.DefaultPipelineConfigSearchPaths = []string{filepath.Join(t.TempDir(), "missing-pipeline.yml")}
-	t.Cleanup(func() {
-		config.DefaultPipelineConfigSearchPaths = origPipelineDefaults
-	})
+	deps, missingPath, executor := guardedMissingPipelineDiscovery(t)
 
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
-	code := Execute([]string{"session", "init", "2026-06-07", "--remote"}, &stdout, &stderr)
+	code := executeWithDependencies([]string{"session", "init", "2026-06-07", "--remote"}, &stdout, &stderr, deps, executor)
 	if code == 0 {
 		t.Fatal("exit code = 0, want non-zero")
 	}
-	if !strings.Contains(stderr.String(), "session init: no pipeline config path provided and no default pipeline config found; searched:") {
+	if stdout.Len() != 0 {
+		t.Fatalf("stdout = %q, want empty", stdout.String())
+	}
+	if got := stderr.String(); !strings.Contains(got, "session init: no pipeline config path provided") ||
+		!strings.Contains(got, "no default pipeline config found") || !strings.Contains(got, missingPath) {
 		t.Fatalf("stderr = %q, want default pipeline searched-path error", stderr.String())
 	}
 }

internal/app/operator_locks.go

@@ -131,7 +131,7 @@ func locksRemoveWithDependencies(ctx context.Context, args []string, out io.Writ
 
 func loadLockCommandContextWithDependencies(ctx context.Context, flags commonConfigFlags, deps commandDependencies) (*config.Config, storage.ObjectStore, string, func(), error) {
 	deps = normalizeCommandDependencies(deps)
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return nil, nil, "", nil, err
 	}

internal/app/operator_session_init.go

@@ -28,6 +28,7 @@ func sessionInitWithDependencies(ctx context.Context, args []string, out io.Writ
 	fs.SetOutput(io.Discard)
 	var pipelinePath, campaignPath, campaignFilePath, sessionID, previousSessionID, date, title, output, audioS3Prefix, audioDir string
 	var profile singletonStringFlag
+	var secretsDir singletonStringFlag
 	var remote, force bool
 	fs.StringVar(&pipelinePath, "config", "", "path to pipeline.yml (optional; defaults searched)")
 	fs.StringVar(&campaignPath, "campaign", "", "campaign ID")
@@ -41,6 +42,8 @@ func sessionInitWithDependencies(ctx context.Context, args []string, out io.Writ
 	fs.StringVar(&audioDir, "audio-dir", "", "local audio directory")
 	profile.name = "profile"
 	fs.Var(&profile, "profile", "named pipeline profile")
+	secretsDir.name = "secrets-dir"
+	fs.Var(&secretsDir, "secrets-dir", "secrets directory (optional; defaults beside pipeline.yml)")
 	fs.BoolVar(&remote, "remote", false, "write session.yml to S3 session prefix")
 	fs.BoolVar(&force, "force", false, "overwrite existing target")
 	if err := parseSessionAwareFlags("session init", fs, args, &sessionID); err != nil {
@@ -56,7 +59,7 @@ func sessionInitWithDependencies(ctx context.Context, args []string, out io.Writ
 		return fmt.Errorf("session init: --audio-dir and --audio-s3-prefix are mutually exclusive")
 	}
 
-	base, err := loadPipelineCampaignConfigWithDependencies(pipelinePath, campaignPath, campaignFilePath, config.PipelineLoadOptions{Profile: profile.pointer()}, deps)
+	base, err := loadPipelineCampaignConfigWithDependencies(pipelinePath, campaignPath, campaignFilePath, config.PipelineLoadOptions{Profile: profile.pointer()}, secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("session init: %w", err)
 	}

internal/app/operator_session_validate.go

@@ -30,7 +30,7 @@ func sessionValidateWithDependencies(ctx context.Context, args []string, out io.
 	}
 
 	findings := []finding{}
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		findings = append(findings, errorFinding("config", err.Error()))
 		return renderFindings(out, "", "", findings)

internal/app/operator_status.go

@@ -31,7 +31,7 @@ func statusWithDependencies(ctx context.Context, args []string, out io.Writer, d
 	if strings.TrimSpace(flags.sessionID) == "" {
 		return fmt.Errorf("status: session_id is required")
 	}
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("status: %w", err)
 	}

internal/app/pipeline_config_path_test.go

@@ -1,10 +1,16 @@
 package app
 
 import (
+	"bytes"
+	"context"
+	"errors"
 	"os"
 	"path/filepath"
 	"strings"
 	"testing"
+
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
 )
 
 func TestResolvePipelineConfigPathWithCandidatesExplicitWins(t *testing.T) {
@@ -55,11 +61,57 @@ func TestResolvePipelineConfigPathWithCandidatesPrecedence(t *testing.T) {
 }
 
 func TestResolvePipelineConfigPathWithCandidatesMissing(t *testing.T) {
-	_, err := resolvePipelineConfigPathWithCandidates("", []string{"/does/not/exist/one.yml", "/does/not/exist/two.yml"})
+	directory := t.TempDir()
+	first := filepath.Join(directory, "one.yml")
+	second := filepath.Join(directory, "two.yml")
+	_, err := resolvePipelineConfigPathWithCandidates("", []string{first, second})
 	if err == nil {
 		t.Fatal("expected error, got nil")
 	}
 	if !strings.Contains(err.Error(), "no default pipeline config found") {
 		t.Fatalf("error = %q, want missing-defaults context", err.Error())
 	}
+	if !strings.Contains(err.Error(), first) || !strings.Contains(err.Error(), second) {
+		t.Fatalf("error = %q, want controlled searched paths", err.Error())
+	}
+}
+
+func TestExecuteMissingExplicitConfigDiscoversExistingDefaultBeforeLoading(t *testing.T) {
+	directory := t.TempDir()
+	missing := filepath.Join(directory, "missing.yml")
+	existing := filepath.Join(directory, "pipeline.yml")
+	if err := os.WriteFile(existing, []byte("workspace:\n  root: ./tmp\n"), 0o644); err != nil {
+		t.Fatalf("write default pipeline: %v", err)
+	}
+
+	controlledErr := errors.New("controlled pipeline loader boundary")
+	loadedPath := ""
+	deps := commandDependenciesForPipelineCandidates(missing, existing)
+	deps.loadPipelineConfig = func(path string, _ config.PipelineLoadOptions) (*config.PipelineConfig, error) {
+		loadedPath = path
+		return nil, controlledErr
+	}
+	deps.loadSecretsDir = func(string) (*configkit.SecretsReport, error) {
+		t.Fatal("absent conventional secrets directory unexpectedly invoked bootstrap loader")
+		return nil, nil
+	}
+
+	var stdout bytes.Buffer
+	var stderr bytes.Buffer
+	code := executeWithDependencies([]string{"config", "validate"}, &stdout, &stderr, deps, func(context.Context, *config.Config, BoundedPlan, RunOptions) (*RunSummary, error) {
+		t.Fatal("config validation unexpectedly invoked pipeline executor")
+		return nil, nil
+	})
+	if code == 0 {
+		t.Fatal("exit code = 0, want controlled loader failure")
+	}
+	if loadedPath != existing {
+		t.Fatalf("loader path = %q, want discovered default %q", loadedPath, existing)
+	}
+	if stdout.Len() != 0 {
+		t.Fatalf("stdout = %q, want empty", stdout.String())
+	}
+	if !strings.Contains(stderr.String(), controlledErr.Error()) {
+		t.Fatalf("stderr = %q, want controlled loader error", stderr.String())
+	}
 }

internal/app/plan.go

@@ -31,7 +31,7 @@ func planWithDependencies(ctx context.Context, args []string, out io.Writer, dep
 		return err
 	}
 	flags := request.Config
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("plan: %w", err)
 	}

internal/app/plan_test.go

@@ -88,7 +88,7 @@ func TestPlanShowsRunAndSkipFromManifest(t *testing.T) {
 	}
 }
 
-func TestPlanDoesNotLoadConfiguredSecrets(t *testing.T) {
+func TestPlanSkipsAbsentConventionalSecretsDirectory(t *testing.T) {
 	workspaceRoot := t.TempDir()
 	configDir := t.TempDir()
 	pipelinePath := filepath.Join(configDir, "pipeline.yml")
@@ -99,8 +99,6 @@ func TestPlanDoesNotLoadConfiguredSecrets(t *testing.T) {
   root: ` + workspaceRoot + `
 storage:
   backend: local
-secrets:
-  env_dir: ./missing-secrets
 whisperx:
   transcribe_url: https://example.com/transcribe
 seriatim:
@@ -128,7 +126,7 @@ inputs:
 	var out bytes.Buffer
 	err := Plan(context.Background(), []string{"2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath}, &out)
 	if err != nil {
-		t.Fatalf("Plan() error = %v, want missing runtime secrets ignored", err)
+		t.Fatalf("Plan() error = %v, want absent conventional secrets directory skipped", err)
 	}
 }
 

internal/app/profile_commands_test.go

@@ -37,10 +37,10 @@ func TestRunProfileSelectionFlowsThroughSharedLoader(t *testing.T) {
 		t.Fatal(err)
 	}
 	if language != "en" {
-		t.Fatalf("default profile language = %q, want en", language)
+		t.Fatalf("base configuration language = %q, want en", language)
 	}
 
-	for _, profile := range []string{"unknown", ""} {
+	for _, profile := range []string{"unknown"} {
 		err := runWithPipelineExecutor(context.Background(), []string{
 			"2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath, "--profile", profile,
 		}, io.Discard, executor)
@@ -48,19 +48,20 @@ func TestRunProfileSelectionFlowsThroughSharedLoader(t *testing.T) {
 			t.Fatalf("profile %q error = %v, want selection rejection", profile, err)
 		}
 	}
+
+	if err := runWithPipelineExecutor(context.Background(), []string{
+		"2026-05-03", "--config", pipelinePath, "--campaign-file", campaignPath, "--session", sessionPath, "--profile", "",
+	}, io.Discard, executor); err != nil {
+		t.Fatalf("explicit empty profile should select base: %v", err)
+	}
 }
 
 func enableCommandTestProfiles(t *testing.T, pipelinePath string) {
 	t.Helper()
-	data, err := os.ReadFile(pipelinePath)
-	if err != nil {
+	dir := filepath.Join(filepath.Dir(pipelinePath), "profiles")
+	if err := os.MkdirAll(dir, 0o755); err != nil {
 		t.Fatal(err)
 	}
-	composition := "composition:\n  default_profile: production\n  profiles:\n    production:\n      overlay: production.yml\n    testing:\n      overlay: testing.yml\n"
-	if err := os.WriteFile(pipelinePath, append([]byte(composition), data...), 0o644); err != nil {
-		t.Fatal(err)
-	}
-	dir := filepath.Dir(pipelinePath)
 	if err := os.WriteFile(filepath.Join(dir, "production.yml"), []byte("whisperx:\n  language: en\n"), 0o644); err != nil {
 		t.Fatal(err)
 	}

internal/app/remote_session_test.go

@@ -10,6 +10,7 @@ import (
 	"strings"
 	"testing"
 
+	configkit "gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/storage"
 	"gitea.maximumdirect.net/eric/narratio/internal/artifacts"
 	"gitea.maximumdirect.net/eric/narratio/internal/config"
@@ -78,7 +79,7 @@ inputs:
 	setSessionConfigSearchPaths(t, []string{filepath.Join(t.TempDir(), "session.yml")})
 	deps, storeInitCalls := remoteSessionLoaderDependencies(fake)
 
-	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, "", config.SessionLoadOptions{SessionID: "2026-05-03"}, deps)
+	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, "", config.SessionLoadOptions{SessionID: "2026-05-03"}, nil, deps)
 	if err != nil {
 		t.Fatalf("loadCommandConfig() error = %v", err)
 	}
@@ -200,7 +201,7 @@ inputs:
 
 	var downloadedPath string
 	deps.downloadObjectToTemp = captureRemoteSessionTempPath(&downloadedPath)
-	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, "", config.SessionLoadOptions{SessionID: "2026-05-03"}, deps)
+	loaded, err := loadCommandConfigWithDependencies(context.Background(), pipelinePath, "", campaignPath, "", config.SessionLoadOptions{SessionID: "2026-05-03"}, nil, deps)
 	if err != nil {
 		t.Fatalf("loadCommandConfig() error = %v", err)
 	}
@@ -246,12 +247,18 @@ inputs:
     prefix: audio/
 `)
 	setSessionConfigSearchPaths(t, []string{filepath.Join(t.TempDir(), "session.yml")})
-	deps := commandDependencies{newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
-		if os.Getenv(accessKeyEnv) != "remote-session-key-id" || os.Getenv(secretKeyEnv) != "remote-session-secret" {
-			return nil, fmt.Errorf("secrets were not loaded before remote session object store init")
-		}
-		return fake, nil
-	}}
+	bootstrapCalls := 0
+	deps := commandDependencies{
+		loadSecretsDir: func(directory string) (*configkit.SecretsReport, error) {
+			bootstrapCalls++
+			return configkit.LoadSecretsDir(directory)
+		},
+		newObjectStore: func(context.Context, *config.Config) (storage.ObjectStore, error) {
+			if os.Getenv(accessKeyEnv) != "remote-session-key-id" || os.Getenv(secretKeyEnv) != "remote-session-secret" {
+				return nil, fmt.Errorf("secrets were not loaded before remote session object store init")
+			}
+			return fake, nil
+		}}
 
 	var stdout bytes.Buffer
 	var stderr bytes.Buffer
@@ -259,6 +266,9 @@ inputs:
 	if code != 0 {
 		t.Fatalf("exit code = %d, want 0; stdout=%q stderr=%q", code, stdout.String(), stderr.String())
 	}
+	if bootstrapCalls != 1 {
+		t.Fatalf("bootstrap calls = %d, want 1", bootstrapCalls)
+	}
 }
 
 func TestExecuteExplicitLocalSessionPrecedenceSkipsRemote(t *testing.T) {
@@ -468,10 +478,13 @@ func addSecretsToPipelineConfig(t *testing.T, pipelinePath, secretsDir, accessKe
 	pipelineYAML := strings.Replace(
 		string(pipelineData),
 		"storage:\n  backend: s3\n  s3:\n    bucket: test-bucket\n",
-		"storage:\n  backend: s3\n  s3:\n    bucket: test-bucket\n    access_key_id_env: "+accessKeyEnv+"\n    secret_access_key_env: "+secretKeyEnv+"\nsecrets:\n  env_dir: "+secretsDir+"\n",
+		"storage:\n  backend: s3\n  s3:\n    bucket: test-bucket\n    access_key_id: "+accessKeyEnv+"\n    secret_access_key: "+secretKeyEnv+"\n",
 		1,
 	)
 	if err := os.WriteFile(pipelinePath, []byte(pipelineYAML), 0o644); err != nil {
 		t.Fatalf("write pipeline: %v", err)
 	}
+	if err := os.Symlink(secretsDir, filepath.Join(filepath.Dir(pipelinePath), "secrets")); err != nil {
+		t.Fatalf("link conventional secrets directory: %v", err)
+	}
 }

internal/app/restore.go

@@ -53,7 +53,7 @@ func restoreWithDependencies(ctx context.Context, args []string, out io.Writer,
 	if strings.TrimSpace(flags.sessionID) == "" {
 		return fmt.Errorf("restore: session_id is required")
 	}
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("restore: %w", err)
 	}

internal/app/restore_execution_test.go

@@ -82,6 +82,28 @@ func TestExecuteRestoreRoundTripsPublishedExtractionAndManifestMetadata(t *testi
 	remoteManifest.RunID = "20260519T010203Z-a1b2c3d4"
 	producerRoot := "/prior/workspace/work/sample-campaign/2026-05-03"
 	remoteManifest.LocalWorkDir = filepath.Join(producerRoot, "runs", remoteManifest.RunID)
+	rawRelative := "transcripts/raw/alice.json"
+	rawOutput := manifest.ArtifactRecord{
+		Kind: "transcript_raw", LocalPath: filepath.Join(producerRoot, filepath.FromSlash(rawRelative)),
+		ProducerRunID: remoteManifest.RunID, Checksum: strings.Repeat("b", 64),
+	}
+	remoteManifest.MarkStageSucceeded("transcribe", time.Now().UTC(), []manifest.ArtifactRecord{rawOutput})
+	remoteManifest.Stages["transcribe"].TranscribeStateVersion = manifest.TranscribeStateContractVersion
+	remoteManifest.Stages["transcribe"].TranscribeSpeakers = map[string]manifest.TranscribeSpeakerRecord{
+		"alice": {
+			FingerprintVersion: manifest.TranscribeFingerprintContractVersion,
+			Fingerprint:        strings.Repeat("c", 64),
+			AudioPath:          "audio/alice.flac",
+			AudioSHA256:        strings.Repeat("a", 64),
+			AudioSize:          12,
+			Output: manifest.ArtifactRecord{
+				Kind: "transcript_raw", LocalPath: rawRelative,
+				ProducerRunID: remoteManifest.RunID, Checksum: strings.Repeat("b", 64),
+			},
+			OutputSize:    15,
+			ProducerRunID: remoteManifest.RunID,
+		},
+	}
 	remoteManifest.Stages["extract"] = &manifest.StageRecord{
 		Name: "extract", Status: manifest.StatusSucceeded,
 		Outputs: []manifest.ArtifactRecord{{
@@ -121,6 +143,13 @@ func TestExecuteRestoreRoundTripsPublishedExtractionAndManifestMetadata(t *testi
 	if want := filepath.Join(sessionRoot, "artifacts", "encounters.json"); lane.LocalPath != want {
 		t.Fatalf("rebased extraction path = %q, want %q", lane.LocalPath, want)
 	}
+	speaker := restored.Stages["transcribe"].TranscribeSpeakers["alice"]
+	if speaker.AudioPath != "audio/alice.flac" || speaker.Output.LocalPath != rawRelative {
+		t.Fatalf("restored speaker paths = %#v, want session-relative paths", speaker)
+	}
+	if _, err := os.Stat(filepath.Join(sessionRoot, filepath.FromSlash(rawRelative))); !os.IsNotExist(err) {
+		t.Fatalf("unpublished raw transcript should not be restored; stat err=%v", err)
+	}
 }
 
 func TestExecuteRestoreIncludeAudioRestoresAudio(t *testing.T) {

internal/app/restore_plan_test.go

@@ -24,6 +24,7 @@ func TestRestorePlanDefaultScope(t *testing.T) {
 	seedCurrentRestoreObject(t, store, current, current.SessionPrefix+"audio/alice.flac", []byte("audio"))
 	seedRestoreObject(store, current.SessionPrefix+"runs/20260519T010203Z-a1b2/manifest.json", []byte("{}"))
 	seedRestoreObject(store, current.SessionPrefix+"logs/publish.log", []byte("log"))
+	seedRestoreObject(store, current.SessionPrefix+"cache/notarius/references/ignored/party.yml", []byte("cache"))
 
 	plan, err := buildRestorePlan(context.Background(), cfg, current, store, RestorePlanOptions{})
 	if err != nil {

internal/app/restore_test.go

@@ -179,13 +179,8 @@ func TestExecuteRestoreLoadsSecretsBeforeObjectStoreInit(t *testing.T) {
 	secretsDir := filepath.Join(t.TempDir(), "secrets")
 	mustWriteSecretFile(t, filepath.Join(secretsDir, accessKeyEnv), "test-access-key-id\n")
 	mustWriteSecretFile(t, filepath.Join(secretsDir, secretKeyEnv), "test-secret-key\n")
-	f, err := os.OpenFile(pipelinePath, os.O_APPEND|os.O_WRONLY, 0)
-	if err != nil {
-		t.Fatalf("open pipeline config for append: %v", err)
-	}
-	defer f.Close()
-	if _, err := f.WriteString("\nsecrets:\n  env_dir: " + secretsDir + "\n"); err != nil {
-		t.Fatalf("append secrets config: %v", err)
+	if err := os.Symlink(secretsDir, filepath.Join(filepath.Dir(pipelinePath), "secrets")); err != nil {
+		t.Fatalf("link conventional secrets directory: %v", err)
 	}
 
 	storeInitCalled := false

internal/app/run.go

@@ -29,7 +29,7 @@ func runWithDependencies(ctx context.Context, args []string, out io.Writer, exec
 		return err
 	}
 	flags := request.Config
-	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), deps)
+	loaded, err := loadCommandConfigWithDependencies(ctx, flags.pipelinePath, flags.campaignPath, flags.campaignFilePath, flags.sessionPath, flags.sessionOptions(), flags.secretsDir.pointer(), deps)
 	if err != nil {
 		return fmt.Errorf("run: %w", err)
 	}

internal/app/run_stage.go

@@ -80,6 +80,8 @@ func runStageWithDependencies(ctx context.Context, args []string, out io.Writer,
 		SessionPath:       flags.sessionPath,
 		SessionID:         flags.sessionID,
 		PreviousSessionID: flags.previousSessionID,
+		Profile:           flags.profile.pointer(),
+		SecretsDir:        flags.secretsDir.pointer(),
 		Force:             force,
 		SelectedArtifacts: normalizedArtifacts,
 	}, executor, deps)
@@ -137,6 +139,8 @@ func analyzeWithDependencies(ctx context.Context, args []string, out io.Writer,
 		SessionPath:       flags.sessionPath,
 		SessionID:         flags.sessionID,
 		PreviousSessionID: flags.previousSessionID,
+		Profile:           flags.profile.pointer(),
+		SecretsDir:        flags.secretsDir.pointer(),
 		Force:             true,
 		SelectedArtifacts: normalizedArtifacts,
 	}, executor, deps)
@@ -192,6 +196,8 @@ func publishWithDependencies(ctx context.Context, args []string, out io.Writer,
 		SessionPath:       flags.sessionPath,
 		SessionID:         flags.sessionID,
 		PreviousSessionID: flags.previousSessionID,
+		Profile:           flags.profile.pointer(),
+		SecretsDir:        flags.secretsDir.pointer(),
 		Force:             true,
 		SelectedArtifacts: normalizedArtifacts,
 	}, executor, deps)
@@ -218,6 +224,8 @@ type singleStageCommand struct {
 	SessionPath       string
 	SessionID         string
 	PreviousSessionID string
+	Profile           *string
+	SecretsDir        *string
 	Force             bool
 	SelectedArtifacts []string
 }
@@ -232,7 +240,8 @@ func runSingleStageCommand(ctx context.Context, req singleStageCommand, executor
 	loaded, err := loadCommandConfigWithDependencies(ctx, req.PipelinePath, req.CampaignPath, req.CampaignFilePath, req.SessionPath, config.SessionLoadOptions{
 		SessionID:         req.SessionID,
 		PreviousSessionID: req.PreviousSessionID,
-	}, deps)
+		Profile:           req.Profile,
+	}, req.SecretsDir, deps)
 	if err != nil {
 		return nil, fmt.Errorf("%s: %w", req.CommandName, err)
 	}

internal/app/runner.go

@@ -211,11 +211,18 @@ func buildDefaultAuditaRunner(cfg *config.Config) (audita.Runner, error) {
 	if a.Report != nil {
 		report = *a.Report
 	}
+	llmAPIKey := ""
+	llmAPIKeyReference := ""
+	if a.LLMAPIKey != nil {
+		llmAPIKey = a.LLMAPIKey.Value()
+		llmAPIKeyReference = a.LLMAPIKey.EnvName()
+	}
 
 	runner, err := audita.NewSubprocessRunnerFromConfigValues(
 		a.Binary,
 		a.Timeout,
-		a.LLMAPIKeyEnv,
+		llmAPIKey,
+		llmAPIKeyReference,
 		append([]string(nil), a.Modules...),
 		a.BaseURL,
 		a.Model,

internal/app/runner_runtime.go

@@ -157,9 +157,6 @@ func initializeLockedInvocation(runtime *invocationRuntime) error {
 	if err := runtime.manifestStore.Save(runtime.ctx, runtime.manifestPath, m); err != nil {
 		return fmt.Errorf("save manifest identity %q: %w", runtime.manifestPath, err)
 	}
-	if _, err := loadSecretsFromConfig(runtime.env.Config, runtime.env.Logger); err != nil {
-		return runtime.persistTerminalFailure(fmt.Errorf("load secrets from files: %w", err))
-	}
 	return nil
 }