Protect transcript refinement resume semantics

This commit is contained in:
2026-08-30 13:03:13 +00:00
parent 4d6086fefb
commit 7ee637803d
12 changed files with 512 additions and 16 deletions

View File

@@ -207,6 +207,13 @@ runner marks the aggregate record stale and executes it. Analyze's validator
can still accept a partial selection when only unrelated artifact records are
stale.
The current production semantic contracts cover prepare, transcribe, merge,
polish, normalize, trim, and render. They record resolved choices Narratio can
observe, not operational runner tuning. External model, module, prompt,
profile, and configuration-file contents that a tool privately loads remain
outside the manifest contract when their configured identifier is unchanged;
operators must force the affected stage after such a private content change.
Session manifest is the authoritative stage-progress ledger across invocations.
Run manifest is invocation-scoped audit state.

View File

@@ -26,9 +26,17 @@ Normalize polished transcript into final transcript using Seriatim.
- final transcript must validate as processed transcript JSON (`segments` array).
- normalize defaults are applied when `pipeline.normalize` is unset.
## Resume Semantics
The versioned semantic fingerprint covers the Seriatim normalize operation,
output schema and canonical output identity, plus the configured transcript
transformations. Seriatim's executable and timeout and optional report
generation are operational and do not invalidate the normalized transcript.
## Related Contracts And Tests
- [Seriatim](../integrations/seriatim.md) owns subprocess and output semantics.
- [Configuration](../config.md#pipeline) owns normalize fields and defaults.
- Implementation and tests: `internal/stage/normalize.go`,
`internal/stage/normalize_test.go`
`internal/stage/normalize_test.go`,
`internal/stage/semantic_contracts_refinement.go`

View File

@@ -28,10 +28,26 @@ Run Audita polishing on base transcript and produce polished transcript.
- polished transcript schema validation is mandatory.
- report output is config-gated.
## Resume Semantics
The versioned semantic fingerprint covers the Audita service endpoint, model,
validation model, module set, transcript description, output schema, selected
external configuration path, and canonical polished-transcript identity. Module
ordering is normalized because the configured modules form a set. Audita's
executable, timeouts, concurrency, report and debug behavior, work retention,
and credential environment name are operational and do not invalidate a
successful result.
Narratio can fingerprint a selected model, module, or configuration identifier,
but it cannot inspect content that Audita privately resolves behind that stable
identifier. Force `polish` after changing such private content without changing
its identifier.
## Related Contracts And Tests
- [Audita](../integrations/audita.md) owns subprocess, validation, and failure
semantics.
- [Configuration](../config.md#pipeline) owns operator-selected Audita values.
- Implementation and tests: `internal/stage/polish.go`,
`internal/stage/polish_test.go`
`internal/stage/polish_test.go`,
`internal/stage/semantic_contracts_refinement.go`

View File

@@ -26,7 +26,8 @@ records that may consume rendered transcripts.
- records input provenance, output paths, adapter metadata, logs, and generated config refs.
- when `pipeline.render.enabled=false`, completes successfully with no outputs
and records explanatory metadata. This is not an explicit self-skip: both
manifests record success, and enabling render later requires a forced run.
manifests record success. Because enablement is fingerprinted, enabling
render later automatically makes the prior result non-resumable.
## Failure Semantics
@@ -40,9 +41,20 @@ records that may consume rendered transcripts.
- only `format: markdown` is supported.
- render stage owns production of built-in Markdown transcript sources.
## Resume Semantics
The versioned semantic fingerprint covers enablement, final format, resolved
title (including the session-title fallback), timestamp, segment-ID and
metadata inclusion, both canonical input identities, and both Markdown output
identities. Seriatim's executable, timeout, and report behavior are operational
and do not invalidate rendered transcripts. A render-only change leaves the
independent `extract` sibling reusable while invalidating their shared
downstream consumers.
## Related Contracts And Tests
- [Seriatim](../integrations/seriatim.md) owns render subprocess behavior.
- [Configuration](../config.md#pipeline) owns render fields and defaults.
- Implementation and tests: `internal/stage/render.go`,
`internal/stage/render_test.go`
`internal/stage/render_test.go`,
`internal/stage/semantic_contracts_refinement.go`

View File

@@ -33,6 +33,19 @@ When `trim.enabled=false`:
- bounds output exists only in enabled trim path.
- render-debug output is diagnostic and not a declared stage output.
## Resume Semantics
The versioned semantic fingerprint covers enablement, the bounds prompt and
profile identifiers, the Scriptorium configuration identity, transcript input
name, sticky session variable, bounds and trimmed output identities, and the
Seriatim trim operation. Diagnostic bounds rendering, diagnostic output paths,
timeouts, executable paths, and optional reports are operational and do not
invalidate the canonical trimmed transcript.
Narratio cannot inspect prompt, profile, or configuration content that
Scriptorium or Seriatim privately resolves behind a stable identifier. Force
`trim` after changing such private content without changing its identifier.
## Related Contracts And Tests
- [Scriptorium](../integrations/scriptorium.md) owns bounds generation and
@@ -40,4 +53,5 @@ When `trim.enabled=false`:
- [Seriatim](../integrations/seriatim.md) owns transcript trimming behavior.
- [Configuration](../config.md#pipeline) owns trim fields and defaults.
- Implementation and tests: `internal/stage/trim.go`,
`internal/stage/trim_test.go`
`internal/stage/trim_test.go`,
`internal/stage/semantic_contracts_refinement.go`