Document Notarius cache reuse workflow

This commit is contained in:
2026-09-16 18:14:40 +00:00
parent 8ff74e1251
commit a0312ed034
11 changed files with 228 additions and 31 deletions

View File

@@ -187,7 +187,20 @@ 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:
## 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
@@ -260,9 +273,14 @@ 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
@@ -297,6 +315,28 @@ 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
@@ -509,6 +549,7 @@ 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
@@ -519,6 +560,13 @@ 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: