Trim and validate scene caveats

This commit is contained in:
2026-07-05 13:30:32 +00:00
parent 95218218e2
commit d3f790095e
6 changed files with 49 additions and 9 deletions

View File

@@ -88,7 +88,8 @@ last, exact source-unit IDs, sequential contiguous scenes, and no overlap. It
assigns chunk IDs such as `scene-000001` and stores scene metadata including
title, primary mode, participants, summary, boundary note, confidence, boundary
unit IDs, and unit count. Boundary caveats become warnings with reason code
`scene_boundary_caveat`.
`scene_boundary_caveat`. Whitespace-only caveats are treated as malformed
structured output rather than silently dropped.
Malformed model output fails explicitly rather than falling back to another
chunker. The chunker exposes prompt and response-schema provenance through

View File

@@ -183,6 +183,7 @@ Symptoms include:
- `dnd scenes chunker`
- `malformed structured output`
- `boundary_caveats`
- `start_unit_id`
- `end_unit_id`
- `gap`
@@ -199,6 +200,8 @@ Fix:
- Inspect retained diagnostics for the run error and resolved pipeline.
- If the error names malformed structured output, retry with a model that
follows structured response schemas reliably.
- If the error names `boundary_caveats`, check for blank or whitespace-only
caveat text in the scene response.
- Scene boundaries must use exact source-unit IDs, cover the full source
document, be contiguous, and not overlap.