Task 094: Transcription-Report Sync Validation

Status: DONE

Motivation

CLAUDE.md specifies multiple times: "When transcription corrections are made, update corresponding dialogue quotes in episode reports. Run validation to check consistency between dialogue data files and report citations."

Currently, the validation tests check:

• Speaker names in reports match dialogue registry ✓
• Structural integrity of reports ✓
• Referential integrity (evidenceQuoteIds → dialogueQuotes) ✓

Missing validation:

• Dialogue quote text in reports vs actual dialogue text in epXX_dialogue.json
• Dialogue quote timestamps in reports vs actual dialogue timestamps
• Speaker attribution for quoted text matches the attributed speaker in dialogue data

This creates a risk of drift: when transcriptions are corrected (Task 089 just completed a full re-review), report quotes may become stale.

Scope

1. Add test: For each dialogue quote in an episode report, find the closest matching entry in dialogue data and validate:

- Speaker name matches

- Text is a substring or close match (quotes may be abbreviated)

- Timestamp is within reasonable tolerance (±5s)

1. Report which quotes have no matching dialogue entry (possible fabrication or paraphrase)
2. All new validation as tests in report-data-validation.test.ts

Notes

• Dialogue quotes in reports are often abbreviated or paraphrased — exact match is not expected
• Timestamp matching needs tolerance since VTT timestamps are approximate
• This is a guard against drift, not a strict equality check