Data type
Transcript correction
Stage up to 200 transcript text, speaker, and timestamp corrections; preview their impact; recover conflicts; compare revisions; and compensate the latest ordinary batch without rewriting history.
Implemented task
Public task
Open a ready source and choose Revisions.
Expected outcome: A previewed correction batch becomes one immutable revision or one explicit conflict recovery flow.
Declared product contract
Inputs and task boundary
Inputs
- A readable transcript revision
- Up to 200 existing-line text, speaker, start-time, or end-time edits
Input constraints
- The batch edits existing lines only and must target the current readable revision before preview.
- Every conflict choice requires another impact preview before commit.
User actions
- Open Revisions from a source or line shortcut.
- Stage one or more distant line edits and preview downstream impact.
- Resolve any field conflict and save one immutable revision.
- Compare revisions or compensate the current ordinary batch.
Outputs
- One immutable transcript revision with impact and audit metadata.
- A paged field-level or structural revision comparison.
Decision boundary
Where it fits
- Researcher-reviewed corrections to text, speaker labels, and time ranges.
Outside the boundary
Where it does not fit
- Structural insertion, deletion, split, merge, or reorder as a correction write.
Current scope
Limitations
- Correction batches edit existing lines only; insert, delete, split, merge, and reorder remain read-only diff outcomes from structural reprocessing.
Verification
Product proof
journey
proof.x-r03-transcript-revision
- Verified
- Review due
- Expires
- A 2,000-line transcript completes multi-row impact preview, atomic correction, 409 recovery, arbitrary diff, and compensation.
- Old reviews and public sharing remain historically bound without fabricated review.
- Reviewer read-only, outsider denial, reduced motion, bounded DOM, and zero browser errors are recorded.
Failure boundary: Correction batches edit existing lines; structural insert, delete, split, merge, and reorder remain read-only diff outcomes.
benchmark
proof.m10-transcript-correction-benchmark
- Verified
- Review due
- Expires
- Three real PostgreSQL preview and commit repetitions each preserve 2,000 transcript lines and change exactly 200 existing lines.
- The artifact records the input digest, environment, repetition timings, broad noise tolerance, and bounded feature scope.
- The diagnostic timing boundary is explicitly not a production latency objective, guarantee, or SLA.
- Planned saved-query and report builder features are not benchmarked or promoted.
Failure boundary: This benchmark covers only the implemented transcript-correction preview/commit path for one deterministic synthetic workload; it excludes planned features and states no production SLA.
OpenVerbatim is an open-source (Apache-2.0) qualitative data analysis platform for coding and analyzing interview transcripts. AI-suggested codes stay marked as suggestions until a human reviewer confirms or rejects them, and every decision is kept in an audit trail. The full feature set is available when self-hosted; there is no paid feature wall.