Revert "feat: updating minor features" - #107
Conversation
This reverts commit 5a6e2f7.
|
| Filename | Overview |
|---|---|
| app/api/v1/routes/call_imports.py | Restores catalog-only row mutations, breaking delete, diarisation cancellation, and speaker swapping for shard-resident rows. |
| app/workers/tasks/evaluate_call_import_row.py | Forces all evaluation workers to score diarised transcripts, including retries of persisted production-source runs. |
| app/api/v1/routes/call_import_evaluations.py | Restricts new runs to diarised transcripts and removes source-aware row serialization without compatibility handling for existing production runs. |
| app/workers/tasks/process_call_import_row.py | Unconditionally chains post-import diarisation; validated new runs supply the required configuration, while legacy-source effects are covered by the evaluation compatibility finding. |
| app/services/reporting/call_import_evaluation_pdf_report.py | Reverts report-section classes and pagination-oriented CSS, with no independently publishable correctness failure established. |
| frontend/src/pages/callImports/CallImportDetail.tsx | Removes production-source evaluation selection and transcript copy controls to align the UI with diarised-only creation. |
| frontend/src/pages/metrics/MetricsManagement.tsx | Removes metric clipboard import/export controls and fixes newly created categories to single-choice mode. |
Reviews (1): Last reviewed commit: "Revert "feat: updating minor features (#..." | Re-trigger Greptile
| row = ( | ||
| db.query(CallImportRow) | ||
| .filter( | ||
| CallImportRow.id == row_id, | ||
| CallImportRow.call_import_id == call_import.id, | ||
| ) | ||
| .first() | ||
| ) |
There was a problem hiding this comment.
When sharding is enabled, get_db supplies the catalog session while call-import rows reside on shard databases, so this direct query returns 404 for existing rows and prevents deletion. The same catalog-only lookup also breaks single-row diarisation cancellation and speaker swapping.
|
|
||
| production_transcript = (source_row.transcript or "").strip() | ||
| diarised_transcript = (source_row.diarised_transcript or "").strip() | ||
| eval_source = ( | ||
| (evaluation.transcript_source or "diarised").strip().lower() | ||
| ) | ||
| if eval_source == "production": | ||
| transcript = production_transcript | ||
| missing_label = "production" | ||
| else: | ||
| transcript = diarised_transcript | ||
| missing_label = "diarised" | ||
| transcript = diarised_transcript | ||
| missing_label = "diarised" |
There was a problem hiding this comment.
When a persisted production-source evaluation is retried, this worker unconditionally scores diarised_transcript while the run can still report its source as production, causing the retried metric results to use a different input than the original run.
Reverts #106