Fetching metadata from the HF Docker repository...
-
Phase 11: Telemetry instrumentation + OffsetMapper tests
Task 5: OffsetMapper validation suite (36 tests, all passing)
- Covers: insertions, deletions, replacements, Arabic text, multi-edits, chained mutations
- FINDINGS: Delete-boundary off-by-one behavior documented
- FINDINGS: Tanween deletion end-position mapping loss documented
Task 1: Grammar filter telemetry instrumentation
- Added [FILTER-TEL] JSON events at every grammar filter decision point
- Events: grammar_raw_output, grammar_diffs_extracted, grammar_diff,
filter_reject (StageLocker|Jaccard_03|Jaccard_05|DigitGuard|TanweenGuard|
PunctuationGuard|LatinGuard|DirectionalBlock|IVtoOOV), patch_accepted
- NO correction logic changes
Task 8: Local telemetry capture tool
- telemetry_capture.py: runs benchmark, fetches HF logs, extracts events
- Stores: artifacts/telemetry.jsonl, artifacts/telemetry_summary.json