During the live repair-d window the migration's spot-check reported ok=false for 11 of 50 samples. Lead verification against zoneinfo ground truth showed all 50 stored conversions CORRECT — the checker, not the data, is wrong (likely it re-converts without provenance context or mishandles DST windows; all 11 false flags were Asia/Jerusalem-converted whatsapp rows). A self-check that false-negatives will train operators to ignore it. Fix the checker to replicate the provenance-aware conversion path exactly, add a DST-boundary test matrix. Evidence: docs.local/repair-d-timestamp-live-result.json + wave25 collab receipt DONE_REPAIR_D_LIVE_MAIN.
🤖 Generated with Claude Code
During the live repair-d window the migration's spot-check reported ok=false for 11 of 50 samples. Lead verification against zoneinfo ground truth showed all 50 stored conversions CORRECT — the checker, not the data, is wrong (likely it re-converts without provenance context or mishandles DST windows; all 11 false flags were Asia/Jerusalem-converted whatsapp rows). A self-check that false-negatives will train operators to ignore it. Fix the checker to replicate the provenance-aware conversion path exactly, add a DST-boundary test matrix. Evidence: docs.local/repair-d-timestamp-live-result.json + wave25 collab receipt DONE_REPAIR_D_LIVE_MAIN.
🤖 Generated with Claude Code