Permanent regression check (Phase 1, Test 5): a fixed set of series and a fixed historical window, re-diffed on every harvest. Any value that changes for a period already published proves silent restatement on FactIQ's side.
Baseline established: 2026-07-28. See findings/drift-baseline.json for
the exact snapshot (3 series: BLS nonfarm payrolls, BEA real GDP % change,
China Customs preliminary total exports — chosen to span three unrelated
providers). Diff logic lives in src/validation/drift_check.py
(diff_snapshots), fully deterministic, no LLM in the comparison.
Open issue: this is not yet running automatically. FactIQ is only reachable via MCP tool calls inside a Claude Code session in this project — there's no plain REST endpoint for a standalone cron process to call. The Phase 3 harvester needs to resolve how a scheduled process authenticates to FactIQ before this check can run on a real daily cadence; until then, new entries below are appended from manually re-run snapshots.
No entries yet below this line — the first re-diff will be the first line of real evidence either way.
No restatements detected against the baseline.
Password required.