Well, we compute per-service factors per reconciliation run, and each run's factors are saved with their date window (start_date, end_date, generated_at) in a report artifact. But restatement currently applies whatever factor you pass at execution time. There's no calibration_adjustments table tracking what corrections have already been applied to which date ranges. So double-application is a real risk if you're not disciplined about reusing the saved report. That's a gap we didn't notice, a correction history table is the right fix, thanks for that catch.