# Data audit (P1)

File: data/canada_gdp_employment.csv

• SHA-256: f4c8e4976fb5f5f0c326e40bcb4f85b8b22963e5be89e2729e1a430c198f2bed (matches data/README.md)
• 345 data rows + header, 5 columns: month (date, YYYY-MM-01), gdp_level (float, M chained 2017 CAD, SAAR), gdp_yoy (float, %), emp_level (float, thousands), emp_yoy (float, %)
• Date range 1998-01-01 .. 2026-09-01, contiguous monthly, 0 duplicate months.
• Missing: gdp_level and gdp_yoy missing for 2026-08 and 2026-09 only (not yet published, as documented). emp complete.
• Consistency: recomputed 100*(L_m/L_{m-12}-1) rounded to 4 dp equals the given yoy exactly (max abs diff 0.0) for both series wherever the 12-month lag is in the file (from 1999-01). The 1998 yoy values depend on 1997 levels not in the file and cannot be verified; accepted as given.
• Anomalies: COVID-19 shock 2020-03..2021-12: gdp_yoy ranges −16.2 to +18.5, emp_yoy −15.6 to +16.1 (base effects in 2021). Real, not errors, but extreme leverage for linear regressions.
• Information availability: the file is a current vintage (revised data); timing is simulated by restricting rows per origin (GDP ≤ t−2, emp ≤ t−1). No input column violates rules once that filter is applied; the post-origin rows are used only as actuals.
• Handling decisions: (1) no row deleted from the file; (2) per-origin information filter; (3) the model's estimation sample excludes regression rows whose target month or any regressor month falls in 2020-03..2021-12 (pandemic window; decided before computing any out-of-sample error). Logged in out/exclusions.csv for the last estimation sample.