# Data audit (P1)

File: data/canada_gdp_employment.csv

• SHA-256: f4c8e4976fb5f5f0c326e40bcb4f85b8b22963e5be89e2729e1a430c198f2bed (matches README and bundle.json)
• 345 data rows x 5 columns: month (date, YYYY-MM-01), gdp_level (float, M chained 2017 $, SAAR), gdp_yoy (%, 4 dp), emp_level (float, thousands), emp_yoy (%, 4 dp)
• Range 1998-01-01 .. 2026-09-01; monthly, contiguous (no gaps: consecutive diffs 28-31 days); 0 duplicated months.
• Missing: gdp_level and gdp_yoy missing for 2026-08, 2026-09 only (not yet published) — expected. emp columns complete.
• yoy recomputation from levels (where 12-month lag exists in file) matches given yoy to <=5e-5 (rounding).
• Anomaly / doc mismatch: README says gdp_yoy is empty for the first 12 months, but gdp_yoy and emp_yoy are populated for 1998 (presumably computed from pre-1998 levels not in the file). Handling: use the given values as-is (they are the task's defined target/predictor); noted in data_issues.
• Anomaly: COVID-19 shock 2020-03..2021-06 (gdp_yoy from -16.2 to +18.5; emp_yoy -15.6 to +16.1) and base effects through 2021. Handling: pre-specified (before seeing any out-of-sample errors) exclusion from the model's training sample of pairs whose target or regressor month lies in 2020-03..2021-12; logged per origin; sensitivity with no exclusion reported. Naive baseline unaffected.
• Information availability: the file holds data to 2026-09 (beyond every origin). Handled by masking inside each origin loop: GDP months <= t-2, employment months <= t-1. All 96 target months (2022-06..2026-07) have observed actuals.