Data audit
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/UNRATE.csv | 621ff6fadf50910cb7a21ee20ba610c247cb043b29b4e05c4c1a72fadca2a06c (matches README & bundle.json) | 945 | 2 |
- Schema:
observation_date(YYYY-MM-01 string),UNRATE(float, percent of civilian labour force, SA). - Date range 1948-01-01 to 2026-09-01; consecutive months, no gaps in dates, no duplicate dates.
- Values: 944 non-missing, min 2.5, max 14.8 (2020-04, COVID), mean 5.65. No implausible values.
- Missing: 1 value, 2025-10-01 (empty field; consistent with no October 2025 CPS household survey release during the 2025 US federal government shutdown — stated as context, not verified from the data).
- Consequences: (a) origin 2025-10-01: naive uses 2025-09-01 value; model must forecast from 2025-09; (b) targets 2025-10-01 are missing for (origin 2025-09-01, h=1) and (origin 2025-07-01, h=3) → kept in ledger, excluded from error statistics; (c) later training windows contain a gap; differences involving 2025-10 are NaN.
- Information-availability: the file is the final vintage, so all values are revised data; no input violates the date-≤-t rule as long as we slice by date. Handling: evaluation reported as pseudo-out-of-sample on revised data, no real-time claim.
- Anomaly: COVID spike 2020-04 (14.8, +10.4 pp month-on-month) is a genuine observation; kept (no outlier removal), noted as a possible influence on AR coefficients.