# Data audit (P1)

All SHA-256 values computed with `shasum -a 256` match data/README.md.

| file | SHA-256 | data rows | cols | range | missing | dup dates | gaps |
|---|---|---|---|---|---|---|---|
| vintages/..._2024-01-31.csv | 7ce218fb…f243 | 307 | 2 | 1947-04-01..2023-10-01 | 0 | 0 | 0 |
| vintages/..._2024-04-30.csv | 62c1dfce…97cb | 308 | 2 | ..2024-01-01 | 0 | 0 | 0 |
| vintages/..._2024-07-31.csv | 6f942591…79da | 309 | 2 | ..2024-04-01 | 0 | 0 | 0 |
| vintages/..._2024-10-31.csv | c4c51ede…0003 | 310 | 2 | ..2024-07-01 | 0 | 0 | 0 |
| vintages/..._2025-01-31.csv | 10a5e02e…aee2 | 311 | 2 | ..2024-10-01 | 0 | 0 | 0 |
| vintages/..._2025-04-30.csv | 1f19c975…5085d1a | 312 | 2 | ..2025-01-01 | 0 | 0 | 0 |
| vintages/..._2025-07-31.csv | f1e7c556…57a9 | 313 | 2 | ..2025-04-01 | 0 | 0 | 0 |
| vintages/..._2025-12-31.csv | c111c070…42a8d4a7 | 314 | 2 | ..2025-07-01 | 0 | 0 | 0 |
| current_vintage.csv | 1af2590d…0562 | 318 | 2 | 1947-04-01..2026-07-01 | 0 | 0 | 0 |
| first_release.csv | ab843c9f…95bb | 13 | 3 | quarters 2023-07-01..2026-07-01 | 0 | 0 | – |

Units: percent change from preceding quarter, SAAR (pp of annualised growth). Value range in vintages about -28.1 .. 35.2 (2020Q2/Q3 pandemic).

Anomalies / notes:
- Pandemic outliers 2020Q2 (≈-28) and 2020Q3 (≈+35) in all vintages → set to missing in the AR(1) estimation sample only (see exclusions.csv; per-origin in exclusions_by_origin.csv).
- Values of earlier quarters differ across vintages (revisions), as expected for real-time data; each origin uses only its own file.
- Last quarter of each vintage matches README; and equals first_release value of that quarter (consistent: the vintage dates are the first-release dates).
- Eighth origin is 2025-12-31 (shutdown delay); h=1 target 2025Q4 first released 2026-02-28.
- Every target's first_release_vintage_date is ≥ origin date (h=2 target of origin 2025-07-31 is 2025Q3, first released 2025-12-31 — after the origin). Asserted in analysis.py.

Information-availability: current_vintage.csv violates the real-time rule for estimation; it is read only for the revision statistic. No other input violates the rules.