# Data audit (P1)
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/DGS10.csv | 8187f89ab6523b3c964255e1736cfcf0e2ca854ed9e50343e04f4dc2945c5cd7 (matches README and bundle.json) | 5217 | 2 |
| data/README.md | provenance only (not analysed) | – | – |

- Schema: `observation_date` (YYYY-MM-DD), `DGS10` (percent, 2 decimals). Range 2000-01-03 … 2019-12-31; sorted; 0 duplicate dates; 0 weekend dates.
- Missing: 215 empty DGS10 fields (US market holidays, incl. 2001-09-11 and 2001-09-12). No '.' markers, no non-numeric values.
- Levels: 1.37 … 6.79 %, mean 3.43 %. 5002 available observations → 5001 changes (bp): mean −0.093, sd 5.72, min −51 (2009-03-18, FOMC QE1 announcement), max +25.
- Calendar gaps between available obs: 1 day 3899, 2 days 58, 3 days 889, 4 days 155. 214 changes span a holiday.
- Extreme changes (|Δy| ≥ 25 bp): 2001-11-15, 2008-09-15, 2008-11-20, 2009-02-17, 2009-03-18 — all crisis/policy dates, genuine; kept.
- Availability rules: no input violates them; the only rule-relevant step is the split at 2009-12-31/2010-01-01 enforced in code (asserted). Exclusions in out/exclusions.csv: 5217 − 215 − 1 = 5001 analysed.