Data audit (P1)
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/gc5y_daily.csv | 2cdda6ccb2147c8861997166cb0d32cf85ff2744e1e7c0f47eded0f354c2db51 (matches README) | 2498 | 3 |
| data/README.md | 7716f8b58d8f4729504a7713a8a3d651cd2f7298f53f668b377215a5b2af1e34 | n/a | n/a |
- Schema: date (YYYY-MM-DD string), yield_5y (percent, float, range 0.48–3.20), post_cut (int).
- Date range: 2010-01-04 to 2019-12-31; sorted ascending; 0 duplicate dates; 0 missing values.
- Gaps between consecutive rows: 1–5 calendar days (weekends/holidays); largest gaps (5 days) around year-end holidays – normal for business-daily data.
- Largest absolute daily change: 27 bp; no implausible values.
- Critical: post_cut = 0 for all 2498 rows. The README's retrieval URL used end_date=2019-12-31, so the file contains no observations on/after 2020-03-04. The treated window is empty; the estimand cannot be computed.
- Information-availability: no input violates the rules; the problem is missing coverage, not leakage. Handling: declare premise_invalid, report only the descriptive baseline.