# Data audit (P1)

• data/gc5y_daily.csv SHA-256 2cdda6ccb2147c8861997166cb0d32cf85ff2744e1e7c0f47eded0f354c2db51 (matches README). 2498 rows, 3 cols (date str, yield_5y float %, post_cut int).
• data/README.md SHA-256 7716f8b58d8f4729504a7713a8a3d651cd2f7298f53f668b377215a5b2af1e34.
• Dates 2010-01-04 .. 2019-12-31, strictly increasing, no duplicates, no missing values. Yield range 0.48–3.20 %.
• Gaps between rows: 1d 1962, 3d 440 (weekends), 2d 10, 4d 79, 5d 6 (holidays). Nothing anomalous.
• Critical: post_cut is 0 on all 2498 rows. The file ends 2019-12-31, before the 2020-03-04 treatment date (the README's retrieval URL uses end_date=2019-12-31). The flag agrees with the rule "date ≥ 2020-03-04" on every row, so the flag is not wrong; the treated period just isn't in the sample.
• Decision: the estimand can't be computed. Conclusion is premise_invalid and feasibility is infeasible. Effect, CI, model_metric and n_analysed are null. I will not use a proxy (for example, a pseudo-event date). Only the descriptive baseline and the table/figure are produced.
• No input breaks the availability rules.