Data audit
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/gc5y_daily.csv | 2cdda6ccb2147c8861997166cb0d32cf85ff2744e1e7c0f47eded0f354c2db51 (matches README) | 2498 | 3 |
| data/README.md | 7716f8b5...e34 (matches bundle.json) | — | — |
- Schema: date (str YYYY-MM-DD), yield_5y (float, percent), post_cut (int).
- Date range 2010-01-04 .. 2019-12-31, strictly increasing, 0 duplicate dates, 0 missing values.
- yield_5y: min 0.48, max 3.20, mean 1.574 %. One-day changes: mean -0.043 bp, sd 4.24 bp, min -27, max +17 bp (largest moves Aug 2011 and 2015-01-21, the BoC surprise cut; these are plausible real moves).
- Calendar gaps between rows: 1962×1 day, 10×2, 440×3 (weekends), 79×4, 6×5 days (holidays). Nothing anomalous.
- post_cut: all 2498 rows = 0. n_treated = 0. README shows the retrieval used end_date=2019-12-31, so the file stops before the 2020-03-04 event. The flag matches the stated rule (date >= 2020-03-04) on every row, so it is right but has nothing to flag.
- Decision: the estimand cannot be computed. I will not use a placebo or proxy date and will not pull in outside data (the rules forbid both). Result: premise_invalid / infeasible, with nulls for the effect, CI, model_metric and n_analysed. The descriptive baseline, table and figure are still produced.