1# Data audit — F6-conf-012| file | SHA-256 | rows (excl. header) | cols |3|---|---|---|---|4| data/DGS10.csv | 8187f89ab6523b3c964255e1736cfcf0e2ca854ed9e50343e04f4dc2945c5cd7 (matches README & bundle.json) | 5217 | 2 |5| data/README.md | 32b825f02387ef3d552df6def82c585d7f248980667a28f30e58506e472e561a | provenance text | – |67- Schema: `observation_date` (YYYY-MM-DD, all parse, strictly increasing, no duplicates, all weekdays: Mon 1044, Tue 1044, Wed 1043, Thu 1043, Fri 1043); `DGS10` (percent, 2 decimals).8- Date range 2000-01-03 … 2019-12-31.9- Missing: 215 empty DGS10 fields (holidays: Jul-4 16, Dec-25 15, Jan-1 14, Nov-11 13, other Monday holidays...). No non-numeric non-empty values.10- Level range 1.37–6.79 %, mean 3.43 %.11- Changes (bp, n=5001): mean −0.093, sd 5.72, min −51 (2009-03-18, FOMC QE announcement), max +25. Heavy tails; kept (no outlier rule in TASK).12- 155 changes span >3 calendar days (holiday gaps); per TASK rule they are treated as one-day changes attributed to the weekday of t. Mondays therefore have fewer changes (919) because Monday holidays are dropped.13- Information availability: only DGS10.csv used; split train ≤ 2009-12-31 (last 2009-12-31), test ≥ 2010-01-01 (first 2010-01-04). Forecast weekday means use train only. No violations.14