Data audit
| file | sha256 | rows×cols |
|---|---|---|
| data/ip_predictors.csv | 7e388665caf3e4e7305ae6ed8f5625ab12a6c8df5c5696c1b5dd5b86da62a39b (matches README & bundle.json) | 417 × 7 |
| data/README.md | 225f52077afb6f435f67e2b398cbe59963ab1ce6d0d0d4e8257fd27c777ede6e | — |
- Schema: month (YYYY-MM-01, parsed date), ip_growth, oil_chg, permit_chg, neworder_chg (percent per month, 100·Δln), fedfunds (percent, level), tcu (percent of capacity, level). All float64.
- Date range 1992-01-01 .. 2026-09-01, contiguous monthly, all day=1, no duplicate months.
- Missing: only neworder_chg at 1992-01 and 1992-02. No other NaN.
- Anomaly / documentation discrepancy: README says the first month of each change series is empty, but ip_growth, oil_chg, permit_chg have 1992-01 values. Used as given (only ip_growth 1992-01 enters, as lag 2 of the first training row).
- Extremes: ip_growth min -14.14 (2020-04, COVID); evaluation-period actuals (2021-10..2026-09) lie in [-1.38, 1.33].
- Availability: the file contains data through 2026-09; month 2026-09 is used only as the target of the last origin (2026-08), never as a feature/scaling input. All scaling statistics are computed per origin from training rows only (verified by perturbation test, see checks.md).
- Vintage: current vintage, publication lags ignored by task design → not real-time.