# Data audit

• File: data/ip_predictors.csv; SHA-256 7e388665caf3e4e7305ae6ed8f5625ab12a6c8df5c5696c1b5dd5b86da62a39b (matches README).
• 417 data rows + header, 7 columns: month (date, YYYY-MM-01), ip_growth (% per month), oil_chg (% per month), fedfunds (% level), permit_chg (% per month), neworder_chg (% per month), tcu (% of capacity). All numeric float.
• Range 1992-01-01 … 2026-09-01; months unique, consecutive, no gaps, no duplicates.
• Missing: neworder_chg 1992-01 and 1992-02 only (2 cells). README says the first month of each change series is empty, but ip_growth/oil_chg/permit_chg are populated in 1992-01 — documentation discrepancy, harmless (only ip_growth 1992-01/02 enters, as lags of the 1992-03 row, as the task spec implies).
• Anomalies: large COVID moves (ip_growth 2020-04 −14.14, 2020-06 +6.35; oil_chg ±55 in 2020-03..05), GFC 2008-09 ip −4.50. Genuine; retained (ridge spec fixed).
• Availability: data extend to 2026-09, which is only used as the actual for the last origin (2026-08). No input violates the rules given the spec; the script enforces s+1 ≤ t for training targets (asserted).
• Vintage: current vintage; publication lags ignored by design.