# Data audit (P1)
| file | SHA-256 | rows (excl. header) | cols |
|---|---|---|---|
| data/cpi_quarterly.csv | c8652dec1567ec35e3b84ee8fa55545e9bf0ae4990a4d6d05b89a6d22f6801ea | 127 | 3 (quarter, cpi_avg, infl_annualised) |
| data/cpi_monthly.csv | 855d069176369d30d02cbbe0cc6456b76d2f3967cdaabc1b362c2461ef9015a2 | 381 | 2 (month, cpi) |

SHA-256 values match data/README.md.
- Quarterly: 1995-01-01..2026-07-01, regular quarterly spacing, no duplicates. `infl_annualised` missing only for 1995Q1 (by construction). Units: percent at annual rate. Range −6.04 to 12.15, mean 2.13, sd 2.39. Strong seasonality: mean by quarter Q1 2.32, Q2 4.18, Q3 1.66, Q4 0.30.
- Monthly: 1995-01..2026-09, no missing, no duplicates, index 2002=100 (1 decimal).
- Cross-check: quarterly means recomputed from monthly match cpi_avg to 3.3e-7; 400·Δln(cpi_avg) matches infl_annualised to 3.3e-6 (rounding of cpi_avg to 6 decimals). Not an issue; the given `infl_annualised` is used as the target.
- Availability: the file contains data after the last origin (to 2026Q3); these are used only as realised targets (latest target 2026Q3 = 2025Q3 + 4). Every model fit at origin t is restricted in code to quarters ≤ t (asserted).