# Data audit
| file | SHA-256 | rows | cols | dates | missing | dups |
|---|---|---|---|---|---|---|
| cpi_vintage_2016-07-15.csv | 281fbcdd6a6eeb3e9cb7c1f20476e6bd1f7b925bf140606fe46538fe83f2c83e | 834 | date,cpi | 1947-01..2016-06, contiguous | 0 | 0 |
| cpi_vintage_2026-10-28.csv | 8220d5b4d70ca32fc2c7b1975651c515967db8a87053cf725159c63ed274cfba | 153 | date,cpi | 2014-01..2026-09, contiguous | 1 (2025-10 empty) | 0 |
| cpi_assembled.csv | ff732b201a525d6fb171d4e8b2d470f986863c1547a1928e42bb197c0777d44d | 957 | date,cpi,vintage | 1947-01..2026-10 | 1 (at 2025-11) | 0 |
| README.md | ac136bff13bbdb5afac60e0404212ea1f73f992bd5cca1c1e7a6599df78f8965 | — | — | — | — | — |

Units: index 1982-84=100, 3 decimals (pre-2007 values in 2016 vintage have 1 decimal). Note: SHA-256 in README refer to raw downloads, not these reformatted files (README says reformatted) — not comparable.
Overlap 2014-01..2016-06 (30 months): vintages differ by up to ~0.5 index points (revised seasonal factors); no gross discrepancy (>1).
Max monthly % change in 2026 vintage ~1.26% (2022), plausible.

**Assembled file errors:** (1) overlap months use the 2016 vintage, violating the most-recent rule; (2) 2016-07 missing; (3) from 2016-08 onward values are shifted one month later (value of month m placed at m+1), so a spurious 2026-10 row and the missing value appears at 2025-11. **Decision:** the assembled file is read only for audit; the series is rebuilt from the two vintage files. No input violates the availability rules except the assembled file's construction.