Data audit
| file | SHA-256 | data rows | cols | date range | empty cpi | dup dates |
|---|---|---|---|---|---|---|
| cpi_vintage_2016-07-15.csv | 281fbcdd6a6eeb3e9cb7c1f20476e6bd1f7b925bf140606fe46538fe83f2c83e | 834 | 2 (date, cpi) | 1947-01-01..2016-06-01 | 0 | 0 |
| cpi_vintage_2026-10-28.csv | 8220d5b4d70ca32fc2c7b1975651c515967db8a87053cf725159c63ed274cfba | 153 | 2 (date, cpi) | 2014-01-01..2026-09-01 | 1 | 0 |
| cpi_assembled.csv | ff732b201a525d6fb171d4e8b2d470f986863c1547a1928e42bb197c0777d44d | 957 | 3 (date, cpi, vintage) | 1947-01-01..2026-10-01 | 1 | 0 |
| README.md | ac136bff13bbdb5afac60e0404212ea1f73f992bd5cca1c1e7a6599df78f8965 | - | - | - | - | - |
Units: index 1982-84=100, monthly, SA. SHA-256 of files match bundle.json; they differ from README 'raw download' hashes because the files were reformatted (documented in README).
Findings
- Vintage 2016-07-15: 834 rows 1947-01..2016-06, contiguous, no missing, no duplicates, no jumps >3% m/m.
- Vintage 2026-10-28: 153 rows 2014-01..2026-09, contiguous; 2025-10 cpi empty (source reports no value; consistent with the 2025 US federal shutdown when the October 2025 CPI was not published). Handling: g(2025-10) not computable; reported missing. g(2026-10) would also need it but is outside the window.
- Overlap 2014-01..2016-06 (30 months): values differ between vintages (abs diff mean 0.19, max 0.455 index points) due to seasonal-factor revisions. Rule: new vintage used.
- 12-month ratios for 2014-01..2014-12 use a new-vintage numerator and an old-vintage denominator (2013 months not in new extract). Prescribed by the rule; documented as a data issue.
- cpi_assembled.csv contains assembly errors: (a) 2014-01..2016-06 taken from the old vintage instead of the latest; (b) 2016-07 row missing and every new-vintage value from 2016-07 onward is labelled one month late (e.g. 240.101 = new-vintage 2016-07 labelled 2016-08; the empty value labelled 2025-11 instead of 2025-10); (c) a spurious 2026-10 row carrying the 2026-09 value. Handling: assembled file not used for computation; series rebuilt from the two authoritative vintage files.
Information-availability
No input violates the rules per se; the assembled file would violate the latest-vintage rule if used, so it is excluded from computation.