Data audit (P1)
| file | SHA-256 (as found in data/) | rows | cols | date range | missing cpi | duplicate dates | gaps |
|---|---|---|---|---|---|---|---|
| cpi_vintage_2016-07-15.csv | 281fbcdd6a6eeb3e9cb7c1f20476e6bd1f7b925bf140606fe46538fe83f2c83e | 834 | 2 (date, cpi) | 1947-01..2016-06 | 0 | 0 | none |
| cpi_vintage_2026-10-28.csv | 8220d5b4d70ca32fc2c7b1975651c515967db8a87053cf725159c63ed274cfba | 153 | 2 (date, cpi) | 2014-01..2026-09 | 1 (2025-10) | 0 | none |
| cpi_assembled.csv | ff732b201a525d6fb171d4e8b2d470f986863c1547a1928e42bb197c0777d44d | 957 | 3 (date, cpi, vintage) | 1947-01..2026-10 | 1 (labelled 2025-11) | 0 | 2016-07 missing |
| README.md | ac136bff13bbdb5afac60e0404212ea1f73f992bd5cca1c1e7a6599df78f8965 | — | — | — | — | — | — |
Note: README SHA-256 values refer to the raw downloads before reformatting, so they do not match the reformatted CSVs here (expected per README).
Units: index 1982–84=100, SA, monthly, first-of-month dates. 2016 vintage cpi range 21.48–239.927.
Anomalies:
- 2026 vintage: empty cpi for 2025-10 (source reports no value). Older vintage does not cover it -> P_2025-10 missing; g_2025-10 missing. g_2026-10 would also need it but lies outside the window.
- Overlap 2014-01..2016-06 (30 months): levels differ by up to 0.455 points (revisions/seasonal factors). Rule: use 2026 vintage.
- cpi_assembled.csv errors: overlap months use the OLD vintage (30 months mislabelled vs. rule); 2016-07 omitted; from 2016-08 onward every value is shifted one month later (e.g. labelled 2016-08 = true 2016-07 value 240.101; labelled 2026-10 = true 2026-09 value 336.298); hence the missing value appears as 2025-11. 151 of 955 shared months have a wrong level. Decision: the assembled file is read for diagnostics only (counted in n_input_rows); all estimates come from the two vintage files. No input violates the availability rules beyond these assembly errors.