Data audit (P1)
| file | SHA-256 | rows | cols |
|---|---|---|---|
| data/design.csv | f4ff004ed50d893eece1aaa7edced04d3c0439aaedf1d391ea4642a87d0ecc50 | 9 data rows | 11 |
| data/README.md | 73a6595909d92a69db281e283f193348ac244f255fcd5343ef42ae0353394e67 | (text) | — |
Hashes match bundle.json.
Schema: cell_id (str C1..C9), G (int: 5,10,30), icc (float: 0.0,0.2,0.5), cluster_sizes (str, '|'-separated ints), N (int), R (int, 2000 all), B (int, 399 all), seed (int, 20261101..20261109), mu (0.0), total_variance (1.0), nominal_level (0.95). Units: y is unitless with total variance 1.
Missing values: 0. Duplicate rows: 0. Duplicate cell_id: 0. Consistency: for every row len(cluster_sizes)==G and sum(cluster_sizes)==N (128, 264, 864). Σn_g² = 5504, 11200, 38400. Date ranges: none (synthetic design).
Anomalies / notes:
- Cluster sizes are unbalanced (8..64); pattern 8,16,32,64 repeated and truncated at G, so G=5 has two clusters of size 8 — as documented in README, not an error.
- With G=5 and Rademacher weights there are only 2^5=32 distinct weight vectors (16 distinct |t*| values since w and −w give the same |t*|); the B=399 draws therefore sample with replacement from a tiny support and the 0.95 quantile is coarse. Not a data error; recorded as a data_issue / limitation, procedure kept exactly as specified.
- No information-availability violations: the only input is the design file; no observed data. No exclusions are applied (exclusion ledger records a zero-removal step).