Data audit (P1)
| file | SHA-256 | rows | cols |
|---|---|---|---|
| data/design.csv | f4ff004ed50d893eece1aaa7edced04d3c0439aaedf1d391ea4642a87d0ecc50 | 9 | 11 |
| data/README.md | 73a6595909d92a69db281e283f193348ac244f255fcd5343ef42ae0353394e67 | (text) | — |
| Hashes match bundle.json. |
Schema: cell_id(str C1..C9), G(int 5/10/30), icc(float 0/0.2/0.5), cluster_sizes(str '|'-sep ints), N(int), R=2000, B=399, seed(int 20261101..20261109), mu=0.0, total_variance=1.0, nominal_level=0.95. No dates (synthetic design). Missing values: 0. Duplicate rows: 0. Duplicate cell_id: 0. Consistency: for all 9 rows len(cluster_sizes)=G, sum(cluster_sizes)=N (128, 264, 864), and sizes follow 8,16,32,64 repeated.
Anomalies / notes (not errors):
- ICC=0.0 cells: u_g ~ N(0,0) is degenerate (u_g ≡ 0); handled by drawing u_g = sqrt(icc)*Z, which is exactly 0.
- G=5 with Rademacher weights: only 2^5=32 distinct weight vectors, and |t*| is invariant to flipping all signs, so at most 16 distinct |t*| values; the B=399 bootstrap quantile is therefore highly discrete (known WCB limitation with very few clusters). Kept as specified; reported in data_issues.
- Unequal cluster sizes (8..64) make cluster-size heterogeneity part of every cell. Information availability: only design.csv is used; no violations possible (no observed data). Exclusions: none (all 9 cells × 2000 replications simulated); ledger in out/exclusions.csv.