# Data audit (P1)
| file | SHA-256 | rows x cols |
|---|---|---|
| data/design.csv | f4ff004ed50d893eece1aaa7edced04d3c0439aaedf1d391ea4642a87d0ecc50 | 9 x 11 |
| data/README.md | 73a6595909d92a69db281e283f193348ac244f255fcd5343ef42ae0353394e67 | documentation |

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, all 2000), B (int, all 399), seed (int, 20261101..20261109, all distinct), mu (0.0), total_variance (1.0), nominal_level (0.95). Units: dimensionless (simulated y with unit total variance). No dates.

Checks run: missing values = 0; duplicate rows = 0; duplicate seeds = 0; for every cell len(cluster_sizes)==G, sum(cluster_sizes)==N, sizes follow 8,16,32,64 repeated pattern (all True). Σn_g² = 5504 (G=5), 11200 (G=10), 38400 (G=30).

Anomalies / notes: none in the file. Methodological note (not a data error): with G=5 Rademacher weights there are only 2^5=32 distinct weight vectors (16 distinct |t*| values since |t*| is invariant to w→−w), so B=399 draws heavily repeat and the 0.95 quantile of |t*| is coarse. Recorded in results `data_issues`.

Information availability: the design file is synthetic and fully allowed; no violation. No exclusions are needed (all 9 cells × 2000 replications simulated); exclusions.csv records a single no-op step.