# 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 (C1..C9, unique), G (int: 5,10,30), icc (0.0,0.2,0.5), cluster_sizes ("|"-separated ints), N (int), R=2000, B=399, seed (20261101..20261109, distinct), mu=0.0, total_variance=1.0 (unitless; variance of y), nominal_level=0.95. No dates (synthetic design). No missing values, no duplicate rows.

Consistency: for every row len(cluster_sizes)=G and sum(cluster_sizes)=N (128, 264, 864) — asserted in analysis.py. Sizes follow the pattern 8,16,32,64 repeated (G=5: 8,16,32,64,8; G=10: ...,8,16; G=30: 7 full patterns + 8,16). Clusters are strongly unbalanced (one 64-obs cluster carries half of N at G=5).

Anomalies / issues: none in the file. Methodological note recorded in results `data_issues`: at G=5 Rademacher weights have only 32 distinct sign patterns, so the WCB distribution is discrete and coarse.

Information-availability: design.csv contains only design parameters (no outcomes); nothing violates the rules. Exclusions: none (out/exclusions.csv, 9 -> 9).