# Data audit

• File: data/panel.csv, SHA-256 968410b993df4d696194ba82b35a030d9c4f5f69e3c2de2c16c7b304d55ce5a8
• 2000 rows x 3 columns: firm_id (int64), x (float64, dimensionless), y (float64, dimensionless). No dates (not a time-indexed panel; no within-firm time variable).
• Missing values: 0 in every column. Exact duplicate rows: 0. (One repeated x value and one repeated y value across different rows — coincidental, not duplicate records.)
• firm_id 0..49: 50 firms, exactly 40 rows each (balanced).
• x: mean 0.161, sd 1.325, range [-4.34, 4.99]; y: mean 1.573, sd 1.782, range [-3.50, 7.20]. No implausible values.
• Strong within-firm clustering: share of variance of firm means ≈0.45 for x and ≈0.53 for OLS residuals → clustering matters.
• Information-availability: only input is panel.csv; no violations.
• Exclusions: none needed (ledger in out/exclusions.csv, all n_removed=0).