# Data audit (P1)

File: data/design.csv — SHA-256 b1a17a529973e7994bbedad5ec1bc73fe968c777301ca8745657ea300e3c74f0 (matches bundle.json).

• 30 rows × 3 columns: i (int64, 1..30, unique, contiguous), x (float64, dimensionless, range -1.3942..1.9919, mean 0.2823, sd 0.8212), sigma (float64, dimensionless, range 0.6034..1.7903, all > 0).
• Missing values: 0. Duplicate rows: 0. Duplicate i: 0. No dates (fixed synthetic design).
• Anomalies: none. Note: sigma is (almost exactly) a deterministic function of x^2 (corr 0.99999999), i.e. variance rises with |x| — the classic heteroskedasticity pattern that makes conventional OLS SEs too small. This is a feature of the design, not a data problem.
• Information availability: the file is the only permitted input; no violations. x and sigma are used as-is, never redrawn.
• Exclusions: none (see out/exclusions.csv).