# Data audit

• data/design.csv — SHA-256 b1a17a529973e7994bbedad5ec1bc73fe968c777301ca8745657ea300e3c74f0
• 30 rows × 3 columns: i (int 1–30), x (float, dimensionless, range −1.3942 to 1.9919, mean 0.2823), sigma (float, dimensionless, range 0.6034 to 1.7903, all > 0).
• Missing values: 0. Duplicate i: 0. No dates (synthetic fixed design).
• Note: σ_i increases with |x_i| (smallest near x≈0, largest at extreme x), so high-leverage points have the largest error variance. This is the structure that makes the conventional SE too small. It is a feature of the design, not a data defect.
• Information-availability: there is only one input, it is allowed, and no violations were found.
• Exclusions: three checks (missing, σ≤0, duplicate i) removed 0 rows; see out/exclusions.csv. n_raw=30=n_analysed.