Reader-level annotations and analysis code for: Binary Referable Evaluation Conceals Half the Annotation Error in the APTOS 2019 Diabetic Retinopathy Benchmark

Reader-level annotations and complete analysis code supporting a grade-stratified audit of annotation error in the APTOS 2019 Blindness Detection dataset. Contents: release/ — A de-identified reader-level dataset covering the 190 sampled images: the dataset label, the sampling stratum, each of three retinal specialists' independent ICDR grades (0-4, or U for ungradable), the adjudicated reference grade, and a duplicate-pair flag. Fundus images are not redistributed; they are obtained from Kaggle under the original terms. scripts/ — Sixteen Python scripts reproducing every reported value, including the design-weighted estimator, the Dirichlet-Jeffreys posterior used for interval estimation, the coverage simulation, the duplicate audit, and two independent verification scripts. REFERENCE_STANDARD_RULE.md — The adjudication rule, fixed in writing on 31 July 2026 before the deciding reads were collected, with its original timestamp intact. PROTOCOL_AMENDMENTS.md — Dated log of every deviation from the pre-specified analysis plan. Reproduce the headline table with no images and no withheld files: python3 scripts/final_intervals.py --source release Verify the hand-written statistics against scikit-learn and NumPy: python3 scripts/validate_against_reference_impls.py The analysis uses only the Python standard library. scikit-learn, NumPy and SciPy are required for the verification scripts alone, not to reproduce any reported result.