Data-changing release relative to v1.0.2: the deterministic dataset selection seed is changed from 20260815 to 0. Raw data, source revision, difficulty definitions, split quotas, schema, leakage controls, validation rules, and core dataset description remain unchanged. All four processed splits are deterministically regenerated and independently verified.
Documentation-only release relative to v1.0.1. Adds a pinned, source-grounded Tdoku difficulty definition, quantitative raw-pool audit, source-confounding analysis, and reproducibility limitations. The seven raw CSV files and four processed Parquet splits are byte-identical to v1.0.1.
Publish v1.0.1 raw snapshot and reproducible build process
Twist Shan
Verified signature
f9d92af
Add the complete 1,415,420-row raw candidate pool used by the v1 protocol, raw provenance and license notices, exact build and verification scripts, and metadata manifests. Processed Parquet split contents remain byte-identical to v1.0.0.
Publish v1.0.0 formal hard-Sudoku splits
Twist Shan
Verified signature
84189a5
Train on balanced original, r0, and r1_4; reserve r5_19 for OOD evaluation. Includes deterministic builder, provenance hashes, and overlap audit.
