CHANGELOG.md

Changelog

1.0.0 — deterministic public release

• Published the 51,389-row SOTA merge release with sft as the default Hugging Face configuration.
• Added 11 explicit Parquet configurations and 64 byte/hash-checked shards.
• Added tokenizer freeze metadata, schema and provenance documentation, quality disclosures, training guidance, citation metadata, and reviewed reference assets.
• Recorded mixed source rights as license: other and documented noncommercial scope, service terms, benchmark-derived development holdouts, and contamination limits.
• Kept the public package no-code: no loading script, no raw admission records, and no preference-shaped configuration or field.