Remove original-label subset Parquets and redundant easy-first variants from the active dataset; retained subset data stays byte-for-byte unchanged. Preserve historical JSONL and manifests, with prior revision linked for recovery.
Add 177 static candidates and source-corrected operation labels
Preserve all original reference bytes and archive files; add corrected-label views, pinned provenance, raw upstream labels, mixed-license attribution and explicit unvalidated runtime/speed status.
8 days ago
Preserve exact seed42 curricula with Parquet configs and original JSONL
