Commit History
Today
-
Keep only v3 and make choice_v3 the default dataset
Verified signatureRemove 19 obsolete original/v2 dataset files and all six old configurations, plus superseded cards, catalogs and audits. Keep all 23 v3 Parquet files unchanged. Move inherited input evidence under v3, preserve source license notices and attribution, and update the card, provenance links and loading checks. Default loading, explicit v3 views, rejected obsolete configurations, document links and unchanged data hashes were verified before this atomic commit.f5396ab
Yesterday
-
Repair v3 supervision, native tool context, and evaluation isolation
Verified signatureAdd explicit v3 choice and native auxiliary configurations. Regenerate semantic pairs, restore Hermes requests and call sets, recover Nemotron schemas, restore legacy validation, mask unsupported NLU++ labels, and preserve frozen default/v2 data. All new configs passed structural, lineage, encoder and datasets loading checks.0e1ce34 -
Add curated v2 semantic, tool-context and NLU++ configurations
Verified signatureAdd choice_v2 with 116,440 training, 8,206 validation and 24,639 test records, plus separate semantic, tool-context and multi-label views. Preserve the four original data files and default config. Include pinned source attribution, licenses, parent isolation, paired no-match construction, and validation audits.11e6f65