Publish the complete frozen base (three verified shards), tokenizer/processor assets, unchanged A2000 adapter and B2000 branch. Retain original training history and source attribution. Implementation code is hosted on GitHub. Remove the optional affiliation disclaimer.
Unchanged A1613 and B2000 weights, model card, Apache-2.0 license, attribution, and SHA256SUMS. Code and evaluation evidence remain on github.com/Quinn-Ma/OriginX.
