What is here, and where it came from.

zetlyn-entities-1.json
    The entity recognizer the Open Zetlyn Corpus Specification pins as `zetlyn-entities-1`: a
    per-token BIO classifier over twelve lexical features. Produced by Zetlyn and distributed with
    it. A corpus built with any other bytes is not conforming under that name — which is what the
    SHA-256 beside it is for.

multilingual-e5-small/
    The embedding model, redistributed unchanged from `intfloat/multilingual-e5-small` on Hugging
    Face: https://huggingface.co/intfloat/multilingual-e5-small
    Its model card declares the licence `mit`. The repository carries no LICENSE file, so this note
    states what the card states and nothing more.

    It is served from here rather than fetched from Hugging Face so that rebuilding a corpus does
    not depend on somebody else's bucket still existing. The bytes are the same bytes; the SHA-256
    of each is in SHA256SUMS and can be checked against the upstream copy.

Every file here is fixed under its name. A different model is a different name, never the same name
with new contents — so anything downloaded from here can be cached for as long as you like, and a
hash that stops matching means something is wrong rather than something moved on.