Test fixtures copied from neurotic_docx_bench
=============================================

The files under tests/corpus/neurotic_docx_bench/ and
tests/corpus/_to_improve_accepted_changes/ are test inputs copied from the
neurotic_docx_bench benchmark (the public dataset
arthrod/neurotic_docx_bench-fixtures on the Hugging Face Hub, ODC-By 1.0) so
that jubarte's tests do not depend on a checkout of the benchmark. They keep
the relative paths they have in the benchmark. Jandira Technologies, LLC does
not license them under this repository's AGPL-3.0-only license, and nothing
here grants rights in them. They are used only as test inputs.

Where they come from:

- tests/corpus/neurotic_docx_bench/grok_run/fixtures_500/ and the
  hash-named documents under corpus/word/: documents from docx-corpus
  (superdoc-dev/docx-corpus, https://huggingface.co/datasets/superdoc-dev/docx-corpus),
  and Microsoft Word's saves, compares, Accept All and Reject All of them.
  docx-corpus is made available under the ODC Attribution License
  (ODC-By v1.0, https://opendatacommons.org/licenses/by/1.0/). Cite the
  dataset as superdoc-dev/docx-corpus. Built by SuperDoc. ODC-By covers the
  database (the collection, its selection and arrangement). It does not
  license copyright in the individual .docx files, which were collected from
  the public web and may have their own rights (section 2.4 of the license).
- corpus/word_redlines_superdoc/ and the super_editor__, behavior__, evals__,
  doc_api_stories__ and cli_legacy__ documents: test documents of SuperDoc
  (https://github.com/Harbour-Enterprises/SuperDoc, AGPL-3.0), and Word's
  compares of them.
- corpus/word_based/, grok_run/word_based/,
  grok_run/no_comments_pdf_was_generated_by_word/ and grok_run/_fixtures/:
  the jubarte-first fixture documents (some generated for the benchmark,
  some collected under their original names) and Word's saves and compares
  of them.
- grok_run/wr0928/rejected_tracking/ and tests/corpus/_to_improve_accepted_changes/:
  Word's compares of benchmark pairs, and Word's Reject All or Accept All of
  those compares.
- corpus/word/notices/: the benchmark's own selection tables.
