Creative Commons Attribution 4.0 International (CC BY 4.0)

Copyright (c) 2026 Theta Health

The benchmark **data** in this repository is licensed under CC BY 4.0.
Full text: https://creativecommons.org/licenses/by/4.0/legalcode

You are free to share and adapt the material for any purpose, including
commercially, provided you give appropriate credit, link to the license, and
indicate if changes were made.

================================================================================
WHAT COUNTS AS "DATA" HERE
================================================================================

Covered by this file (CC BY 4.0):

  inputs/*.job.yaml          task specifications
  registry/*.yaml            clinical vocabularies, findings, scoring profiles
  probes/**                  question framings
  frozen/*.Q.jsonl           frozen reference packs -- the solver-visible half
  frozen/MANIFEST.json       freeze record for those packs
  docs/**                    documentation
  docs/anchor/*.json         freeze snapshots
  results/** reports/**      generated cases, model responses, per-cell
                             judgements and rendered reports that you produce
                             by running the pipeline on this data

Covered by `LICENSE` (MIT) instead:

  haenv/**  tools/**             the pipeline, judges and self-checks
  verifier_core/**               domain-independent verifier primitives
  frozen/split_wq.py             the Q/W splitter (code, not data)
  config.yaml  pyproject.toml  uv.lock  .gitignore   config and build files

================================================================================
RESIDUAL RULE -- nothing is left unlicensed
================================================================================

  A file is under CC BY 4.0 if it is listed above, or if it is benchmark
  **data** of the same kind: a generated case, a task specification, a model
  response, a per-cell judgement, a rendered report, a freeze snapshot, or a
  clinical vocabulary.

  Everything else -- source code, build and config files -- is MIT
  (`LICENSE`).

  `core/**` is governed by its own licence file, not by either of ours.

================================================================================
THIRD-PARTY CONTENT -- NOT COVERED BY THIS GRANT
================================================================================

Nothing in this repository transcribes published third-party case reports.

================================================================================
ATTRIBUTION
================================================================================

If you use the data, cite the repository and state the **freeze revision** the
numbers came from -- every published board is pinned to one:

    haenv benchmark (Theta Health, 2026), freeze rev<N>,
    judging fingerprint <judging_sha16>.

Both values are printed on every board and stored in
`docs/anchor/freeze-*.json` and in each `eval.jsonl` row (`judging_sha16`).
A number without them is not reproducible -- the judging code changes, and a
board produced under a different fingerprint is a different measurement.

================================================================================
THE DATA IS SYNTHETIC. IT IS NOT MEDICAL ADVICE.
================================================================================

Every patient in this benchmark is generated. No row describes a real person.
See `docs/ETHICS.md` for what that does and does not license you to conclude,
and for the one place where real-EMR-derived material was used (calibration of
baseline value ranges) and how it is handled.

Do not use this data, or models evaluated on it, to make clinical decisions.
