PhysMAP — what each licence covers
==================================

This repository carries two licences, because the code and the data are
different things.

CODE — MIT. See LICENSE.
    Everything under src/, tests/, dev/, examples/, docs/ and protocols/.

DATA — Creative Commons Attribution 4.0 International. See LICENSE-CORPUS.
    The curated corpus as authorship: the seed closure corpus, the closure
    index, the provenance manifests, and the selection, structuring, regime
    mapping and validity-bound assignment throughout.

NOT COVERED BY EITHER — numerical values digitised from third-party
publications.

    This repository contains measurements read off figures and tables in
    published papers, theses and reports. Those measurements were not made
    here, so no copyright in them is claimed; applying a licence to them would
    assert a right that is not ours to assert.

    They are governed by their publishers' terms, by public-domain status (for
    example NACA TN-1451, a US Government work), or they are uncopyrightable
    facts. Every such file carries its own redistribution determination in the
    manifest beside it: the source, the upstream terms, whether those terms
    permit redistribution of derived numerical data, and the decision.

    A citation is attribution. It is not permission.

No source PDF is redistributed. Papers are cited, not copied.

FIVE OF THE SEVEN DATASETS SHIP WITHOUT A LICENCE

    Shipping is not licensing. Two of the seven published datasets carry
    affirmative permission -- naca_tn1451 (public domain) and velazquez_sco2
    (CC BY 4.0). The other five do not:

        marineau_hypersonic_transition   no licence, and no prohibition
        casper_hypersonic_transition     against AIAA terms
        forrest                          against ASME terms
        dirker_water                     against Elsevier terms
        jin_sco2_buoyancy                against Elsevier terms

    All five redistribute NUMBERS ONLY -- no paper, figure or PDF. That is
    enforced by two automated audits and is the foundation the whole position
    rests on.

    Marineau is the only one with no prohibition attached: values transcribed
    from a published table of measurements, in a publicly funded, publicly
    released, government-hosted document.

    The other four are published DESPITE express publisher terms. AIAA
    prohibits using its content to develop machine-learning models. ASME and
    Elsevier require permission to reproduce, and Elsevier's licence forbids
    systematic redistribution of the dataset. Those are risks this project
    accepted knowingly; none of them is a finding that the term does not apply.

A SEPARATE WARNING ABOUT DATA QUALITY

    forrest is published but is NOT benchmark-grade. Its own header says the
    values are visual estimates to be used only as a resolvability triage
    check, and its benchmark cell is degenerate -- one training row, no
    detector fit -- so its DO_NO_HARM outcome is short-circuited rather than
    earned. It ships so the weakness is inspectable, not so it can be cited.

    Being legal to publish and being fit to benchmark on are different
    questions. This file answers only the first.

    data/REDISTRIBUTION.md states the basis for each in full, including the
    arguments against.

REMOVAL ON OBJECTION

    Any rights holder who objects gets the data removed. Open an issue at
    https://github.com/cloudronin/physmap/issues, or contact the maintainer.
    No argument will be made; the files come out.

    Removal reaches this repository and every later release. It cannot
    reach a version already archived: each release is archived on Zenodo
    with its own DOI, and an archived version is a permanent record that
    only Zenodo can withdraw. The packages on PyPI do not include data/.
