Only 15 of 56 corpus pre-prints are full-reuse; 9 grant no reuse at all

[lfnothias]

Surfaced by the new pre-print licence resolver. OPEN_ACCESS_POLICY.md used to assert that pre-prints are "always CC-BY by repository policy". They are not, and the corpus shows it.

Reproduce:

python -m scripts.preprint_license --corpus 'collections/*/v*/corpus.yaml'

What the corpus actually contains

Of 600 DOIs across all collections, 56 are pre-prints (Crossref posted-content or DataCite Preprint). Their real posting licences:

Source reuse	Count	Licences
full — redistribute, quote, derive	15	CC-BY-4.0 (14), CC-BY-SA-4.0 (1)
limited — NC and/or ND; link-only	32	CC-BY-NC-ND-4.0 (21), CC-BY-NC-4.0 (6), CC-BY-ND-4.0 (4), CC-BY-NC-SA-4.0 (1)
none — no reuse rights at all	9	bioRxiv cc_no (8), arXiv nonexclusive-distrib (1)

So only 15 of 56 pre-prints meet the open-access bar the old rule assumed for all of them.

The nine with no reuse rights, all status: included

10.1101/2020.10.30.362533                 NoReuse-1.0
10.1101/2023.02.26.530140v1.article-info  NoReuse-1.0
10.1101/2024.05.30.596727v2               NoReuse-1.0
10.1101/2024.08.17.608400v2               NoReuse-1.0
10.1101/2024.10.07.617094v3               NoReuse-1.0
10.1101/2024.10.31.621429v1               NoReuse-1.0
10.1101/2024.11.13.623458v1               NoReuse-1.0
10.1101/2025.10.15.682422v1               NoReuse-1.0
10.48550/arxiv.2507.06853                 arXiv-1.0

NoReuse-1.0 is the bioRxiv cc_no token: the server displays the pre-print, the author reserves all rights.

Why this is not (yet) an emergency

Every one of these entries carries access.type: repo-oa, i.e. it was admitted on the repository axis, not on the paper's licence — access.license holds the tool's code licence (MIT, GPL-3.0…), not the pre-print's. So nothing was admitted because of the blanket pre-print rule. The exposure is that the pre-print text itself has no recorded reuse basis, and any grounding that quotes it at length would rely on rights we do not have.

This is the same conflation tracked in #23 (repo-oa asserts availability, not licence).

What needs deciding

1. Whether a repo-oa entry may ground on its paper text at all, or only on its repository. If the former, the 41 non-full-reuse pre-prints need access.license corrected to the paper's licence and their grounding restricted to link-only.
2. Whether the 9 no-reuse entries stay included. They have no reuse basis for the paper under any reading.

Separately: 24 malformed DOIs

The resolver could not identify these because the doi field holds a scraped URL, not a DOI. A few examples:

10.1093/bioinformatics/btab060/6125380      <- trailing article id
10.1021/acs.analchem.1c03163?ref=           <- query string
10.1007/s10858-025-00465-y#sec2             <- fragment
10.5281/zenodo.1043226.svg                  <- a badge *image* URL
10.1111/jipb.13774-cyan                     <- badge colour suffix
10.1016/1044-0305                           <- truncated

10.5281/zenodo.363932874 is the known 404 DOI already tracked elsewhere. These should be normalised at ingest; until then every consumer that resolves a DOI will fail on them.