Zenodo curation policy compliance across the register

The register rendering now audits every Zenodo-archived certificate against the [CODECHECK community curation policy](https://zenodo.org/communities/codecheck/curation-policy) (`codecheck` 0.25.0.9002, `make zenodo_check CERT_ID=…` for a single record). The first full run over the register gives the baseline below.

**55 certificates are archived on Zenodo. 7 comply with the curation policy, 48 do not.**

Compliant: 2025-004, 2025-018, 2025-022, 2026-001, 2026-018, 2026-019, 2026-023.

### Findings by requirement

| Requirement | Records affected |
| --- | --- |
| related work (`Reviews` to paper: 37, repository: 28) | 65 |
| machine-readable certificate missing | 48 |
| publisher not `CODECHECK Community on Zenodo` | 45 |
| alternate identifier (`other` scheme) missing | 41 |
| title not spelled `CODECHECK Certificate` | 41 |
| alternate identifier (`url` scheme) missing | 39 |
| licence not CC (17 `other-open`, 2 missing) | 19 |
| no keywords set | 18 |
| resource type not `publication-report` | 7 |
| codechecker recorded as organisation, not person | 6 |
| language not set | 6 |
| no PDF in the deposit | 1 |

### Mechanical corrections

Most findings have a target value that follows from the certificate ID or from `codecheck.yml`, so they can be corrected without judgement: publisher, title capitalisation, both alternate identifiers, language, resource type, the `Reviews` relation where `paper.reference` is a DOI, and the repository relation where the repo is under `codecheckers/`. A dry run over all records says **48 of 55 would be corrected**. These are being applied now; a summary follows in a comment.

### Needs human input

These will be handled one at a time, each with its own comment:

1. **Paper reference is not a DOI (9 records)** — the `Reviews` relation is required, but there is no DOI to point at:

| Certificate | Current `paper.reference` |
| --- | --- |
| 2020-008 | cmmid.github.io PDF |
| 2020-009 | cmmid.github.io PDF |
| 2021-001 | arxiv.org/abs/2104.00718 |
| 2022-001 | iiep.unesco.org publication page |
| 2024-003 | arxiv.org/pdf/2310.19997 |
| 2024-017 | pure.tue.nl PDF |
| 2024-025 | pure.tue.nl PDF (same as 2024-017) |
| 2025-002 | aclanthology.org/2022.lrec-1.328/ |
| 2025-025 | a PDF inside the codecheck repo itself |

2. **Codechecker recorded as an organisation (6 records)** — 2024-003, 2024-017, 2024-018, 2024-019, 2024-020, 2025-028. Not safe to bulk-fix: some entries genuinely are groups, e.g. "Delft 2024-05 participants" in 2024-003.
3. **Title carries extra text (4 records)** — 2024-025, 2025-001, 2025-002, 2025-005 have titles like `CODECHECK certificate 2025-005 for "Verification Report of …"`. Correcting these means deciding whether the extra text is kept.
4. **Repository not under `codecheckers/` or `cdchck` (3 records)** — 2024-023, 2025-005, 2025-025 point elsewhere (Zenodo).
5. **Licence (19 records)** — relicensing a published record is a rights decision for the codechecker, not a metadata fix.
6. **2020-001 has no PDF in the deposit** — needs investigation.
7. **Keywords (18 records)** — the policy says "as you see fit", so this is discretionary.