How PEP 740 attestations spread among actively releasing projects in the top 8,000 PyPI packages by downloads, October 2024 to November 2026: a pre-registered data note
PyPI announced support for PEP 740 attestations on 2024-11-14; the earliest attested upload seen in a pre-study sample is from 2024-10-18. For projects that publish through Trusted Publishing with the canonical GitHub publish action, attestations are generated by default, so an attested release need not reflect a decision by a maintainer to add provenance. Under a plan committed before any outcome was computed, we took a download-ranked frame fixed before launch (hugovk/top-pypi-packages, release 2024.10, 8,000 projects), kept the 5,015 projects that uploaded a file in the year before 2024-10-18, and read upload times and per-file provenance from the PEP 691 JSON simple index on 2026-11-15. Among GitHub-hosted eligible projects (74,599 release opportunities, 1,364 first attested releases), a release of a project ranked 1 to 500 had higher odds of carrying the project's first attestation than a release of a project ranked 2,001 to 8,000 (odds ratio 1.93, 95% CI 1.30 to 2.86), adjusting for calendar quarter; unregistered post hoc refits give top-tier odds ratios from 1.62 to 2.14. By 24 months after 2024-10-18, 28.8% (27.5% to 30.1%) of GitHub-hosted eligible projects had an attested release (Kaplan-Meier): 47.0% of the top tier and 23.8% of the bottom tier. The monthly share of post-launch releases with an attested file rose from 12.3% in 2024-11 to 34.5% in 2026-10. The registered comparison by repository host could not be made, because by the registered rule it needs the date PyPI began accepting non-GitHub attestations, and PyPI does not state it. The results are associations, not causes. The record holds the note (PDF and Markdown), the registered plan (verbatim) with its manifest, the deviations log (redacted and annotated), package-level result tables, the results file, the collection and analysis code (local paths and an internal commit id replaced), the independent review of the draft (redacted), and a trace of every number in the note with the script that checks it; no publisher identity was stored. Analysis and text by Claude (an AI model by Anthropic) for Glasp; an independent AI recomputation matched for the eligible counts, events, ever-attested shares, opportunities, publisher kinds, first non-GitHub dates and the primary and host-indicator logistic models; no human has checked the figures.
Authors
- Keisuke Watanabe (ORCID: https://orcid.org/0000-0001-6374-3918)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24251577
- Primary Topic
- Software Engineering Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00