Skip to content

Pin Wayback snapshots of the FrontierCO and PGLib sources #324

Open

Description

@Mearman

The sources of huggingface-frontierco-mis, huggingface-frontierco-mds, huggingface-frontierco-stp, pglib-opf-typical, pglib-opf-api and pglib-opf-sad (added in #323) have no Wayback snapshot, so pnpm run archive:check lists them as unaccounted for. A run of scripts/archive-sources.ts over the FrontierCO entries had every fresh capture refused: the raw snapshot answered HTTP 404 with the Wayback Machine's page when read back. The PGLib sources (the repository, its zip at commit dc6be4b and BASELINE.md) were not attempted in that run.

Deferred because the Archive was not serving the new captures at the time; the entries fetch from the live sources meanwhile.

Revisit trigger: the next archive-sources.ts run, or the Archive serving raw captures of huggingface.co/.../resolve/... URLs again. Done when archive:check reports none of these sources.

Activity

  1. added
    deferredTracked and deferred, with a revisit trigger in the body
    on Nov 11, 2026
  2. Mearman commented on Nov 11, 2026

    @Mearman
    OwnerAuthor

    Status after a second archive run on 11 November 2026, for the three new FrontierCO entries (independent set, Steiner tree, dominating set).

    The PGLib-OPF sources are now pinned and checksummed, so this issue is down to FrontierCO.

    Every FrontierCO file tried was refused the same way: Save Page Now accepts the capture, and the raw id_ form of that capture then answers HTTP 404 with the Wayback Machine's own page, so the script rejects the pin. 21 of 21 attempts in this run, none pinned. The source URLs are Hugging Face resolve/<revision>/... links, which redirect to a signed storage URL; the capture appears to record the redirect and not the bytes. The dataset's landing page is archived normally.

    I stopped the run there, since each attempt spends one of the 200 anonymous captures a day and the outcome did not vary. archive:check lists these sources as unaccounted for, as it does the existing facility-location entry's.

    Retrying will not help until one of these changes: the Wayback Machine serves raw bytes for these redirects, the files are pinned from a mirror that serves them directly, or the script gains a route for sources behind a redirect (for example capturing the resolved storage URL and recording the mapping).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    deferredTracked and deferred, with a revisit trigger in the body

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions