[bfs2021]

Problem

The weekly link check reports a 502/503/504 as a dead link (site/lib/link-check.mjs:61). When one big host has a short outage, every entry on it lands under "Dead": the 2026-11-09 run (#274) listed 70 github.com entries as dead from HTTP 504s, and a recheck the same day returned 200 for all 70. Fixes #275.

Root cause

classify() treats every status from 400 up as permanent. A gateway answer means the host could not answer at that moment, not that the page is gone, so the report cannot tell an outage from a removed source.

Fix

• classify() returns a new unavailable kind for 502/503/504, before the >= 400 branch. 404/410 and every other 4xx/5xx answer still classify as dead.
• site/check-links.mjs lists those entries under the existing Could not check section (with the API skips) and keeps them out of problems, so an outage no longer opens entries that need a licence re-read.
• The script's header comment and the run summary count now name the transient case.

The optional delayed retry from the issue is not included: the existing immediate retry already covers a blip, and a 30 s wait per entry would lengthen the run without changing the classification.

Testing

• node site/lib/lib.test.mjs: 1710 assertions pass. New cases: 502/503/504 return unavailable (all three failed with "dead" on the old classify), 410 still dead, and a 502 detail names the status.
• npm test passes (274 + 1710 assertions), npm run lint reports "66 scripts parse, no unused imports".

This change was drafted by an agent; the classification and the report section were read in the source before submitting.

[TMHSDigital]

Thanks! Closing in favour of #300, which fixed the same classification and also covers licence pages and the retry. Your write-up of the cause was spot on.