Source-link audit

Corpus-wide probe of every URL cited in a dossier ## Sources section. Standing gate: docs/audits/rolling/2026-W34/03-claim-vs-source.md#F5. Re-run: npx tsx scripts/audit-source-links.ts [--batch=N] — each run probes the N oldest-checked (or never-checked) URLs and persists results to docs/intelligence/audits/source-link-status.jsonl, so full corpus coverage accrues over many runs rather than one slow/abusive sweep. NOT wired into a cron gate yet — land the script, read one honest run, then decide cadence (matches the convention set by audit-outlook-claims.ts).

Last run: 2026-10-06T10:59:05.451Z (batch size 60, probed 21 this run)

100% of the 6688 distinct cited URLs have been checked at least once.

blocked-from-host and unroutable are deliberately NOT counted as dead-at-origin — a 403/429/202/empty-body/timeout means this host is blocking or challenging THIS prober, not that the page is gone. Collapsing them would repeat the absence-as-signal defect this audit exists to catch. Since 2026-08-26, a 2xx response with a near-empty or challenge-page body (status-code-only checking's blind spot — e.g. an Incapsula challenge page) is also blocked-from-host, not ok.

Confirmed dead-at-origin (404/410)

Blocked/unroutable but a Wayback snapshot exists — resolve these first

An archived copy is a click away for these; use it to confirm the citation still supports its claim (and consider citing the archive.org URL instead).

Redirected to a different page (still ok, but final URL != cited URL)

Status alone says these are fine — verify by hand that the page actually landed on still supports the cited claim (nifco.md #76 did not: it 301'd to a generic index with none of the enumerated products).