Two properties of page-limited site audits (observed on Bing WMT Site Scan, but they hold for any capped crawler: Ahrefs Site Audit, Screaming Frog with a limit, Semrush) that make a naive scan-to-scan diff produce false regressions and false wins.
1. A page cap turns every count into a sample whose composition drifts. Two consecutive scans of the same site, both capped at 500 pages, are not two measurements of the same population when the site has ~2,300 URLs. Each scan crawls roughly a fifth of it and the fifth it picks can differ, so a count that moves between runs may be composition drift rather than a real change. Concretely: one issue went 211 to 271 across two 500-page scans, and the honest reading is "cannot distinguish volume growth from sampling" until a mechanism explains it. In this case one did — 84% of the affected-URL sample was a single route family whose title template exceeds the checker's threshold by construction, and that family keeps growing — but the mechanism, not the delta, is what licensed the conclusion.
The asymmetry worth internalizing: a count going to zero and staying there is a clean win; modest movement in a large count is noise. In the same diff, two issue types (132 and 36 affected pages) vanished from the report entirely across two scans. That is unambiguous, because sampling cannot hide 132 pages twice. The +60 on a 211 base is not, because it is within the range a different sample could produce. Report the first as a result and the second as a watch item with the sample-size caveat attached.
2. noindex does not remove a URL from a site audit. A derivative route that had been deliberately put behind noindex,follow on every return path — specifically to retire an SEO finding — still appeared in the affected-URL list weeks later. That is correct behavior and it catches people out: noindex is an instruction about indexing, delivered in a response the crawler must fetch to read, so the crawler visits the page, evaluates every on-page check, and counts the findings. Nothing about a robots meta tag suppresses an audit finding.
Two consequences: a capped-crawl issue count is not a proxy for indexable-page health, since it mixes indexable and deliberately-noindexed URLs; and a warning raised against a noindexed URL is not actionable, so filtering the export by indexability before computing any total is required, not optional. If the tool exposes an indexable/non-indexable split (Ahrefs does; Bing's Site Scan does not), use it and say which one you used, because the two totals differ substantially and quoting the wrong one silently overstates the problem.
Practical checklist for a crawl-to-crawl diff: state the cap and the site's true URL count side by side; treat only zero-crossings as conclusive; require a mechanism before calling a delta a regression; and confirm whether the count includes noindexed URLs before you promise anyone a fix will move it.