Corrections & retractions

Retracted: 22 archive links offered as independent evidence, and 3 claims about sites refusing the Internet Archive. 21 links pointed at web.archive.org/web/<url> with no timestamp, which resolves to the Archive's most recent capture — so a link offered as proof of what a page contained would have shown the page as it is now; on a removal ledger that is precisely backwards. One (trumpaccounts.gov) pointed at a capture of a 403 block page. Separately, three records stated that getactive.gov, moms.gov and techprosperitycorps.gov refused the Internet Archive while serving Daylight's own request normally: each clause was true and the impression was not, since those servers return the same 403 to any non-browser client and Daylight's capture succeeds only because it drives a real browser. Archive links are now pinned to a capture instant and checked against the Archive's public index, and the comparison with our own request is measured rather than implied.

Retracted: 52 page baselines that were not captures of the page, and the 4 high-severity claims derived from them. 48 were a block page — eleven sites, including state.gov, hhs.gov, fcc.gov and usda.gov, refuse Daylight's crawler with HTTP 403, and the refusal was recorded as their homepage: 'no trackers, no privacy notice, no agency seal'. That was false and flattering; fcc.gov, state.gov and usda.gov shared a single DOM hash, which three federal homepages cannot. The other 4 were browser errors, published as 'studentaid.gov now redirects off-domain to chrome-error://chromewebdata/' and the same for state.gov, usda.gov and fcc.gov — those pages simply failed to load. Daylight now records a page only when the browser receives a real 2xx document, and lists pages it cannot capture as exactly that instead of inventing a baseline for them.

Retracted: every change Daylight had published about a watched page — 431 events (Receipts: 45 removals, 71 tracker additions, 38 privacy-notice edits; Floodlight: 264 tracker changes). They described Daylight's crawler, not the sites. Two defects. The page capture waited 8 seconds for a page to finish loading and, when that timed out, inventoried whatever had arrived — so trackers that had not yet fired were published as removals; healthcare.gov's tracker count read 3, 1, 3, 3, 3, 12, 3 across seven captures and produced twelve dated 'tracker removed' findings. And a tracker was identified by its hostname, which rotates for reasons that mean nothing: Microsoft Clarity answers from a-z.clarity.ms, so each capture 'removed' the previous shard and 'added' the next — reported as sba.gov adding Clarity three times. Twenty of the removals are provably false: the thing said to be gone is present in a later capture. A capture now records whether the page finished loading, and no change is published unless both observations were complete; a tracker is now identified by its vendor. Nothing was removed from any of these sites.