Source Verification: Read all 7 sources from raw text. Snapshots decompressed from sources/2026-11/<slug>/ and converted to plain text; every one of the 9 double-quoted passages in the body was machine-matched verbatim (whitespace and curly-quote normalised) against the snapshots (13 quoted passages if inline code spans are counted; the code spans were checked by hand against webkit.org and MDN). (0) webkit.org/blog/18541 (snapshot source-0, status 200, sha256 70294844...): byline Simon Fraser et al., dated 'Nov 3, 2026'. Confirms: 'When enabled by a user, Safari adapts light websites and presents them as dark.'; 'Almost 40% of web page loads are from users in dark mode, and yet only 10% of page loads support it.' (WebKit's own figures, scope = web page loads, no methodology given - the article labels them WebKit's and says no method is stated: correct); sites that 'signal support of dark mode with <meta name="color-scheme" content="light dark"> in HTML or color-scheme: light dark in CSS' - 'Adaptive Dark Mode does not change a thing'; for the rest it analyzes colors and estimates predominantly light or dark, dark pages unchanged, light pages adapted; color matrix inverting luminosity while preserving hue; WCAG 4.5:1; light CSS gradients to dark; most images unchanged, SVG inspected, bitmap pixel histograms to invert simple black-and-white logos; masks and mix-blend-mode turned off in some cases; canvas left unchanged; privacy (JavaScript gets original authored colors, canvas images render in original form); advice to declare color-scheme in the head and/or on :root or body and not inject it via JavaScript; element-level 'color-scheme: light dark' for part of a page; opt-out via meta content="light only" or CSS color-scheme: light only, with the note '(Note: the 'only' keyword is not supported in Safari 27.3 beta 1. Check back as Adaptive Dark Mode evolves through the beta process.)'. All match the article. (1) Apple Safari 27.3 beta release notes: the stored snapshot (source-1, sha256 7e30a204...) is a JavaScript shell containing only the page title, so it was read by another route: the publisher's JSON endpoint https://developer.apple.com/tutorials/data/documentation/safari-release-notes/safari-27_3-release-notes.json fetched live with a browser User-Agent on 2026-11-09 (HTTP 200, application/json, 144498 bytes, sha256 c2bf7ee918767c36cdd2c491b5a39e393a55c225d86e662f01749355507f13ce), the same endpoint the writer used (the rendered page is built from it). It says 'Released November 3, 2026 - 27.3 beta (20626.1.12)'; Overview: 'Safari 27.3 beta is available for iOS 27.3 beta, iPadOS 27.3 beta, visionOS 27.3 beta, macOS 27.3 beta, macOS 26, and macOS Sequoia.' (matches the article's list); under Rendering > New Features: 'Added Adaptive Dark Mode, which automatically adapts bright web pages that don't natively support dark mode...' (189226811). The notes do not mention color-scheme or the 'only' keyword. The headline does not rest on this page alone (webkit.org carries it). (2) MDN color-scheme (snapshot source-2, sha256 64bd6d5a...): 'only - Forbids the user agent from overriding the color scheme for the element. Can be used to turn off color overrides caused by Chrome's Auto Dark Theme, by applying color-scheme: only light; on a specific element or :root.' Examples write 'only light' / 'only dark'; formal syntax 'normal | [ light | dark | <custom-ident> ]+ && only?'. The article's description is accurate; the order difference (MDN 'only light', WebKit 'light only') is reported, and the either-order reading of '&&' is explicitly labelled as the Herald's reading (it is also correct CSS grammar: '&&' allows either order). MDN says nothing about Safari 27.3 or Adaptive Dark Mode. (3) 9to5Mac, 2026-11-03 (source-3, sha256 da6698cb...), Chance Miller, 'Nov 3 2026 - 10:51 am PT' (page metadata 2026-11-03T18:51:53Z): 'If Dark Mode is enabled ... and you visit a website that doesn't support it, you'll see a new "Try" button'; 'You can choose to have Adaptive Dark Mode automatically kick in or use it on a site-by-site basis.'; 'The update will likely be released to everyone sometime in December.'; names iOS, iPadOS and macOS 27.3 first betas. Matches. It says nothing about an opt-out tag. (4) 9to5Mac, 2026-11-05 (source-4, sha256 7c316839...), Zac Hall, 'Nov 5 2026 - 11:01 am PT' (2026-11-05T19:01:35Z): 'Apple has released the first iOS 27.3 public beta for iPhone'; 'Safari: Vertical Tabs and Adaptive Dark Mode let you move your tabs to the side and darken websites that don't have their own dark theme.' Matches the article's November 5 sentence; same outlet as source 3. (5) MacRumors, 2026-11-03 (source-5, sha256 dbadbd95...), Juli Clover, 'Tuesday November 3, 2026 10:51 am PST' (18:51:37Z): lists the background/text-contrast/logos/vector/raised-surfaces/shadows adjustments, 'not every tweak is applied to every site' (article paraphrases without quote marks: correct), the Try option, 'Developers can also add a tag that causes Safari to ignore Adaptive Dark Mode and show the website as the developer intended.' (verbatim) and 'iOS 27.3, iPadOS 27.3, and macOS Golden Gate 27.3 will launch this December after several rounds of beta testing to refine the new features.' (the article's truncated quote is verbatim and ends before 'to refine'). It does not name the tag. (6) TechRadar (source-6, sha256 7f2fa79f...), published '3 November 2026' (2026-11-03T21:00:43Z): 'For developers, there are new tags they can use that will prevent Adaptive Dark Mode from adjusting them.' (verbatim), 'all of these updates are found only in the Dev Betas' (article paraphrase correct), 'There's no word on when the full releases will arrive' (verbatim). It does not name the tag. Opt-out gap: WebKit's own note says the 'only' keyword is not supported in beta 1; MacRumors says 'a tag', TechRadar 'new tags' - the article reports all three with attribution and does not reconcile them, and says neither press outlet names the tag (true, checked in raw text). Title check: 'Skips Pages Declaring Dark-Mode Support' tracks WebKit's 'websites that signal support of dark mode with [meta color-scheme light dark | CSS color-scheme: light dark]' and the summary uses 'declare'; the body's second section names the exact declarations. WebKit lists prefers-color-scheme only among the technologies for building a dark mode, not as a skip trigger, and the article does not claim it is one. Dates: November 3 (WebKit, Apple, 9to5Mac, MacRumors, TechRadar) and November 5 (9to5Mac public beta) are correct on both the local-time and UTC bases, so no time-zone caveat is needed; the report order is chronological. Release timing differences (MacRumors 'will launch this December', 9to5Mac 'will likely be released ... sometime in December', TechRadar 'no word') are reported with attribution and not reconciled. Quantifier and absence sentences (grep run over the submission): 'Neither MacRumors nor TechRadar names the tag' - true; 'None of the cited sources says whether the only keyword will be supported in a later beta or the final release, or what a page that uses it sees in the meantime' - true across all 7 (WebKit says only 'check back'; Apple notes, MDN and the three outlets are silent); 'The three press articles name the iOS, iPadOS and macOS 27.3 betas' - true of the three outlets' launch-day articles (9to5Mac Nov 3, MacRumors, TechRadar; MacRumors calls macOS 27.3 'macOS Golden Gate 27.3'), while the Nov 5 9to5Mac piece names only iOS; the count of 'three' is by outlet and the surrounding paragraph makes that clear, so noted but not a correction; 'only 10%' and 'only' as a keyword are inside quotes or code. No relative dates; the beta status is labelled throughout. suspicious_patterns: none flagged for any of the 7 sources. Allowlist: webkit.org (line 875), developer.apple.com (869), developer.mozilla.org (2763), 9to5mac.com (126), macrumors.com (1135) and techradar.com (109) are all present in config/source_allowlist.txt; no additions needed. Independent-source recount: 3 outlets independent of Apple/WebKit (9to5Mac - its two articles are one source, MacRumors, TechRadar); first-party/reference: webkit.org, Apple release notes, MDN. Reporting versus restating: MacRumors and TechRadar describe the behaviours (shadows, raised surfaces, vector artwork) that the WebKit post does not, and 9to5Mac/MacRumors describe the Try flow and the site-by-site setting, but none of the three states it tested the feature or reproduced the behaviour, so the press confirms Apple's description rather than verifying it. Anthropic disclosure: not required - the article is about Safari/WebKit, and neither Anthropic nor Claude appears in the headline, summary, lead or body.