Source Verification: ALL 11 sources read from snapshots in sources/2026-11/four-developer-tools-host-other-vendors-coding-agents-.../ (gunzip, HTML to text) and every one of the 34 quoted strings was checked programmatically against the normalised raw text of the snapshots: 34/34 found verbatim in the source the article attaches them to (the bare word 'Agents' is a quoted UI label found in Apple's page). Suspicious-pattern scan: suspicious_patterns null for all 11; no injection text found. ROUTES: source-1 (developer.apple.com, DocC page) is a script-rendered shell (64 characters of text), so it was read by another route: live fetch with a browser User-Agent of the underlying DocC JSON https://developer.apple.com/tutorials/data/documentation/xcode/setting-up-coding-intelligence.json (HTTP 200, 16,629 bytes, sha256 ea2d2e797ea7467759223abcfa7d5727062e63f3223f2f99e5256919d814f638, fetched 2026-11-10); all Apple quotes and the Add an Agent (ACP), Account sign-in, ChatGPT and MDM CodingAssistantAllowExternalIntegrations=false statements confirmed in its raw text, and the page contains no usage, billing, spending or audit text. Every other snapshot was HTTP 200, no archive fallback (sha256 per manifest). Supplementary live fetch (browser User-Agent, HTTP 200, sha256 1614ecc1df14104e7f929564770e078ede8ea62f5d2da22d553376b45dcb7471, 2026-11-10) of https://docs.github.com/en/copilot/reference/copilot-billing/request-based-billing-legacy/what-changed-with-billing, a page linked from the sidebar of the cited GitHub docs page (see claim 6 below). PER SOURCE: [0] 9to5Mac, Marcus Mendes, 'Nov 3 2026 - 1:04 pm PT': Xcode 27.3 beta 1 adds GitHub Copilot; release-note line 'GitHub Copilot is now available in the coding assistant.' CONFIRMED; the article's one-line-release-note description is right. [1] Apple: CONFIRMED (see route). [2] GitHub changelog (Xcode): public preview, Xcode 27.3 beta, 'All Copilot plans are supported', separate Copilot for Xcode extension stays; published 2026-11-04T19:10Z. CONFIRMED. [3] Android Developers Blog, 28 Oct 2026, Matthew Warner: Rabbit 2 'now stable and ready for you to use in production', BYOA, 'over 40 coding agents' including Claude Agent, Codex, Junie, Google Antigravity, steps 'authenticate, and submit your first prompt', 'cost efficiency' and 'enterprise agreement' sentences, Gemini Enterprise section with 'Compliance & Audit Logging: Full visibility into agent activities, generated patches, and prompt history' and 'Admin Policies: ... define approved skills, agents, and model endpoints'. CONFIRMED; no usage/quota/limit/billing text on the page. [4] The Decoder, 28 Oct 2026: built-in marketplace, Claude Agent, Codex, Antigravity, ACP. CONFIRMED; it is a restatement of the Google announcements (source line 'Android Dev Summit | Android Studio Rabbit 2'). [5] The Register, Anderson, published tue 10 Mar 2026 // 16:38 UTC: Air in public preview supports OpenAI Codex, Anthropic Claude Agent, Google Gemini CLI, JetBrains Junie; ACP sentence verbatim. CONFIRMED; date March 10 correct. [6] GitHub changelog (Air), 'November 3, 2026', published 2026-11-03T07:11-08:00 (15:11Z), modified 2026-11-06T20:54Z: 'use your existing Copilot subscription', 'Tasks use your Copilot plan rather than JetBrains AI credits', 'Air's governance and analytics features do not include Copilot usage', 'Air administrators can disable GitHub Copilot through a governance policy, but cannot set Copilot spending limits within Air'. CONFIRMED incl. the 'modified on November 6' claim and the Pacific-time basis. These usage/governance sentences sit under 'Cloud task considerations' in GitHub's entry, but JetBrains' FAQ [7] states 'Air doesn't track Copilot credits for local or cloud tasks', so the article's Air-wide statement is supported. [7] JetBrains blog (Vladimir Gromozdin): 'the official GitHub Copilot CLI over the Agent Client Protocol (ACP)', 'the same task list as Claude, Codex, and Junie', 'GitHub, not Air, handles your Copilot account', 'you have to keep track of it on GitHub'. CONFIRMED. [8] GitHub Docs, Third-party coding agents: public preview; two supported agents (Anthropic Claude, OpenAI Codex); enablement in account settings, organization policies or the 'Agents' section of an enterprise's AI controls; 'can be powered by your existing Copilot subscription'; 'Coding agents consume GitHub Actions minutes and AI credits. Each agent session consumes AI credits based on the model used and the number of tokens processed.'; audit-log sentence; VS Code local-agents sentence. CONFIRMED. [9] GitHub blog, 4 Feb 2026 (published 2026-02-04T09:00-08:00): both quotations verbatim ('Each coding agent session consumes one premium request.'). CONFIRMED. [10] InfoWorld, Oct 30 2026, on GitHub Universe: Amit Kumar Jena is 'AI development head at IT consulting firm Kanerika'; 'set credit limits for individual projects' is Jena's direct quote about Copilot Projects; 'monitor agent activity through audit logs' is InfoWorld's indirect speech for Jena about the revamped app ('They should also monitor agent activity through audit logs ... he added'); 'shifting to an AI credit-based consumption model' is InfoWorld's own sentence (Satapathy cited for the effect). The piece covers Copilot's own app and Projects and never mentions third-party/bring-your-own agents (the only Claude/Codex mention is a related-article link), as the article says. CONFIRMED. CLAIM-ACCURACY TESTS: (1) Absence/quantifier sentences, all run against all 11 snapshots: 'None of the cited independent outlets ... reports a test of how any of the four hosts bills, limits or logs an agent the developer adds' HOLDS (the Decoder's token-usage figures are Google's tests of the Android CLI, not hosted-agent billing); 'Apple's page ... does not say how usage or spending for a third-party agent is shown' HOLDS; 'Google's post does not say whether Gemini Enterprise audit logging and admin policies reach BYOA agents' HOLDS; 'Each of Apple, Google, JetBrains, GitHub documents a way to run another vendor's agent' HOLDS per host (Apple: Intelligence settings + ACP Add an Agent; Google: BYOA; JetBrains: Air agent picker; GitHub: Claude and Codex); 'The vendors' pages agree that the developer may choose' HOLDS for all four; the article's 'both/every/each' statements about the four hosts each hold per host. The earlier Herald Air error ('both vendors direct users to GitHub for usage', corrections record 2026-11/06-github-copilot-is-now-selectable-in-jetbrains-air...json) is NOT repeated: the article attributes the changelog statement to GitHub and the 'keep track of it on GitHub' line to JetBrains separately, and does not link that article. (2) Headline/summary/lead: qualifiers kept (preview/beta noted in the lead and in 'Two cautions'); the title's three questions are each answered per host from that host's own page, though for Xcode and Android Studio the answer to 'where usage shows' is a stated absence. (3) Dates: Feb 4 (GitHub post), Mar 10 (Register, 16:38 UTC), Oct 28 (Android blog and Decoder), Oct 30 (InfoWorld), Nov 3 (9to5Mac 1:04 pm PT; GitHub Air changelog, Pacific basis stated) all correct and chronological; no relative dates besides 'now' in present-tense statements anchored to 'as read on November 10, 2026'. (4) No Herald-derived figures. (5) Disagreement between two GitHub pages (one premium request per session in Feb vs AI credits by model and tokens now) is reported with both attributed and dates; see correction 1 for an explanation the Herald did not cite. (6) Independence recounted below. INTERNAL LINKS (opened): /article/2026-03/25-jetbrains-unveils-central... exists, no corrections record, the article does say IntelliJ IDEA 2026.1 added external-agent support through ACP, so the description is exact; /article/2026-11/05-github-copilot-joins-xcode-273... (Briefing) and /article/2026-11/04-android-studio-rabbit-2... (News) exist, no corrections records, and are described as separate reports of the Xcode and Android Studio steps. The 2026-11/06 Air article is not linked. INDEPENDENT-SOURCE RECOUNT (by outlet and by case): 4 outlets independent of the four vendors: 9to5Mac (Xcode; restates Apple's release note and adds the history of earlier integrations), The Decoder (Android Studio; restates the Google announcement), The Register (Air and ACP, 10 Mar 2026; original reporting, but predates Copilot-in-Air and says nothing on accounts or usage), InfoWorld (original reporting on GitHub's Copilot app/Projects and an analyst's advice; says nothing on third-party agents). No syndication or duplicate outlets. The 7 other sources are first-party (Apple docs, Android Developers Blog, GitHub docs, GitHub blog, two GitHub changelogs, JetBrains blog). By case: Xcode 1 independent (9to5Mac), Android Studio 1 (Decoder), Air 1 (Register, March only), GitHub 1 (InfoWorld, adjacent product). Strictly, no case has independent support for the three compared dimensions (who lists, which account, where usage shows); that comparison rests on first-party pages, which the article states plainly in What We Don't Know. The Analysis requirement (4+ sources, 2+ independent by outlet, restating outlets described as such) is met, with that limitation. Allowlist: 9to5mac.com, the-decoder.com, theregister.com, infoworld.com, developer.apple.com, developer.android.com, docs.github.com, github.blog, blog.jetbrains.com are all in config/source_allowlist.txt (the article cites developer.android.com, not android-developers.googleblog.com; both listed); no additions needed. Anthropic disclosure: not required - Claude and Anthropic appear only as named agents in lists of hosted agents (Xcode, Android Studio, Air, GitHub docs, the February 4 quotation) and not in the headline, summary or lead; the article is not about an Anthropic product and has no Anthropic-credit passage of its own.
Factual Accuracy: All 34 quotations verbatim; all dates, versions and attributions verified. Two subordinate imprecisions and one uncited explanation, recorded as clarifications: (1) The 'Two cautions' paragraph and the last What We Don't Know bullet leave the February 4 'one premium request' versus docs 'AI credits' difference unreconciled, tying it only to InfoWorld's October 30 'AI credit-based consumption model'. GitHub's docs (page linked from the cited docs page's sidebar: 'What changed with Copilot billing (legacy)', read November 10) state that before June 1, 2026 billing was premium-request-based and that 'As of June 1, 2026, GitHub replaced request-based billing with usage-based billing'; the February 4 figure therefore describes the earlier model. The article's wording ('none of the cited sources says') is literally true of the cited pages, but a GitHub page explains the difference, which the Claim accuracy rule 6 asks reviewers to check first. (2) The Background says Apple added 'Claude, Codex and Claude Agent in Xcode 26.3'; 9to5Mac says ChatGPT came in Xcode 26, 'followed by Anthropic's Claude', and that Xcode 26.3 'expanded both integrations ... through OpenAI's Codex and Anthropic's Claude Agent'; Claude (chat) is not attributed to 26.3. (3) The quotation 'monitor agent activity through audit logs' is InfoWorld's indirect speech about Jena's advice, not his own words (his direct quote is only 'set credit limits for individual projects'). Not corrected (noted only): Google's setup steps are 'authenticate, and submit your first prompt', so 'end with authenticating' compresses them; Google's Admin Policies line says administrators can 'define approved skills, agents, and model endpoints', which bears on the 'who lists' question, though the article correctly says the post does not tie it to BYOA agents.