Content Quality: Well-structured Analysis with clearly separated Overview/Background/Current situation (facts) and Implications (analysis) sections. 1536 words, within the 800-2000 Analysis range. Neutral tone throughout; no sensationalism or AI self-reference.
Source Verification: All 4 source snapshots were fetched successfully (HTTP 200, no archive fallback) and their sha256 hashes verified against manifest.json before reading. Read in full via gunzip + text extraction: source-0.html.gz (Fortune, 'Exclusive: Cognition bets on Devin Bots...', Nov 17 2026) confirms the Devin Bots description, Scott Wu quotes ('Self-driving software is...', the security-model quote, 'rather than relying on models themselves...'), the Interaction Company/Poke acquisition, the $1B ARR and $48B valuation figures, the tenfold Bots growth claim, and the Citigroup/Jane Fraser 14,000-engineer quote verbatim. source-1.html.gz (The Next Web, 'Cognition and McKinsey team up...', Nov 16 2026) confirms the McKinsey alliance terms, the Kaplan quote, the 334-leader/top-quarter-doubled/30%-fell survey figures, the Hämäläinen quote, the Levy/Ferrari quote (verbatim, confirmed via raw-text search), the six-of-eight-banks and three-of-six-insurers client figures, the Comarch ERP XL detail, and the Stamos/$48bn/$80bn financing timeline. It does NOT support the 'frontier-level performance for 60% less' phrase as a direct Cognition quote -- see findings. source-2.html.gz (SiliconANGLE, Claude Duet launch, Oct 22 2026) confirms Duet's description, availability (web/iOS/Android, Pro/Max), its basis in Anthropic's agent framework (Claude Code, Claude in Chrome), the automatic-classifier detail, and the Karan Ahuja usage detail -- but does NOT quote 'even ship code' verbatim; that phrase is SiliconANGLE's own narration. See findings. source-3.html.gz (TechCrunch, OpenAI Dots launch, Sept 29 2026) confirms the DevDay date, the 'remarkably capable, always-on agents built to handle everything' quote verbatim, the GPT-6 Astra basis, the Pro/Business Premium availability, deployment from Codex, and the customer-feedback-monitoring use case. No suspicious_patterns were flagged in manifest.json for any of the 4 sources; none required review as potential prompt injection. Independent-source count for the Analysis requirement: 4 of 4 sources (Fortune, The Next Web, SiliconANGLE, TechCrunch) are independent news outlets with no commercial or ownership tie to Cognition, McKinsey, Anthropic, or OpenAI -- well above the 2-of-4 minimum. Anthropic disclosure: required - missing (see disclosure finding above; filed as a clarification correction rather than blocking publication, since the lead's single comparative mention of Duet could plausibly be read as the policy's 'incidental... competing models' exemption, but the dedicated background paragraph and implications-section comparison tip the article into substantive treatment that the policy's disclosure rule is meant to cover).
Factual Accuracy: Central-tension claim verified: The Next Web's survey figures (334 leaders, top quarter more-than-doubled productivity, 30% saw productivity fall) are reported by The Next Web without any explanation of what distinguishes the two groups, and Fortune does not address the survey's internal split at all. The article's Implications section states this accurately ('Neither The Next Web nor Fortune reports what distinguishes the organizations in the top quarter from the 30% that saw declines') rather than inventing a reconciliation -- this is the correct, honest framing of an unresolved tension in the cited sources. All dates (Nov 16/17 2026 for Cognition's announcements, Sept 29 2026 for Dots, Oct 22 2026 for Duet, Oct 13 2026 for Citigroup's earnings call) match the source bylines/datelines. All specific figures (revenue run rate, valuation, financing talk, client counts, survey sample size) trace to the sources cited beside them. No orphan URLs: every URL in body_markdown appears in article.sources and vice versa. The cross-reference to the Herald's November 6 orchestration-layer Analysis was checked against the live article file (src/content/articles/2026-11/06-as-coding-agents-multiply-...md): the November 6 piece does carry a passing, unexamined Register reference to 'Claude Duet and Meta Muse' alongside Dots/Grok Bot and states 'The Herald has not independently examined those other products' -- the new submission accurately summarizes that passage and extends it with the Herald's own independent reporting on Duet and Dots, rather than re-narrating the November 6 article's content. Two quote-attribution errors found (see findings): 'frontier-level performance for 60% less' and 'even ship code' are each presented in quotation marks as if verbatim from the cited outlet, when in each case the outlet's own text is unquoted narration/paraphrase. The underlying facts both phrases convey are accurately sourced; only the quotation-mark treatment is wrong.
Overall Assessment: This is a well-sourced, well-organized Analysis that meets the category's requirements (1536 words, 4 independent sources, facts separated from analysis, an examination of the McKinsey-survey tension rather than a release recap) and correctly builds on the Herald's prior November 6 reporting without re-narrating it. Two subordinate quote-attribution errors and a missing Anthropic disclosure sentence are the only issues found; none touch the headline, summary, or lead, and all three are honestly coverable in corrections records. Featured candidacy: 1 - solid analytical treatment of a notable but not industry-defining enterprise AI-coding development, useful to a developer/enterprise-tooling audience but not broadly transformative.