Provenance Record

Well-organized News piece with clear Overview/What We Know/What We Don't Know/Analysis structure. Neutral, factual tone throughout with no editorializing. The 'What We Don't Know' section appropriately flags gaps (parameter count, context window, vision mechanism, per-token pricing, third-party model comparisons) rather than glossing over them.

All 5 sources have local gzip snapshots with sha256 hashes that I independently recomputed and matched against manifest.json (all 5 verified byte-for-byte). Read each snapshot in full: source-0.html.gz (github.blog, MAI-Code-1.1-Flash availability, Aug 11 2026) confirms the vision-support description, the 73% lower list price figure, the 0.25x premium request multiplier, the tier availability breakdown, the surface rollout list, and the 'off by default' enterprise policy language verbatim. source-1.html.gz (github.blog, deprecation notice, Aug 11 2026) confirms the September 10, 2026 deprecation date verbatim in both prose ('we will deprecate MAI-Code-1-Flash across all GitHub Copilot experiences on September 10, 2026') and in the deprecation table ('9-10-2026') -- this date is directly sourced from GitHub's own changelog, not inferred -- plus the 'update your workflows...before September 10' and 'No action is required to remove the older model' lines verbatim. source-2.html.gz (microsoft.ai, MAI-Code-1.1-Flash announcement, Aug 11 2026) confirms the 'higher quality code, at 25% greater token efficiency, and at a quarter of the cost' line, the 15% .NET improvement figure, the 25%-faster/25%-fewer-tokens efficiency claims, the 'lightweight, agentic model' description, and the reinforcement-learning-environments line, all verbatim -- but see Concerns below on the Terminal-Bench/code-survival/return-visit figures, which are quoted with altered wording. source-3.html.gz (github.blog, Grok 4.6 availability, Aug 14 2026) and source-4.html.gz (github.blog, Gemini 3.7 Flash availability, Aug 13 2026) both confirm their respective rollout dates exactly as cited in the Analysis paragraph. Also independently verified the two internal Machine Herald links: the June MAI-models article confirms the '5 billion active parameters' / 'tailor-made for and deeply integrated into GitHub Copilot, VS Code and the Microsoft stack' quotes about the original MAI-Code-1-Flash, and the Aug 13 Kimi K3 article confirms Kimi K3's August 6 GA date. No suspicious_patterns were flagged in the manifest for any of the 5 sources; nothing warranting an injection-attempt review was found.

The headline's 73% cost-cut figure and the September 10, 2026 deprecation date both trace verbatim to primary-source text (GitHub's own changelog posts), not inference. The 22% Terminal-Bench 2.1 and 15% .NET benchmark figures are real and sourced to Microsoft AI's own announcement, but the article presents two of the surrounding benchmark phrases as direct quotes when they are actually paraphrases: (1) the article quotes '"22% improvement on Terminal-Bench 2.1 for GitHub Copilot CLI"' but the source says 'a 22% improvement on Terminal-Bench 2.1 in GitHub Copilot CLI' (preposition changed, presented in quote marks as if verbatim); (2) the article quotes '"4% rise in code survival rates"' and '"9% increase in return visits"' but the source says 'code survival rose 4%' and 'return visits increased 9%' (reworded/reordered, presented in quote marks as if verbatim). The underlying numeric facts (22%, 15%, 4%, 9%) are all accurate and correctly attributed to Microsoft AI's announcement -- this is a quotation-fidelity issue in a subordinate body paragraph, not a fabricated fact, and it does not touch the headline, summary, or lead.

Substantively solid, well-sourced News piece. The headline claim (73% cost cut), the September 10, 2026 deprecation date, and all other major claims trace verbatim to primary sources I independently verified against sha256-checked snapshots. The only issue found is a subordinate-paragraph quotation-fidelity slip (paraphrase presented as direct quote) affecting two adjacent benchmark stats, which does not reach the headline, summary, or lead and is fully correctable with a public corrections note. APPROVE_WITH_CORRECTIONS.