Provenance Record

Clean, well-structured News piece in the standard Overview / What We Know / What We Don't Know format. Appropriate technical depth on worktrees and the Copilot SDK integration; 498 words, well within the News 400-1200 range. Every substantive claim is attributed inline to the cited source.

Both automated snapshots FAILED at chief:review time (manifest.json records file: null for both sources — github.blog returned 'fetch failed' and windowsforum.com returned HTTP 403 bot-block). No .html.gz snapshots exist to read. As permitted for failed snapshots, I fell back to live WebFetch to verify. source-0 (github.blog changelog, the primary source): live WebFetch CONFIRMED all nine attributed claims/quotes verbatim — the summary line, macOS/Windows availability, the worktree description, the 'coding agents ... isolated, parallel sessions' line, the copilot-instructions.md/AGENTS.md custom-instructions claim, the merge-conflict 'review, accept, or edit' language, the Copilot SDK 'shared foundation' claim, the model-picker + BYOK sentence, and the 'free to download and use' pricing line. source-1 (windowsforum.com, secondary corroboration): live WebFetch CONFIRMED the June 26, 2026 release date, the BYOK statement, and the caution that a clean AI-suggested merge 'can still be wrong.' Both sources support what the article attributes to them.

No hallucinations detected. Every direct quote appears verbatim in the primary changelog. The version number (3.6.0), platforms (macOS/Windows), date (June 26, 2026), and all feature specifics trace to the cited sources. The 'What We Don't Know' section correctly frames the limits of GitHub's disclosure without inventing facts.

Content is factually clean and every claim is verbatim-verified against its cited source; the only issues are provenance-mechanical (failed auto-snapshots, one off-allowlist secondary source), not editorial. APPROVE_WITH_CORRECTIONS with a clarification note documenting the source-verification circumstances.