Provenance Record

Well-structured News piece (Overview / What We Know / What We Don't Know / Analysis). Clear, neutral prose at 710 words, within the News 400-1200 word range. The topic is well-chosen and timely — it covers a concrete governance change (a real commit) rather than speculation, and the 'What We Don't Know' section appropriately flags open questions instead of papering over them.

Read and gunzip-verified all 6 snapshots in sources/2026-09/systemds-new-agentsmd-rule-adds-an-ai-canary-to-force-human-review-of-ai-written-code/ (all sha256 hashes in manifest.json matched the decompressed content): (1) source-0.html.gz — GitHub commit 73aa952, confirms commit title 'Ensure contributors engage with their AI written code', the exact commit-message quote 'Let's add a canary so that we can distinguish AI written code that was not looked at in any way or form by its author easily in pull requests,' the diff adding the HARD RULE text to AGENTS.md, and authoredDate 2026-09-02T16:35:03+02:00 (confirms 'September 2, 2026' and author Daan De Meyer / daandemeyer). (2) source-1.html.gz — AGENTS.md file view, confirms the HARD RULE paragraph verbatim (README.md marker instructions, 'before any other work', 'No exceptions', prohibition on the agent removing it) and the Legal-section quote 'Only human beings can ever be credited within commit messages... no Co-Developed-By or Co-Authored-By or anything similar that lists an AI model instead of a human being' verbatim. (3) source-2.html.gz — GitHub commit 27c5a2f2 ('docs: Update AI usage policy'), confirms authoredDate 2026-06-17, and all three direct quotes used in the article ('written from a standpoint that AI models are not very good...', 'now that AI models have become actually good...', 'It's up to the author to make sure they're not wasting our time by submitting unreviewed, untested code upstream, regardless of whether that code is written by an AI or not') appear verbatim (modulo line-wrap whitespace), plus the inspired-by link to lxc/incus/pull/3506. (4) source-3.html.gz — Phoronix 'systemd 262-rc2 Adds An AI Canary...', published 8 September 2026, confirms the HARD RULE canary text is quoted correctly by Phoronix and confirms the rc1-then-rc2 release cadence, BUT does NOT contain the phrase attributed to it in the submission ('a detection mechanism modeled after NetworkManager's approach') — see finding above; this is the one issue driving APPROVE_WITH_CORRECTIONS. (5) source-4.html.gz — GitHub release tag v262-rc2, confirms the release exists as cited (used only to support the 'second release candidate' framing, not linked in body). (6) source-5.html.gz — lxc/incus PR #3506 ('Update AI/LLM policy' by stgraber), confirms the PR exists as the cited inspiration link. No snapshot failures; no WebFetch fallback was needed.

All specifics (commit hashes, commit titles, authorship, dates, quoted commit messages, quoted AGENTS.md rule text, the Incus inspiration link) trace correctly to their cited sources and were verified verbatim against the raw snapshots. The one exception is the Phoronix 'modeled after NetworkManager's approach' quote, which is fabricated/misattributed — see finding. Everything else in the article, including the nuanced 'What We Don't Know' hedges, held up under verification.

Substantively strong, well-sourced News piece with one recoverable issue: a fabricated direct quote attributed to Phoronix describing the NetworkManager comparison. The issue sits in the body (not the headline, summary, or Overview/lead), the underlying fact is true and independently sourced, and a single corrections note can honestly inform readers — so APPROVE_WITH_CORRECTIONS rather than REJECT.