Skip to content

docs(agents): record what let an unverified conjunction reach users - #432

Merged
doublegate merged 1 commit into
mainfrom
docs/two-true-facts-lesson
Aug 20, 2026
Merged

docs(agents): record what let an unverified conjunction reach users#432
doublegate merged 1 commit into
mainfrom
docs/two-true-facts-lesson

Conversation

@doublegate

Copy link
Copy Markdown
Owner

Follow-up to #431, which retracted the claim. This records what let it happen, because the instance is the least interesting part.

The mechanism

Both inputs were real and each had been checked independently:

  • release_anchor_audit.rs genuinely is absent from the v2.3.9 tag.
  • v2.3.9 genuinely is where the anchor drift was found.

What was never checked was the conjunction — that v2.3.9's body therefore described the audit. Checking it was one command:

gh release view v2.3.9 --json body

That body says the eight drifted documents "All now read v2.3.9" — a manual re-synchronisation, which is in its tag — and contains no occurrence of release_anchor_audit, "standing", "gate", or "15 anchors". Verifying every premise is not verifying the conclusion.

The two propagation mechanisms, which matter more

AGENTS.md is a notes file, not an oracle. The sentence was written here during the v2.4.0 work, and the v2.4.1 release summary was then drafted from here. That is the laundering step: an unverified claim entering this file is indistinguishable from a verified one by the time the next release quotes it — and it then reaches a published release body, a commit body, a PR body, and a report to the maintainer in one pass. This is the first entry that says so about the file itself.

The claim had the shape this project rewards — a recursive irony, "the same failure one level up". That shape suppresses the check, because it reads as an insight rather than an assertion needing evidence. The house style's appetite for findings of that form is precisely why it needs a verification bar, not a looser one.

The bar

Before writing that a document, release, commit or person says something, open it and grep for the words.

One bullet, inserted beside the sibling traps it belongs with — the reviewer-claim bullet it now precedes is the same failure with a different source. Docs only.

#431 retracted the claim. This records the cause, because the instance is the
least interesting part of it.

Both inputs to the false statement were real and each had been checked
independently: release_anchor_audit.rs genuinely is absent from the v2.3.9 tag,
and v2.3.9 genuinely is where the anchor drift was found. What was never
checked was the conjunction -- that v2.3.9's release body therefore DESCRIBED
the audit -- and checking it was one command:

  gh release view v2.3.9 --json body

That body says the eight drifted documents "All now read v2.3.9", a manual
re-synchronisation which is in its tag, and contains no occurrence of
"release_anchor_audit", "standing", "gate", or "15 anchors". Verifying every
premise is not verifying the conclusion drawn from them.

Two propagation mechanisms matter more than the instance, and the bullet names
both.

The sentence was written into AGENTS.md during the v2.4.0 work, and the v2.4.1
release summary was then drafted FROM AGENTS.md. That is the laundering step:
an unverified claim entering this file is indistinguishable from a verified one
by the time the next release quotes it, and it then reaches a published release
body, a commit body, a PR body and a report to the maintainer in a single pass.
AGENTS.md is a notes file, not an oracle, and this is the first entry that says
so about itself.

The claim also had the shape this project rewards -- a recursive irony, "the
same failure one level up". That shape actively suppresses the check, because
it reads as an insight rather than as an assertion needing evidence. The
house style's appetite for findings of that form is exactly why it needs a
verification bar rather than less of one.

The bar stated: before writing that a document, release, commit or person SAYS
something, open it and grep for the words.

Nothing else changes -- one bullet, inserted beside the sibling traps it
belongs with (the reviewer-claim bullet it now precedes is the same failure
with a different source).
Copilot AI lite review requested due to automatic review settings August 20, 2026 23:16
@coderabbitai

coderabbitai Bot commented Aug 20, 2026

Copy link
Copy Markdown

Important

  • 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: d3816f5a-942b-445c-ba25-4419f6284e79


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

@github-actions

Copy link
Copy Markdown

Antigravity review (Gemini via Ultra)

This PR adds a post-mortem entry to the agents documentation detailing how an unverified claim bypassed checks to reach release notes, establishing a new requirement to grep source material before attributing statements.

Blocking issues

None found. This is a trivial documentation change.

Suggestions

  • AGENTS.md, line 266: This addition is a single, dense ~330-word paragraph. Even if this matches the verbose style of surrounding entries, it dilutes the actual rule being established. Consider breaking the "Two propagation mechanisms matter more..." section into sub-bullets so the actual takeaways are readable.

Nitpicks

  • AGENTS.md, line 266: "re-synchronisation" uses British spelling ("-isation"). Verify if this matches the project's accepted locale, as US English is typical.

Automated first-pass review by agy on a self-hosted runner -- not a human review.

@doublegate
doublegate merged commit c565aee into main Aug 20, 2026
18 of 19 checks passed
@doublegate
doublegate deleted the docs/two-true-facts-lesson branch August 20, 2026 23:19
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants