docs(adr): scope verification-guard corpus to deployment surface

[kyle-sexton]

Fixes #1438

ADR 0003 decision rule 1 now distinguishes the authoring repository's corpus from a guard's deployment surface, names the two readings (marketplace-confined vs consumer-shipped), and cites the asserted-path-verify consumer-config finding as the motivating case.

Related

• ADR 0003 verification-guard corpus scope (docs(adr): ADR 0003's corpus rule does not scope the sweep to a hook's actual deployment surface #1438)

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[claude]

Claude finished @kyle-sexton's task in 2m 7s —— View job

---

Review complete

• Gather context (diff, changed file, CLAUDE.md/REVIEW.md guidelines)
• Understand the request (code review of ADR corpus-scope clarification)
• Review changed file for correctness and alignment with guidelines
• Post findings

Summary

This is a docs-only change to docs/adr/0003-verification-guards-earn-default-on-by-measured-precision.md (+14/-1), adding a clarification to decision rule 1. I read the full ADR (Context, Decision, Consequences, Sources) to check the addition against the rest of the document, not just the diff hunk.

Findings: none blocking. The addition is accurate and internally consistent:

• The new "deployment surface, not necessarily the authoring repository" framing directly resolves the ambiguity rule 1 previously had — the original wording ("the real corpus") didn't say whose corpus, and the two named readings (marketplace-confined vs. consumer-shipped) map cleanly onto the guard categories the ADR already discusses.
• The motivating-case paragraph (lines 71–74) restates the path guard's dominant failure class from the Context section (lines 40–42) almost verbatim. That's consistent with how the rest of rule 1 already works — it also restates the "61 green cases / 23.7%" stat from Context — so the Decision section stays self-contained without a reader needing to cross-reference. Not a defect, just noting the pattern is intentional rather than accidental duplication.
• Markdown structure is valid: the new bullets use - (matching this repo's MD004: dash style) and the continuation/bullet indentation is 3 spaces, correctly aligned under the 1. marker, so it nests inside list item 1 rather than breaking out as a sibling.
• No code, tests, or security-relevant surface is touched, so the REVIEW.md code-review-lane checks (audit logging, multi-file atomicity) don't apply here.

One thing I could not verify: I attempted to run markdownlint-cli2 (this repo's configured linter, per .markdownlint-cli2.jsonc) against the file via npx, but the sandbox blocked the network-fetching command before it could run. I did not confirm lint-clean status with a tool — my markdown-structure read above is manual, not tool-verified. If you want that confirmed, CI's own markdownlint job will catch it, or grant this lane network/bash access for npx.

No inline comments filed — I didn't find a finding tied to a specific line worth anchoring.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f5d572f30f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[github-actions]

Claude has reviewed this PR 1 time. The lane skips further automatic reviews after 5; deleting this comment resets the count.

[kyle-sexton]

Addressed P2 path-guard verdict in rebase onto main; ready for merge when review completes.

[github-actions]

Warning

Automated review did not complete — this is an infrastructure failure, not a review verdict.

Treat any Claude comment on this PR (including a placeholder like "I'll analyze this and get back to you") as incomplete, not "no findings."

• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/31559976081
• Failure class: other
• Last SDK result: (no execution file was produced)

Re-run the job to retry the review. A new push re-triggers this lane only if the caller's pull_request triggers include synchronize (the canonical caller omits it).
An automatic retry may already have run — it is skipped when a partial review could duplicate comments, or when the failure class needs an operator (auth).