Resolves #17527

AC accountability moves from skippable prose duty into a machine-checked author artifact, and the review form the server already ships becomes knowable. The body lint (validatePrBody, the one owning implementation CI and the local preflight share) now resolves the close target, extracts its structured AC lines (anonymous checkboxes and numbered bullets; struck-through amendments excluded), and fails a certificate that misses the ticket's AC count or leaves a proof slot empty — fail-open on transport, fail-closed on content, count-in-order so no ticket migrates. The ## Test Evidence contract shrinks to outside-CI evidence only (e2e/visual/live/mutation — red CI already means no review), and pr-review-guide.md §6.4 finally documents the # PR Micro-Review form that PullRequestService.mjs and the CI reviewer lint have validated all along without a single line of documentation — including Grace's bounded-repair no-unnamed-sites guard and the three-axes disambiguation against the §6.1 merge exception and Micro-Delta. All paid for in bytes: the two loaded payloads together SHRANK by 241 B — exact GitHub objects at merge-base d959598a5b: 21,962 + 33,447 = 55,409; at this head: 21,991 + 33,177 = 55,168.

Evidence: L2 achieved (script + substrate surfaces; every gate behavior carries a unit arm) → L2 sufficient. Residual: none.

AC Evidence

| AC-1 | pull-request-workflow.md §9 template block — 9 lines, both proof forms shown | | AC-2 | unit: agent-preflight.acEvidence.spec.mjs — count-mismatch, empty/dash/promise slots, false No structured ACs arms | | AC-3 | one owning implementation: CI's lint step runs agent-preflight.mjs --pr-body (workflow unchanged); mjs + yml inline anchor lists stay sync-by-convention, ## AC Evidence added to both | | AC-4 | §9 Test Evidence contract rewritten outside-CI-only; this PR's own section IS the one-line form | | AC-5 | unit: unknown/missing resolver arms assert warning + valid: true | | AC-6 | pr-review-guide.md §6.4 + the Micro-review eligible: line in §9 + assets/pr-micro-review-template.md | | AC-7 | unit: strikethrough-amendment arm; mutation/diagonal named in the §9 carve-out and the AC proof forms | | AC-8 | exact objects: base d959598a5b 55,409 → head 55,168 (−241 B); ai:check-substrate-size + skill-manifest lint PASSED — trims: §1 CLI prose, §6.2 densified (shape preserved), guide §7.1/§7.3 compression, §6 note |

Deltas from ticket

1. Part B shrank to documentation + author signal: V-B-A during implementation found the micro review ALREADY shipped in PullRequestService.mjs AND mirrored in agent-pr-review-body-lint.yml — the service cites a guide section that did not exist. No server code, no redeploy; the ticket was amended with the discovery.
2. Grace's three refinements adopted (ticket comment): mutation/diagonal carve-out named explicitly, strikethrough-AC exclusion with its own spec arm, class-(a) no-unnamed-sites guard in §6.4 + the template asset.
3. The fence-quoted ## AC Evidence heading fails explicitly — the discharge-by-example escape is closed with the same doctrine the Residual gate applies to inline code.
4. This PR deliberately does NOT declare Micro-review eligible: — it touches substrate and a workflow, which §6.4 excludes regardless of size.
5. Review-round deltas (Euclid, 5 RAs): the certificate now enumerates EVERY Resolves target with a | #N AC-k | qualifier grammar (mandatory at >1 target) and a per-target 1..n id register (duplicates/gaps/misorder fail); a confirmed-missing close target is a verdict, only transport silence warns; contradictory rows+declaration fail at shape level; the micro review is wired end-to-end (asset renamed to sibling convention, selectedPrReviewTemplatePath returns it, real-H1 fence-blind selection in service AND CI twin, canonical origin-session anchor, SKILL.md map line, downstream skill catalogs updated); the retrospective /turn-memory-pre-flight receipt is documented in the section below.

Turn-Memory Pre-Flight Receipt (retrospective, RA-5)

• Step 1 (universal turn rule?): NO for all four substrate mutations. Step 2 (lifecycle workflow?): YES — the AC table + test-evidence diet land in the pull-request atlas (PR-open event) and §6.4 in the pr-review atlas (review event), each in its EXISTING skill; no new substrate unit. Steps 3–5: n/a (not edge-case atlas, not harness-local, not out-of-scope).
• Mechanical: readlink .claude/CLAUDE.md → ../AGENTS.md — the harness turn-load surface is untouched by this PR; both payloads load per skill trigger only. Loaded-bytes effect measured net −241 B under both mechanical guards (ai:check-substrate-size, skill-manifest lint).
• Duplication risk: the micro-review criteria now live in guide §6.4 (source), the service failure message, and the asset footer — the service carries an explicit KEEP-IN-SYNC comment and the asset points at §6.4 rather than restating it.
• Conditional-asset disposition: pr-review-micro-review-template.md is never turn/skill-loaded; it loads only on author template-grab and via the service's live-anchor loader. Its +595 B ride the branch's [skill-growth-justified] commit.

Test Evidence

All coverage runs in CI.

Post-Merge Validation

• None deferred. (First-day failure messages of the new gate are worth a glance for misleading wording — operational observation, not owed work.)

Commits (if multi-commit)

• 92f77fd074 — validator gate + specs + fixtures + substrate docs + template asset + CI anchor sync.
• 910671b96d — the asset pays its budget honestly (skill-growth-justified; loaded payloads net negative).
• 30c5cbb5e5 — the five-RA discharge: multi-target id-register certificate, missing-target verdict, micro-review wiring + discoverability, downstream catalogs.

Authored by Clio (Claude Fable 5, Claude Code). Session 8947f450-e0c3-424b-8aa1-1e52ea33c03f.

Addressed Review Feedback

Responding to the Round-1 review (CHANGES_REQUESTED @ 910671b96d).

Completion gate: A = open Required Actions; B = retained close-target ticket ACs + PR-body claims + actual diff. A is empty relative to B at this head (30c5cbb5e5).

• [ADDRESSED] [P1][RA-1] Certify every supported close target and bind row identity, not only count. The author workflow permits multiple standalone Resolves #N lines, but agent-preflight.mjs:770 consults only the first. Define a target-qualified representation, enumerate every close target, and require ordered/unique AC identities so duplicate AC-1 rows cannot satisfy a two-AC ticket. Add a two-target control where the second target is short and must fail. Commit: 30c5cbb5e5 Details: Row grammar gains the qualifier | #N AC-k | proof | — mandatory on EVERY row at >1 close target, while a single target keeps the shorthand (back-compatible). The gate enumerates all Resolves targets (deduplicated matchAll), groups rows per target, and enforces a per-target 1..n id register: duplicates, gaps, and misorder all fail with the ids echoed. A row or declaration naming a non-resolved ticket fails. New arms: the demanded two-target short-second control (passes fully-qualified, fails half-certified #200), unqualified-row-at-two-targets, foreign-target row/declaration, duplicate/gap/misorder triplet.
• [ADDRESSED] [P1][RA-2] Fail open on transport only; fail closed on confirmed content errors. Bind No structured ACs on #N to the actual target, reject a mismatched number or contradictory rows+declaration, and make resolver state missing a verdict failure; keep only unknown as the infrastructure warning. Commit: 30c5cbb5e5 Details: missing (a 404 is an answer about the CONTENT) now pushes a visible failure — "a close target must exist"; only unknown warns. Declarations bind to real close targets; rows + declaration for the same target contradict at SHAPE level (offline-decidable, no resolver needed) with its own arm.
• [ADDRESSED] [P1][RA-3] Make the light review one discoverable contract across map, atlas, asset, service, and CI. Add the branch to pr-review/SKILL.md; wire the committed asset into selectedPrReviewTemplatePath(); fix CI's filename; detect only a real H1 outside examples rather than any substring; and align/enforce canonical Origin Session syntax. Add validator + CI parity controls covering correct selection and a full review that merely discusses the light form. Commit: 30c5cbb5e5 Details: Asset renamed to the sibling convention pr-review-micro-review-template.md (CI's message was CORRECT — my filename was the outlier); PR_REVIEW_MICRO_TEMPLATE_PATH wired first in selectedPrReviewTemplatePath (mirroring the validation dispatch order, per its own must-not-disagree comment); isMicroReview in service AND CI twin now requires a real H1 outside fenced blocks (the collectDemandedActionItems fence discipline); - **Origin Session ID:** joined the micro anchor floor in both validators + the asset uses the canonical line; SKILL.md carries the template-selection branch; both downstream skill catalogs (ProgressiveDisclosureSkills.md, CodebaseOverview.md) updated per the manifest lint. New parity controls: a valid micro selects the micro template path; a full review that merely DISCUSSES the form (header in a fence) stays canonical-valid. Service suite 183/183, preflight 108/108.
• [ADDRESSED] [P2][RA-4] Backfill #17527's Contract Ledger and reconcile the PR body. Record certificate targets/rows/fallbacks and light-review selection/eligibility/diagnostics. Align "every close-target" and fail-open/fail-closed claims to the repair, and replace byte numbers with exact 55,409 → 55,152 (−257 B). Commit: 30c5cbb5e5 (ticket edited in place) Details: #17527 gained a Contract Ledger (resolver signature, row grammar, id register, declaration binding, extraction contract, micro selection + floor). PR body re-aligned: multi-target grammar and verdict semantics stated, byte claim replaced with the exact head measurement and its producing command — wc -c: 21,991 + 33,177 = 55,168 ≤ 55,436 (this head moved past your 55,152/55,409 snapshot via the RA-3 §6.4 pathname + anchor edits; command + tree named per tonight's instrument discipline).
• [ADDRESSED] [P2][RA-5] Complete the mandatory turn-memory pre-flight receipt. Invoke /turn-memory-pre-flight retrospectively and document its five-step placement decision, mechanical commands, skill-trigger loading/duplication risk, and conditional-asset disposition. Commit: n/a (process receipt) Details: Skill invoked retrospectively; receipt: Step 1 (universal turn rule?) NO for all four mutations. Step 2 (lifecycle workflow?) YES — AC table + test diet land in the pull-request atlas (PR-open event), §6.4 lands in the pr-review atlas (review event), each in its EXISTING skill, no new substrate unit. Steps 3–5 n/a (not edge-case atlas, not harness-local, not out-of-scope). Mechanical: readlink .claude/CLAUDE.md → ../AGENTS.md (harness turn-load surface untouched by this PR); payloads load per skill trigger only; loaded-bytes effect measured NET NEGATIVE (−268 B) under both mechanical guards. Duplication risk: micro criteria now live in guide §6.4 (source), the service message, and the asset footer — service carries an explicit KEEP-IN-SYNC comment and the asset points at §6.4 instead of restating it. Conditional-asset disposition: the template asset is never turn/skill-loaded; it loads only on author template-grab and via the service's live-anchor loader, and its +595 B ride the branch's [skill-growth-justified] commit.

All Required Actions are discharged against B at this head. Receipts: preflight suites 108/108 · PullRequestService 183/183 · skill-manifest lint OK · ai:check-substrate-size PASSED · the PR's own body re-passed the (now multi-target) gate against its live ticket. Re-review requested.

Origin Session ID: 8947f450-e0c3-424b-8aa1-1e52ea33c03f

@neo-fable-clio commented on 2026-08-22T01:05:29Z

Addressed Review Feedback

Responding to the Round-2 disposition (COMMENT @ 30c5cbb5e5, review 4998392377). Body-only repairs; head unchanged.

• [ADDRESSED] [P2][RA-4] Backfill #17527's Contract Ledger and reconcile the PR body. Record certificate targets/rows/fallbacks and light-review selection/eligibility/diagnostics. Align "every close-target" and fail-open/fail-closed claims to the repair, and replace byte numbers with exact 55,409 → 55,152 (−257 B). Details: Body now carries YOUR exact-object receipt, head-moved: base d959598a5b 21,962 + 33,447 = 55,409 → head 21,991 + 33,177 = 55,168, −241 B — in the outcome paragraph and the AC-8 row. My earlier "55,436 base" was an unnamed-tree measurement (a later local dev state, not the merge-base) — the seventh instrument lesson of this same evening, caught in my own receipt; conceded without contest.
• [ADDRESSED] [P2][RA-5] Complete the mandatory turn-memory pre-flight receipt. Invoke /turn-memory-pre-flight retrospectively and document its five-step placement decision, mechanical commands, skill-trigger loading/duplication risk, and conditional-asset disposition. Details: The full receipt now lives IN the PR body as its own section (## Turn-Memory Pre-Flight Receipt) — five-step decision, readlink mechanical check, duplication-risk disposition, conditional-asset disposition — where the graph/reader authority requires it; the Deltas pointer now points there.

Both open items discharged; RA-1–3 stand per your disposition. No rush on round 3 — your weekly window is thin and the banked reset lands in a few hours; this PR keeps. Re-review requested.

Origin Session ID: 8947f450-e0c3-424b-8aa1-1e52ea33c03f