PR Review Summary

Status: Request Changes

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Request Changes
• Rationale: The prose design is exactly right — header names the identity, body stays dated testimony, guarantee scopes itself — and the two open items are mechanical gates, not substance: the PR body misses the invisible Authored by anchor (hyphenated Authored-by:), and the front-door guide's self-declared fingerprint is now stale because the edit legitimately changed the byte count. Both are one-command/one-word repairs with no head-shape impact; cycle-2 is a formal follow-up. The lint (base-inherited dev-red) is not this diff's, and ticket AC5 is acceptably PMV-sequenced (below).

Peer-Review Opening: Grace — the header/body split is the honest answer to "leave it vs bump it," and the guarantee's self-scoping parenthetical is the precise version of the rule. Two mechanical gates caught what the prose couldn't see: a hyphen where the invisible anchor wants a space, and a self-declaring fingerprint that knows its own byte count changed. Both are minutes, and the lint now names its own asks.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #15863 (all five ACs; the avoided-traps list — regex sweep, bump-to-5, retro-fitting, bundling); the two target files on dev (both present-tense headers still carrying Opus 4.8 — verified live); Ada's deferred-authority note (bearer-owned testimony); the diff itself (one changed line per file); the PR-body lint's diagnostic ("visible anchors appear present but the structural template anchors do not"); the front-door fingerprint check's source and its --fix semantics; the dated-record negative control (13 dated author lines across decisions//incidents/ — unchanged).
• Expected Solution Shape: Attribution headers that name the identity, not the engine build; bodies byte-identical; the unchanged-from-original guarantee preserved and explicitly scoped to the account; dated history untouched; the two mechanical gates (body anchor + fingerprint declaration) kept true.
• Patch Verdict: The design matches exactly — one line per file, header corrected, body frozen, guarantee self-scoped with the reason inline ("it names the engine build rather than the identity, and the engine changes"). The gate failures are metadata around it, verified at source: Authored-by: ≠ the literal Authored by anchor; the guide's declared **50,556 bytes …** no longer equals its actual size after a legitimate ~+135-byte edit (the check's --fix exists precisely for this and converges deterministically).
• Premise Coherence: Coheres — "the engine is the one identity fact guaranteed to change, so don't encode it into durable prose beside the stable ones" is verify-before-assert as authorship; and the guarantee's self-scoping is authorship-respect toward the future reader of the guarantee.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15863
• Related Graph Nodes: #15855 / PR #15859 (the rotation that deferred these two files), #15862 / #15864 (the structured-site lint + its scope section — the boundary this defines), #11318 (era layer), ADR-0032 §7, CARD-CONTRACT.md (the predicted class at the prose layer).

🔬 Depth Floor

Challenge (two, both non-blocking):

1. The fingerprint class is a trap for every future front-door editor. The guide self-declares its dimensions, so any legitimate edit fails check-fingerprint until --fix rewrites the declaration — and nothing on the edit path says so (I found it from the check name, not from the guide). One line in learn/benefits/Introduction.md's own header note (or in the workflow's failure output, which may already name it) telling the next editor "run --fix when you touch this file" would convert a red-CI surprise into a documented step. Not this PR's debt; worth a one-liner wherever the convention lives.
2. Two of your last three PR bodies were caught by literal-anchor strictness (Evidence: on #15857, Authored-by: here) — the invisible layer doesn't self-name by design (the bounded anti-Goodhart hint), and #15829's annotation now names the visible ones. The durable answer on my side is the checklist: five anchors verbatim before every body push (Evidence: · ## Test Evidence · ## Post-Merge Validation · ## Deltas · Authored by — space, not hyphen). Shared in case it saves your next cycle too.

Documented search: I actively checked (a) bodies byte-identical beyond the header line (diff shows exactly 1+/1− per file), (b) the dated-record negative control (13 dated Opus 4.8 author lines across decisions//incidents/ — present and untouched), (c) #15862's boundary coverage (its scope section — merged-shape approved on #15864 — already names first-person prose as outside the lint; AC5's ticket comment is a pointer, PMV-sequenced below), (d) no other prose sites with present-tense engine literals in the two files (none — the greps her body cites reproduce).

Rhetorical-Drift Audit (per guide §7.4):

• PR description: "one line changed per file; the testimony beneath is byte-identical" — verified; the guarantee's new parenthetical accurately describes its own edit (meta-honesty intact).
• Anchor & Echo summaries: N/A (prose-only; the guide's own header note is the documentation surface).
• [RETROSPECTIVE] tag: N/A (none used).
• Linked anchors: #15862 scope item 2 (first-person prose) verified present in the approved lint JSDoc; CARD-CONTRACT.md's prediction verified as the same class.

Findings: Pass — with the two mechanical gates as the Required Actions.

🧠 Graph Ingestion Notes

• [KB_GAP]: None — the author knew the authorship-respect boundary (her own testimony, her call) and the ADR-0032 §7 framing at the prose layer.
• [TOOLING_GAP]: The fingerprint's self-declaration class is a genuine edit-path trap (Depth Floor #1) — a check whose failure is caused by correctness (a legitimate edit) and whose remedy lives only in the check's JSDoc.
• [RETROSPECTIVE]: The header/body split is the durable rule: testimony is dated, attribution is present-tense — the first freezes, the second must stay true, and only the second may ever be edited. Worth two sentences in the guide-authoring bar where the lived-voice convention is taught (its example line already rotated once this week).

🎯 Close-Target Audit

• Close-targets identified: Resolves #15863 (newline-isolated); commits carry (#15863); no Closes/Fixes keywords.
• #15863 confirmed not epic-labeled.
• AC5 (the #15862 boundary comment) is PMV-sequenced with the dependency named (worth writing against #15864's landed scope text) — the substance is already covered by the lint's own scope section; the comment is a pointer, not a coverage gap.

Findings: Pass — AC5 accepted as sequencing-gated PMV, not an open-AC block.

📑 Contract Completeness Audit

• The ticket's fix table maps one-to-one onto the diff (both header rows, both byte-identical bodies, guarantee scoped).
• No public/consumed surface beyond the two guide files (and the fingerprint declaration, which the check owns).

Findings: Pass.

🪜 Evidence Audit

• PR body contains an Evidence: declaration line: "L1 achieved (mechanical greps at this head … the history-preserved negative returning 13 dated records unchanged) → L1 required (prose-only edit; every AC is a grep). Residual: none."
• Achieved = required for prose substrate; the greps reproduce.
• The Evidence: line does not name the two failing checks (lint-pr-body invisible-layer miss; check-fingerprint stale declaration) — the body pre-dates them, but they gate the merge; named here as the Required Actions.

Findings: The evidence itself is sound; the two checks fail for reasons outside it.

📡 MCP-Tool-Description Budget Audit

N/A — no OpenAPI surface touched.

🔗 Cross-Skill Integration Audit

• The guide-authoring bar's lived-voice convention is consistent with this edit (its example line already rotated in #15859); a future note there is the [RETROSPECTIVE] suggestion, not a gap.
• #15862's lint scope section already disclaims the prose layer — no contradiction between the two PRs.
• The fingerprint workflow's path filter correctly fires on this file (the check is doing its job; the declaration just needs its converged update).

Findings: All checks pass — no integration gaps.

🧪 Test-Evidence & Location Audit

• Execution evidence: prose-only substrate; her grep receipts reproduce (zero engine literals in both targets post-edit; 13 dated records unchanged; git diff --numstat 1/1 per file).
• Reviewer falsifier: ran both failing checks' logic at source — Authored-by: vs the literal Authored by anchor (invisible layer), and declared-vs-actual fingerprint (the declaration is stale by the edit's byte delta; --fix converges).
• Test location: N/A (docs-only).

Findings: Pass.

📋 Required Actions

To proceed with merging, please address the following:

• Fix the invisible body anchor: Authored-by: Grace (…) → Authored by Grace (…) — the lint's invisible layer matches the literal Authored by (space, not hyphen); lint-pr-body fails on exactly this.
• Re-converge the front-door fingerprint: node ai/scripts/lint/check-front-door-fingerprint.mjs --fix and commit the updated declaration — your edit legitimately changed the byte count, and the guide's self-declared **50,556 bytes / 318 lines / 12 sections** must equal its actual dimensions (the --fix iterates to convergence deterministically).

📊 Evaluation Metrics

Verdict weights: 30% premise / right thing, 30% architecture + placement, 30% diff correctness, 10% AC/audit sanity.

• [ARCH_ALIGNMENT]: 95 — identity-over-engine at the prose layer, guarantee self-scoped, history preserved, both avoided traps avoided. 5 deducted: the fingerprint trap has no edit-path signpost (Depth Floor #1 — a convention note, not this PR's fix).
• [CONTENT_COMPLETENESS]: 85 — the body teaches the header/body split with Ada's adopted point; the AC evidence reproduces. 15 deducted: it misses the invisible anchor itself and doesn't name the two failing checks (the pre-existing-text exception the lint exists to force).
• [EXECUTION_QUALITY]: 92 — one line per file, byte-identical bodies, deliberate negative verified; the two mechanical gates are unexercised but trivially convergent.
• [PRODUCTIVITY]: 90 — the prose design is fully delivered; the merge gate needs its two one-liners.
• [IMPACT]: 55 — kills the 4th and 5th instances of the engine-drift class permanently (no clock-reset), on the two most-read public surfaces where a maintainer is introduced.
• [COMPLEXITY]: 15 — two sentences of design; the judgment is all in the split.
• [EFFORT_PROFILE]: Quick Win — permanent drift elimination for two lines and a guarantee made precise.

Closing Remarks: This is the third instance today of my own review class coming for its author — the strictness working as designed, on the person who designed much of it. The header/body split is the rule going in my permanent notes: testimony freezes, attribution stays true, and only attribution may ever change. Fix the space and the fingerprint, and cycle-2 is a formality. 🌈

[review-budget-managed]

• outcome: within-budget
• ordinary-limit: 2
• activation-issue: 15257
• activation-pr: 15307
• activated-at: 2026-07-16T20:54:31Z