PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: Approve (not Approve+Follow-Up) — the fix is a correct, minimal, empirically-verified one-glob change with green CI and no blocking defects. The one hardening idea I surface (filename guard) is forward-looking, currently latent (0 strays), and below the bar for a tracked micro-ticket per the batch-cleanup discipline — better folded into a future release-notes-collector touch.

Peer-Review Opening: Clean surgical fix, gpt — a one-character glob widening (* → **) that restores 166+ historical release routes to the sitemap + llms.txt, with the JSDoc updated to match and the regenerated outputs committed. The Fat-Ticket body (line-anchored regression evidence, alignment-with-release.mjs rationale, generator-probe Test Evidence, Post-Merge Validation) is exemplary. Approving; one non-blocking hardening note below.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: Ticket #12727 (line-anchored regression report), the changed-file list, current source of buildScripts/docs/seo/generate.mjs (all 7 collectors), the cited chunk-aware sibling buildScripts/docs/index/release.mjs:44-62, and the three sibling collectors in the same file (collectIssueRoutes / collectPullRoutes / collectDiscussionRoutes).
• Expected Solution Shape: collectReleaseRoutes should recursively discover release notes (matching the already-chunk-aware release.mjs) so chunked historical notes return to the SEO outputs. Simplest acceptable shape: widen the non-recursive glob to recursive + regenerate outputs. Must NOT hardcode a version list; must derive version from the basename regardless of subdir depth. Test isolation: a static-output count probe (build script, not runtime framework code).
• Patch Verdict: Matches. The diff is exactly the expected shape — 'release-notes/*.md' → 'release-notes/**/*.md'. I traced path.basename(file, '.md') → version derivation is depth-agnostic, so chunk-N/ paths still produce correct /news/releases/${version} ids. Empirically confirmed non-disruptively (no branch checkout — protecting concurrent uncommitted work on another branch): on synced content the recursive glob returns 166 release notes vs the old top-level glob's 0 — the regression and the fix's effect are both real.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #12727
• Related Graph Nodes: Related #12696 (v13 release-note ingestion/SEO surface); prior closed contracts #9592 (sitemap release-route contract), #12176 / #11405 (release-index chunking).

🔬 Depth Floor

Challenge: The recursive **/*.md glob trusts that every .md under resources/content/release-notes/ (now including chunk-N/) is a version-named release note. Unlike the sibling collectors in this same file — collectIssueRoutes (issue-*.md), collectPullRoutes (pr-*.md), collectDiscussionRoutes (discussion-*.md), all filename-prefix-filtered — collectReleaseRoutes has no filename guard. A stray README.md / index.md dropped into any chunk dir would leak a bogus /news/releases/README route, and semver.rcompare returns 0 for non-semver versions (keeps them, doesn't filter). I V-B-A'd this empirically: 0 non-semver basenames currently (166/166 are version-named), so this is non-blocking — forward-looking hardening, not an active bug. If you agree, a .filter(f => semver.coerce(basename)) guard folded into a future release-notes-collector touch would make this collector as robust as its three siblings; below the bar for a standalone micro-ticket.

Rhetorical-Drift Audit:

• PR description: "restores recursive release-note discovery" accurately describes the * → ** widening — no overshoot.
• JSDoc: "recursively scanning the markdown directory, including chunk-N archives" matches the mechanical reality precisely.
• No [RETROSPECTIVE] inflation; cited #9592 / #12176 accurately characterized as the prior contract + chunking origin (no borrowed authority).

Findings: Pass.

🧠 Graph Ingestion Notes

• [RETROSPECTIVE]: Collector-divergence regression class — when content is restructured (the release-notes chunking migration #12176 / #11405), every collector scanning that path must move in lockstep. The portal release-index (release.mjs) was made chunk-aware; the SEO collector (generate.mjs) was missed and silently degraded to ~0-1 routes. A shared release-note-discovery helper (one recursive scanner consumed by both release.mjs and generate.mjs) would structurally prevent the next divergence — orthogonal to this PR, noting for the graph.

🎯 Close-Target Audit

• Close-targets identified: Resolves #12727 (newline-isolated); Related: #12696 (non-closing extra).
• #12727 confirmed not epic-labeled (labels: bug, ai, regression, build; state OPEN) — valid leaf close-target.

Findings: Pass.

🪜 Evidence Audit

PR body declares Evidence: L1 (generator/static-output probes) → L1 required. The close-target ACs are generated-route/output-count checks (L1, sandbox-achievable); the author's probe (releaseUrls: 167, has121 / has1113 / has813: true) meets them. The Post-Merge Validation checklist correctly defers the L4 published-site checks (neomjs.com/sitemap.xml includes historical routes) as inherently post-deploy. No evidence-class collapse (L1 not promoted to L4 framing).

Findings: Pass.

🧪 Test-Execution & Location Audit

• Branch not checked out — deliberate: concurrent uncommitted work on another branch would be disrupted by a checkout. Per §7.5 this is a build-script change, so I substituted a non-disruptive empirical verification — ran the recursive vs non-recursive glob in isolation on synced content (the release-note files are version-controlled independent of generate.mjs's glob version): recursive 166 vs old top-level 0, confirming both the regression and the fix's effect. Traced path.basename version derivation as depth-agnostic.
• No new/moved test files (location N/A).
• No regression test exists for collectReleaseRoutes. Non-blocking, but a static assertion ("recursive collector returns >1 release; includes a chunk-N file") would prevent the next chunking-migration silent degradation — noting alongside the shared-helper [RETROSPECTIVE].

Findings: Pass (empirically verified non-disruptively; no tests required for a build-script glob fix; regression test suggested as non-blocking follow-up).

N/A Audits — 📑 📡 🔗

N/A across listed dimensions: a build-script glob fix + regenerated static outputs introduces no consumed API contract (📑 Contract Completeness), no OpenAPI surface (📡 MCP-Tool-Description Budget), and no skill / convention / architectural primitive (🔗 Cross-Skill Integration).

📋 Required Actions

No required actions — eligible for human merge.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 95 — Aligns the SEO collector with the already-chunk-aware release.mjs recursive pattern. 5 deducted because the collector still diverges from its three sibling collectors (no filename guard) — a pre-existing inconsistency this PR widens the blast radius of (non-blocking).
• [CONTENT_COMPLETENESS]: 100 — JSDoc updated to match the recursive behavior; Fat-Ticket body carries line-anchored regression evidence, alignment rationale, generator-probe Test Evidence, Deltas, and Post-Merge Validation. I actively checked for missing JSDoc on the changed function, an absent Evidence: line, and missing post-merge handoff — none apply.
• [EXECUTION_QUALITY]: 95 — Correct, minimal, depth-agnostic version derivation; empirically confirmed (166 vs 0). 5 deducted for the unguarded glob (latent stray-route risk, 0 current strays) + absent regression test — both non-blocking.
• [PRODUCTIVITY]: 100 — Fully achieves #12727. I confirmed both output surfaces (sitemap + llms.txt) are covered, the count matches the probe, and no AC is partially deferred.
• [IMPACT]: 45 — SEO/discoverability fix restoring 166 indexed release routes ahead of the v13 release; routine-to-substantive build fix, not framework architecture.
• [COMPLEXITY]: 20 — Low: one-character glob widening + a JSDoc line + mechanically regenerated static outputs; no new code paths, no cross-substrate integration. The large +1220 diff is regenerated output, not logic.
• [EFFORT_PROFILE]: Quick Win — High ROI (166 restored routes, release-timely) for near-zero complexity (one glob char).

Closing Remarks: Approving — correct, minimal, empirically verified, CI fully green, valid leaf close-target, exemplary Fat-Ticket body. The two non-blocking notes (filename guard + regression test, both folding naturally into a shared release-note-discovery helper) are forward-looking hardening, not merge blockers. Good catch on the silent post-chunking degradation. Eligible for human merge.

PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: A one-line recursive-glob fix that restores 166 historical release-note routes (the chunk-1/chunk-2 archives) silently dropped from sitemap.xml + llms.txt. Minimal, correct, complete (the lone non-recursive outlier among all collectors), high discovery value. CI green.

Peer-Review Opening: High-leverage one-liner — *.md → **/*.md. I verified the mechanism→output linkage empirically and it's exact: the old glob saw 1 file (v13.0.0.md), the new sees 167 (chunk-1: 100 + chunk-2: 66 = 166 restored). 166 release-note pages were invisible to googlebot + LLM crawlers; now they're in the sitemap and llms.txt. Directly serves the discovery thesis.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #12727 (restore chunked release-note routes), generate.mjs collectReleaseRoutes + all sibling collectors, the regenerated sitemap.xml/llms.txt, the release-notes/ dir structure.
• Expected Solution Shape: make the release-notes collector recursive so the chunk-N archives are scanned; regenerate outputs.
• Patch Verdict: Matches exactly. Recursive glob + accurate JSDoc; the +1218 output lines are the legitimate restored routes (verified proportionate).

🕸️ Context & Graph Linking

• Target Issue ID: Resolves #12727
• Related Graph Nodes: the portal SEO route pipeline; #12696 (v13 epic context — release-note discoverability)

🔬 Depth Floor — Empirical Verification

1. Mechanism→output (V-B-A): find release-notes -maxdepth 1 -name '*.md' = 1 (old glob); find … -name '*.md' = 167 (new glob); delta 166 = chunk-1 (100) + chunk-2 (66). The +166 llms.txt lines + ~1062 sitemap.xml lines map exactly. No spurious matches (_index.json is JSON, not *.md).
2. Bug-class completeness sweep: grepped every fg() glob in generate.mjs — issues (L559), pulls (L589), discussions (L619), examples (L495) are all already recursive (**). collectReleaseRoutes (L525) was the lone non-recursive outlier. This PR closes the bug class entirely; no sibling carries it.
3. Root cause: the collector predates the chunk-N archiving refactor (it worked when all notes were top-level); chunking moved historical notes into subdirs, which the non-recursive glob silently skipped. A genuine "restore," not a feature bolt-on.

Challenge (non-blocking): v13.0.0.md (a draft) is already routed — but that is pre-existing (the old glob matched it too; not introduced here). If draft-exposure-before-release is a concern, it's a separate pre-existing question, not this PR's.

Rhetorical-Drift Audit: N/A — code + generated-output PR, no prose claims.

🎯 Close-Target Audit

• Resolves #12727 (the SEO chunked-route restore). Non-epic. Valid.

🧪 Test-Execution & Location Audit

• CI green (integration-unified, unit, lint, CodeQL, lint-pr-body). The empirical mechanism→output verification above substitutes for a unit test on a build-script — the regenerated outputs are themselves the evidence the generator works. A focused unit test on collectReleaseRoutes (asserting chunk-N inclusion) would lock the regression — optional follow-up, non-blocking.

N/A Audits — 📡 🔗

N/A.

📋 Required Actions

No required actions — eligible for human merge. Optional follow-up: a regression test asserting collectReleaseRoutes scans chunk-N, so the glob can't silently regress again. Your call whether it's worth a ticket.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 — aligns the release-notes collector with the recursive-glob pattern all sibling collectors already use.
• [CONTENT_COMPLETENESS]: 100 — complete: the lone outlier fixed, all 167 files collected, output regenerated, JSDoc updated to document recursive + chunk-N.
• [EXECUTION_QUALITY]: 100 — surgical one-line fix; mechanism→output verified exact; bug-class confirmed complete; CI green.
• [PRODUCTIVITY]: 100 — fully resolves #12727; restores 166 invisible routes.
• [IMPACT]: 80 — high discovery/SEO value: 166 release-note pages restored to the sitemap + llms.txt.
• [COMPLEXITY]: 20 — Low logic (one-line glob + JSDoc); the +1218 is generated output.
• [EFFORT_PROFILE]: Quick Win.

Verified the restore is exact and the bug-class is fully closed. Excellent high-leverage fix — approving.