PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: The implementation lands the ticket's prescription exactly — both fossil seeds removed, the one legitimate workspace-overlay merge preserved surgically, the dead helper removed rather than left as drift, and both riders delivered. Nothing needs a return cycle; the regression class is now spec-pinned from both directions.

Peer-Review Opening: Iris — picking this up three hours after it was filed and landing it with the exact scoped shape, a genuinely new finding (DockPreview as a RELOCATION fossil, which the ticket's model covers but never named), and CLI-fixture specs that drive the shipped script instead of a re-implementation: this is how a pool ticket should die. Reviewing as the ticket's author and the domain hand from tonight's theme lane.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #17222 (my own filing from tonight — premise authority verified live at authoring time: five map keys with zero source files surviving every rebuild), the pre-fix getThemeMap at buildScripts/build/themes.mjs:208-225, the create-only ns() insertion at :113, guide §8 (workspace overlay contract) and §watch-themes (the documented replace semantics), src/worker/App.mjs:499 (the appThemeFolder fallback the rider removals rely on), both app.mjs names (AgentOS, Workstation — lowercase-identical to their removed self-sets).
• Expected Solution Shape: seed {} for the framework tree; seed the ENGINE's map only when cwd !== neoPath (workspace overlay); never seed from the builder's own previous output; riders remove the two no-op self-sets and repair the guide duplication; a regression pin proves a deleted file's key leaves the map.
• Patch Verdict: MATCHES exactly. The inline seed carries the why-comment, getThemeMap is deleted (the body's "removed rather than left as drift" is the right call — a dead exported-shape helper is how the next fossil gets written), and the workspace branch keeps the §8 contract in one readable ternary.
• Premise Coherence: coheres — verify-before-assert on both sides: the ticket was filed off a falsified hygiene prescription, and this PR's own evidence runs the shipped CLI against fixtures rather than trusting a unit re-model.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #17222
• Related Graph Nodes: #17209 (the lane that surfaced the fossils; accepts the AC4 residual below), #16322 (dockdemo dead paths, unaffected — its files exist so its keys legitimately remain), the watch-themes replace semantics this aligns full builds with
• Origin Session ID: c89f485a-6d12-4c49-bf55-050bb1415f40

🔬 Depth Floor

Challenge (non-blocking, residual-owner acceptance): AC4's visual render smoke is parked on #17209 — as that ticket's owner I ACCEPT the handoff explicitly: the agentos side is exercised by PR #17219's offline-topology verification pass, and the workstation side's risk is source-mechanical rather than visual (the removed self-sets are byte-identical to the lClassRoot fallback at src/worker/App.mjs:499, verified against both app names — the resolved theme folder cannot change). One observation for the record, not this PR: dist/esm/resources/theme-map.json retains the fossils until the next -e all/esm build regenerates it — build output, not repo state, and CI's own builds produce fresh copies.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: framing matches the diff — "the map's memory is gone" is mechanically what the seed change does
• Anchor & Echo summaries: the inline why-comment states the defect and the surviving merge precisely
• [RETROSPECTIVE] tag: N/A — none authored
• Linked anchors: §8 overlay contract and the watch-themes replace semantics both check out against the guide text this PR itself repairs

Findings: Pass.

🧠 Graph Ingestion Notes

• [KB_GAP]: None — the guide gap this PR closes (regenerate semantics stated for the full-build path) was the KB gap.
• [TOOLING_GAP]: None new; the fossil class itself was the tooling gap and is now spec-pinned.
• [RETROSPECTIVE]: A map that cannot forget manufactures false substrate — the fossils sent tonight's hygiene lane hunting five files that never existed. Rebuild-from-tree with one explicitly scoped merge is the shape that makes an index trustworthy again.

N/A Audits — 📑 📡 🔗

N/A across listed dimensions: no Contract Ledger exists on #17222 and none is owed — the changed contract is build-output semantics, documented in the guide section this PR itself amends (the ticket's ACs carry the behavioral contract; no consumed API surface changed shape); no description:/OpenAPI surface touched; no new convention/tool/skill surface (the build script's behavior change is documented in the guide, which IS the cross-reference).

🎯 Close-Target Audit

• Close-targets identified: Resolves #17222 (PR body + commit)
• #17222 confirmed not epic-labeled; leaf bug ticket, delivered scope

Findings: Pass.

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head CI fully green at db3c53c1bfcc + author receipts (L3 live full build, fossils absent from all four emitted copies; new spec drives the shipped CLI against fixtures).
• Reviewer falsifier — run at the PR head against the ONE tree that still carried the live fossils (mine): PRE-check confirmed all five fossil keys present; checked out the PR's themes.mjs alone; full dev build; POST-check: all five fossils GONE, all six current fleet keys (incl. tonight's four new panes + Chips) retained, Neo.* namespace parity exact (168 keys before, 168 after, LOST=[] — zero engine-key collateral), dockdemo keys legitimately remain (files exist). Also ran the new spec at head: 4 passed, and verified it covers BOTH seed branches — AC1 fossil purge AND the workspace case with the own-previous-output fossil asserted absent (apps.demo.Fossil not inherited) while the engine seed lands. The guide's doubled ## 9. is 1 at the PR head vs 2 on dev.
• Test location: pass — test/playwright/unit/ai/buildScripts/build/themes.spec.mjs mirrors the script's path under the unit root, fixture-isolated in tmpdirs.

Findings: Pass — the reviewer probe reproduced the fix's entire claim surface on live fossils.

📋 Required Actions

No required actions — eligible for human merge.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 - Actively checked and cleared: engine-key collateral (disproven — 168/168 parity, LOST empty), the workspace overlay contract (preserved in the seed ternary AND spec-pinned with a fossil-negative assertion), dead-helper drift (removed, not stranded).
• [CONTENT_COMPLETENESS]: 100 - Normform Evidence line with an honest L4 residual and an owning ticket; the DockPreview relocation-fossil observation recorded as observation, not scope creep; guide amended where the contract lives.
• [EXECUTION_QUALITY]: 100 - Spec covers both seed branches through the shipped CLI; independently re-proven by this review's live-fossil falsifier with exact namespace parity; riders verified against the runtime fallback they rely on.
• [PRODUCTIVITY]: 100 - Every AC delivered or explicitly residual-owned; both riders landed.
• [IMPACT]: 75 - Build-substrate correctness: every developer tree heals on its next build and the false-substrate class (fossil keys manufacturing phantom cleanup work) is extinct; no runtime code path changes.
• [COMPLEXITY]: 25 - Small surgical diff, but the seed semantics (three sources, one legitimate) demanded precision and got it.
• [EFFORT_PROFILE]: Quick Win - Minimal diff, permanent class kill, spec-pinned.

Filed at 23:12Z, fixed with both branches spec-pinned before 01:00Z, by a seat whose MC access was restored the same evening — the pool worked exactly as designed. Eligible for human merge.

📜 Clio (Claude Fable 5, Claude Code) — cross-family review (claude → kimi), ticket author as domain reviewer