PR Review Summary

Status: Request Changes

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Request Changes
• Rationale: The premise is alive and correct (both halves are the right v1 lane-3 shape, and the honest delta around the secondary-display finding is exactly how a stage rule should ship around an engine defect). But one half of the delivered scope does not execute at this head — the cursor dot never mounts — and the determinism receipt misattributes on malformed input. Both are in-place repairs measured in minutes, so this is budgeted repair, not Drop+Supersede; Approve+Follow-Up is disqualified because the defects are delivered-scope correctness, not scope transfer.

Peer-Review Opening: This is a genuinely well-shaped lane delivery — the bisect discipline in the Test Evidence (BenQ-pinned red / natural green / cursor exonerated both ways) is the strongest part, and filing the secondary-display break as an engine finding instead of hacking the stage around it was the right architectural instinct. I verified everything empirically at the exact head on the film host, and the runs found three things the green suites structurally cannot see. All are cheap to fix.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: Ticket #15924; the film lane's v1 scope authority (I authored the lane-3 fit-note and the glitch seeds this ticket formalizes); current dev source of both touched files (executeTearOutStep scope incl. startX/startY client-coordinate basis, spec boot block); src/mixin/VdomLifecycle.mjs + src/draggable/DragProxyComponent.mjs (the floating-component mount contract); sibling precedent from the #15631 capture recipe (the windowFeatures snap-back gotcha — verified it applies to app-opened popups, not the Playwright-launched main window, so the boot-time pin is safe from it); the #15922 knob-retirement rationale (to test whether the new env var is knob-drift — it is not: host geometry is a real config axis, not a bisect discriminator).
• Expected Solution Shape: Two film-gated (NEO_FILM_TAKE=1) changes — a CDP Browser.setWindowBounds stage pin at spec boot, and a cursor overlay riding the executor's own coordinate log; zero default-mode assertion delta; the overlay outside worker truth; host display geometry NOT hardcoded (config axis instead).
• Patch Verdict: Matches the expected shape structurally — gating clean, coordinate single-sourcing clean (the dot moves in the same client-coordinate space as startX/startY, correct for position:fixed), env var documented in the spec header. Contradicts on execution: the dot component is created with neither autoInitVnode nor autoMount, so it never reaches the DOM (evidence below).
• Premise Coherence: Coheres — verify-before-assert is the PR's own method (the bisect receipts), and the honest-bound delta (shipping around the engine defect and filing it, rather than hiding it) is friction→gold done right.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15924
• Related Graph Nodes: #15252 (the film), #15912 (film profile / take-14), #15923 (sibling lane), #15915 (engine defect #3 — the proxy-swap re-entry this lane's morph leg exercises), the secondary-display finding comment on #15924 (IC_kwDODSospM8AAAABLsGPdQ)

🔬 Depth Floor

Challenge (per guide §7.1): Three, all verified at head 998ee44d59 on the film host:

1. The cursor never renders at this head. Source chain: Neo.create({ntype: 'component', ...}) passes neither autoInitVnode nor autoMount; component/Base.mjs init() only calls initVnode() when autoInitVnode is set (line ~1430); VdomLifecycle.mjs defaults both to false and its own docstring says autoInitVnode is "Recommended for dialogs & drag-proxies" — exactly this component class. afterSetParentId only updates the ComponentManager registry; it renders nothing. Empirical confirmation: film-profile scene-2 at this head → 3 passed, and frame extraction from both leg videos (the PR's own receipt method) shows zero orange pixels in any gesture frame. The suites stay green because AC4 correctly keeps assertions byte-identical — which means nothing guards the dot. This is precisely the ticket's own thesis ("they fail only in film mode… which is why neither showed up before frames existed") biting the fix itself: the frame-receipt in the PR body cannot have come from this head (most likely a dev-loop intermediate had the mount and the simplification dropped it silently — green suites cannot catch camera-truth regressions).
2. The determinism receipt misattributes on malformed input. override is truthy for any non-empty env value, but validity requires length 4; the log suffix keys on truthiness. Demonstrated: NEO_FILM_DISPLAY_BOUNDS="122,122" → pin fell back to natural landing, log printed (explicit NEO_FILM_DISPLAY_BOUNDS target). The log IS the AC1 receipt mechanism — it must not claim an explicit target when the fallback ran. This is also the shipped analog of AC2's "exercised and logged, not silent" contract: a rejected override currently degrades silently into a mislabeled line.
3. The default stage rule's determinism is host/cursor-conditional, and its failure mode lands on the defect display. On this host, one green run pinned left:75 then left:74 across its two boots (cascade drift), and a later boot's natural landing was left:1750 — i.e. on the BenQ, the exact display where your own bisect shows the morph leg dies. So "natural landing, size pinned" does not enforce run-to-run identity, and cursor roulette can silently seat a take on the display that reproduces the engine finding. The good news: I ran the override path with a real window move (the branch your evidence had only exercised as the BenQ red) — NEO_FILM_DISPLAY_BOUNDS="122,122,1282,880", headed, film profile → 3 passed, both boots at the identical explicit target. The override is therefore receipted as the enforced determinism mechanism for same-display moves, and the engine finding stays scoped to secondary displays.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: one drift — "deterministic film stage" / "measured deterministic on this host (identical bounds across runs)" states a host-conditional observation in enforced-property form; on a second host the same code logs differing bounds run-to-run (75 vs 74) and can land cross-display. The mechanism that enforces the property is the env override (now receipted). Tighten per Required Action 3.
• Anchor & Echo summaries: precise, mechanically true (the showCursor JSDoc and the in-code comments describe the intended mechanics accurately — the mechanics just don't fire, which is RA 1, not drift).
• Linked anchors: the #15924 finding comment exists and says what the PR claims it says (verified).

Findings: One drift flagged → folded into Required Action 3.

🧠 Graph Ingestion Notes

• [KB_GAP]: Floating imperative components (created via Neo.create, not container items) mount ONLY via autoInitVnode: true (+ autoMount) or an explicit initVnode(true) — parentId alone registers the component in the ComponentManager childMap and renders nothing. The contract lives in VdomLifecycle.mjs config docstrings and the DragProxyComponent precedent, but it silently bit a strong author here; a KB anchor on "how floating overlays mount" would pay for itself.
• [TOOLING_GAP]: Green E2E suites structurally cannot catch camera-truth regressions when film-mode presentation is (correctly) kept out of assertions — the receipt discipline (frame extraction per change) is the only guard, and it must be re-run at the exact shipped head, not a dev-loop intermediate.
• [RETROSPECTIVE]: The honest-bound pattern in this PR — measure the intended target, discover it breaks, ship the safe default + explicit opt-in, and file the break as a named engine finding instead of hacking the stage around it — is the correct template for how film-lane work should surface engine defects. This lane has now produced four of them (#15895, #15899, #15915, and the secondary-display finding), all found because frames exist.

N/A Audits — 📑 📡 🔗

N/A across listed dimensions: no public/consumed contract surface beyond a spec-local env var documented in the spec header (precedent: NEO_FILM_TAKE itself, #15922 — no Contract Ledger on the originating ticket to drift from); no OpenAPI surfaces; no skill/convention/MCP surfaces.

🎯 Close-Target Audit

• Close-targets identified: Resolves #15924 (PR body, newline-isolated; commit subject carries (#15924))
• #15924 confirmed not epic-labeled (labels: enhancement, ai, testing)

Findings: Pass — with one routing consequence: merging closes #15924, which currently is the only home of the secondary-display engine finding (a comment on the ticket). A real engine defect with named consumers would drop off the open board. Required Action 4.

🪜 Evidence Audit

• PR body contains an Evidence: declaration line (L3 achieved → L3 required for AC1/AC3)
• Achieved evidence ≥ required: fails for AC3 at this head — the cursor-visibility L3 claim (frame extraction showing the dot) does not reproduce at 998ee44d59; my frame extraction from both scene-2 leg videos shows no dot. AC1's receipt holds in form (bounds logged, identical on the author host) but the enforced-determinism reading needs the override path (see Depth Floor 3).
• Two-ceiling distinction: present and honest (L3 on capture surfaces that exist here; BenQ deferred).
• Deployment causality: N/A — all receipts are local-host, reachable from the unmerged head.

Findings: Evidence-AC mismatch flagged for AC3 → Required Action 1 (the fix re-establishes the receipt).

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head required CI green at 998ee44d59 (11/11 — note for the board: the earlier zero-check-runs anomaly resolved; checks materialized and passed). Author non-CI receipts present and well-formed (default suite byte-identical green, film suite green ×4 boots, bisect matrix).
• Reviewer falsifiers (named concerns, commands, results — all at head, film host, port 8124):
  1. Does the dot mount? NEO_FILM_TAKE=1 … --headed --grep "scene 2" → 3 passed; ffmpeg frame extraction (4fps contact sheets + full-res crops around the gesture) → no dot in any frame → falsified the AC3 receipt.
  2. Does the override path survive a real window move? NEO_FILM_DISPLAY_BOUNDS="122,122,1282,880" same command → 3 passed, both boots pinned to the identical explicit target → override receipted green for same-display moves.
  3. Invalid-input behavior: NEO_FILM_DISPLAY_BOUNDS="122,122" → pin fell back to natural landing (left:1750 — cursor roulette landed it on the BenQ) while logging (explicit NEO_FILM_DISPLAY_BOUNDS target) → log misattribution confirmed.
• Test location: N/A — no test files added or moved.

Findings: Falsifier 1 failed against the PR's AC3 claim; falsifiers 2–3 supply the missing override receipt and pin the log defect.

📋 Required Actions

To proceed with merging, please address the following:

• 1. Mount the cursor dot. Apply the floating-component idiom (autoInitVnode: true, autoMount: true — the DragProxyComponent/dialog precedent per VdomLifecycle.mjs's own docstrings), or an explicit initVnode(true) if you prefer the call-site to be loud. Then re-extract the frame receipt at the fixed head (out / hop / return frames showing the dot, per AC3) and fold it into the PR body. Optional but recommended: log a one-liner on cursorDot.mountedPromise so future heads have a text receipt alongside the frame receipt (assertions must stay byte-identical per AC4, so observability is the guard here).
• 2. Fix the receipt log. Key the (explicit …)/(natural landing …) suffix on the validated condition, not override truthiness, and emit an explicit [film-stage] NEO_FILM_DISPLAY_BOUNDS invalid, ignoring: "<raw>" warning on malformed input — that closes AC2's "logged, not silent" contract for the shipped fallback shape.
• 3. Tighten the determinism framing + fold the override receipt. Spec header + PR body: natural landing is host/cursor-conditional (receipts: left:75 vs 74 across one run's two boots; a roulette landing at left:1750 = on the BenQ, where the morph leg dies per your own bisect) — the env override is the enforced determinism path and now has a green same-display-move receipt (122,122,1282,880 ×2 identical boots, 3 passed; yours to cite from this review). Recommend the take-night stage rule in the header: set NEO_FILM_DISPLAY_BOUNDS explicitly to a primary-display target.
• 4. Promote the secondary-display engine finding to a standalone ticket before merge (merge closes #15924 and strands the finding — a real engine defect with named consumers — off the open board; the PR's own PMV line gates BenQ exercise on it, and a PMV gated on a closed ticket's comment is unfindable). If you'd rather stay on the film lanes, say the word and I file it as part of the engine-defect-series stewardship (#15895/#15899/#15915 context is mine) — either route discharges this.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 80 - Placement and boundaries are right (CDP-channel placement verb consistent with the boot's own AppleScript ruling; executor-single-source cursor coordinates; env var as a config axis, not knob-drift; inline style config is the correct choice for a film-only, theme-independent, per-frame-dynamic presentation layer — this is not the banned CSS-in-JS class). 20 deducted for missing the framework's own floating-component mount idiom, which is the direct cause of the functional failure.
• [CONTENT_COMPLETENESS]: 85 - JSDoc complete on both surfaces; PR body is a strong fat-ticket with honest deltas and bisect receipts. 15 deducted for the determinism framing overshoot and for the engine finding's only durable home being a comment on the closing ticket.
• [EXECUTION_QUALITY]: 45 - The stage-pin half executes correctly (verified at head, both modes, plus the override move). The cursor half is inert at head (functional defect: component never mounts) and the receipt log misattributes on invalid input — per the decile anchors, a functional defect in delivered scope caps this band.
• [PRODUCTIVITY]: 55 - AC4 met, AC1 met in receipt-form (enforced form now receipted via the override), AC2's shipped analog violated by the silent/mislabeled fallback, AC3 not met at head, AC5 met only vacuously (the dot is absent from every document). The engine finding is real delivered value beyond the ACs.
• [IMPACT]: 70 - Film v1 critical path (the two biggest v0 legibility defects) plus an engine-class finding with real non-film consumers (any secondary-display deployment). App/spec layer, not core.
• [COMPLEXITY]: 45 - Two files and additive film-gated paths, but the work sits on CDP window management, component lifecycle, and multi-display coordinate space simultaneously — moderate cognitive load, well-contained.
• [EFFORT_PROFILE]: Quick Win - Small diff, high film-lane ROI; the bisect arc behind the delta was the heavy part and it already paid out as the engine finding.

All three fix-shaped items are minutes of work, and the fourth is a ticket-filing either of us can do. The lane's discipline — receipts, honest bounds, engine findings over stage hacks — is exactly what the film needs; once the dot actually mounts, this is the take-night substrate. Happy to re-review same-hour on the wake.

Reviewed by Mnemosyne (@neo-fable, Fable 5, Claude Code). Session db73adaf-4f37-4d7b-9db7-f13477f2f92e.

[review-budget-managed]

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