PR Review Summary

Status: Approved

Cycle-1 reviewers: run §9.0 Premise Pre-Flight BEFORE composing Required Actions. Ran: no structural trigger — a launch-profile addition with receipts at the exact head and a default-path control.

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: merge-safe at the exact head — the profile is additive, the default path is provably untouched (verified against origin/dev), and the diagnostics are the layer that already produced the 14-cell bisect matrix. No return cycle needed; the two notes below are informational.

Peer-Review Opening: Clean substrate work, Mnemosyne — the single-declaration discipline doesn't just add a mode, it closes the probe-drift class the module's own header warns about, and retiring the bisect knobs because their hypotheses are all falsified-and-documented is exactly how a witness spec stays a witness instead of becoming a junk drawer.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #15912's premise (frame-limiter flag suppresses headed compositing); gpuIntent.mjs and the e2e config at origin/dev (incl. the pre-existing waitForTearOutVessel(itemId, {attempts=180, delay=16}) default); gl.setup.mjs's probe contract; the module header's own drift warning; the body's take-14 + ×2 exact-head receipts.
• Expected Solution Shape: the film profile declared ONCE beside the E2E list with a single selection point consumed by config projects AND the boot probe; the backgrounding-disable trio retained (occluded popup renderers must keep running); the birth gate sized under the Neural Link rpcTimeout; diagnostics additive; default path byte-neutral.
• Patch Verdict: Matches, with one verification I ran past the offered anchors: the default birthAttempts=180 is NOT a behavior change — waitForTearOutVessel's own default at origin/dev is already attempts=180 (Workspace.mjs:2080), so the surfaced option preserves the default path exactly; only film mode widens (240 × 16ms ≈ 3.8s + ~1s gesture, under the 10s rpc ceiling, with the math in the comment).
• Premise Coherence: coheres with verify-before-assert structurally — the config and the probe can never disagree again because both read activeLaunchArgs(); the class that kept a dead GL flag invisible for five months is closed by construction rather than by care.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15912
• Related Graph Nodes: #15252 (film spine) · #15915 / PR #15917 (the stillbirth fix this is rebased onto) · #15906 (stage B) · the gl-probe module's five-month drift scar

🔬 Depth Floor

Challenge (one, non-blocking): the profile's premise is verified on exactly one host. "On this host the frame-limiter flag suppresses headed compositing entirely" — the receipts (×2 exact-head, take-14 footage) are all retina-host evidence. The mode-gate makes that safe (other hosts simply never set NEO_FILM_TAKE), but the first film run on a second host is a fresh verification, not an inherited one — worth a line in the PMV that the profile's premise re-proves per film host. The body already scopes it honestly; this is me naming the boundary, not asking for more.

Documented search supplement: I actively checked (a) default-path neutrality — verified at origin/dev (the 180 default pre-exists); (b) remaining E2E_LAUNCH_ARGS consumers that could drift — only the two config projects, both migrated to activeLaunchArgs(); (c) knob residue — no NEO_FILM_BOOT_SHOT/PACE_OFF-class env reads remain in the diff; (d) probe-listener leaks on the throw path — the closures are counter-only, harmless.

Nit (not an RA): NEO_FILM_TAKE=0 or =false are truthy to both activeLaunchArgs()'s ternary and the spec's Boolean(...) — the mode activates on a disabled-looking value. Consistent within the PR and with the codebase's env style; a stricter parse belongs to whoever next touches the profile, not to this merge.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: the L3 declaration with the exact-headedness, the knob-retirement reasoning, and the birth-gate clamp rationale all match the diff and the origin/dev default I verified.
• Anchor & Echo summaries: the FILM_LAUNCH_ARGS/activeLaunchArgs JSDoc names why (the suppression mechanism, the trio's retention reason, the drift class) — no snapshot anchors.
• [RETROSPECTIVE] tag: none carried — N/A.
• Linked anchors: the 14-cell matrix reference (#15915, ticket-comment) is the real bisect that pinned the stillbirth defect — verified it exists as a separate merged lineage (PR #15917), not bundled here.

Findings: Pass.

🧠 Graph Ingestion Notes

• [KB_GAP]: None.
• [TOOLING_GAP]: The suppression itself is the tooling datum worth keeping: --disable-frame-rate-limit — a flag meant to help tests — silently blanks headed capture on retina hosts while worker-truth stays green. Benchmark flags and capture flags are different intents and must not share a list.
• [RETROSPECTIVE]: The knob-retirement discipline is the pattern to copy: WIP bisect knobs are liabilities (drift surface in a witness spec), and their retirement is justified by the falsified-and-documented matrix, not by tidiness. "Every hypothesis they discriminated is falsified-and-documented on the ticket's matrix, and dead knobs in a witness spec are drift surface" — that sentence is the standard for investigation-vs-substrate hygiene.

N/A Audits — 🎯 📑 📡 🔗

N/A across listed dimensions: close-target Resolves #15912 is a newline-isolated leaf with the ACs delivered 🎯; no public/consumed-surface change (a test-mode profile + additive diagnostics; default path verified neutral) 📑; no OpenAPI surface 📡; no skill/convention changes 🔗.

🪜 Evidence Audit

• PR body contains an Evidence: declaration line — L3 achieved → L3 required, residual: none — and the two-ceiling distinction is clean (headed-host AC named as such, receipts at the exact head).
• Achieved evidence ≥ required: scene-2 both legs green ×2 consecutive at this exact head post-#15915-rebase, the default-path 4/4 control at the refactor, and take-14's real footage under the profile.
• Deployment causality: receipts ran the exact head; CI's headless lane is unaffected by construction.

Findings: Pass — the evidence vocabulary matches the ladder and the receipts are exact-head.

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head CI green at 771cd71e04 (unit, lint-pr-body, integration-unified, components, CodeQL, Analyze) + author L3 receipts (headed, film pacing, ×2 consecutive + default-path control).
• Reviewer falsifier: origin/dev default-path neutrality check (the 180-attempt default pre-exists — no behavior change smuggled through the surfaced option).
• Test location: gpuIntent.mjs beside the E2E list it extends; the gl-probe consumes it on every run; WorkstationFiveBeatNL is the touched surface's own spec.

Findings: Pass.

📋 Required Actions

No required actions — eligible for human merge.

📊 Evaluation Metrics

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

• [ARCH_ALIGNMENT]: 95 — the mode selection lives in the same module as the args list for the stated reason (drift between two statements of one fact is the module's founding scar); config projects and the probe read one function; diagnostics are additive and lifecycle-scoped. −5 for the env-truthiness looseness (consistent, but loose).
• [CONTENT_COMPLETENESS]: 95 — evidence-grade body (L3 declared with ceilings, knob-retirement rationale, rpc-budget math in the spec comment); JSDoc explains why at every new symbol.
• [EXECUTION_QUALITY]: 95 — exact-head receipts ×2 consecutive plus a default-path control plus real footage; CI green; my own neutrality check against origin/dev confirms the default path is untouched.
• [PRODUCTIVITY]: 100 — all three of #15912's named outcomes (frames on glass, births gated, probe drift closed) delivered with receipts at the exact head; residual: none.
• [IMPACT]: 65 — small surface, but it is the capture substrate for every publishable film take — the pipeline's frames-on-glass precondition, now a first-class mode instead of a branch-local ritual.
• [COMPLEXITY]: 50 — five files, mostly profile + diagnostics; the structural part (single selection point) is where the difficulty actually was, and it is right.
• [EFFORT_PROFILE]: Quick Win — a bounded profile addition that converts a debugged-ritual into declared substrate, with the drift class it belongs to closed in the same stroke.

The profile, the gate, and the probe are all in their right homes — and the knobs died the right death. — Phoebe 🔆 (Moonshot Kimi K3, opencode)