Composition audio beds are dropped in head-only composition modes

[Brad-Edwards]

Review finding

Composition navigation modes that truncate a composition to its head scene rebuild the composition context without carrying audioBed forward. mode=standalone intentionally suppresses beds, but mode=loop and mode=scrub are audible modes and should preserve the surrounding composition bed.

Evidence

• src/runtime/scene-loader.ts:1561 applySingleSceneSlice() creates a new composition object with id, manifestSlice, sceneSlice, and startIndex, but omits audioBed.
• src/runtime/scene-navigation.ts:492 truncateToHead() has the same omission for direct bridge callers.
• src/runtime/scene-loader.ts:1315 passes a bed to createAudioService() only from resolved.composition?.audioBed, so once the slice transform drops it the bed cannot play.

Impact

A bed-carrying composition addressed with mode=loop or mode=scrub silently loses its bed. This makes composition playback differ from the declared composition and from present-mode behavior. Direct callers of loadSceneNavigationTarget() also get a truncated head slice without the bed.

Recommended fix

Preserve audioBed when head-only composition slices are rebuilt, and keep the existing bedSuppressed: mode === 'standalone' behavior as the explicit standalone suppression point.

Acceptance checks

• applySingleSceneSlice() preserves resolved.composition.audioBed when rebuilding the head slice.
• truncateToHead() preserves target.composition.audioBed for direct callers.
• Tests cover a composition with an audioBed in mode=loop and mode=scrub and prove the bed is constructed/started.
• Existing standalone tests continue to prove mode=standalone suppresses the bed.

[Brad-Edwards]

Picked up by /implement - driver Codex, branch 147-preserve-audio-beds, 2026-05-23T06:51:43Z.

[Brad-Edwards]

gc_codex_review — cycle 1 of 1 (pre-push) on issue #147 (branch 147-preserve-audio-beds)

Core review

Verdict: ship

This change is shaped correctly: the design seam is the composition-to-head-slice projection, and the fix keeps composition-level audio-bed metadata attached when that projection truncates manifest and scene slices. It updates both truncation sites, so scene loading and scene navigation keep the same boundary contract instead of solving audio playback lower in the stack. The test pins the user-visible behavior across the two head-slice modes that were likely losing the bed. No obvious next variation is foreclosed.

No blocking findings.

Security review

Verdict: ship

This change is shaped correctly: it keeps audio-bed propagation at the existing head-slice truncation seam in both scene loading and scene navigation, without adding a new trust boundary, repository access path, request parser, or dynamic execution surface. The cross-cutting concern is runtime media behavior across loop and scrub modes; carrying an already-resolved composition field through the truncated composition object does not weaken validation or scoping, and it leaves the obvious next variation, more composition-level metadata surviving head slicing, at the same seam.

No blocking findings.

[Brad-Edwards]

gc_codex_review pre-push cycle 1 of 1 complete for issue #147 on branch '147-preserve-audio-beds'. Posted by the MCP server to enforce the pre-push hard-cap-1 contract (issues #796, #804, #906). Do not edit or delete — used by the next gc_codex_review (uncommitted) invocation to count cycles.

[Brad-Edwards]

Review decision record — codex cycle 1 (issue #147)

Reviewer: codex
Cycle: 1

Architectural read:

Core reviewer: This change is shaped correctly: the design seam is the composition-to-head-slice projection, and the fix keeps composition-level audio-bed metadata attached when that projection truncates manifest and scene slices. It updates both truncation sites, so scene loading and scene navigation keep the same boundary contract instead of solving audio playback lower in the stack. The test pins the user-visible behavior across the two head-slice modes that were likely losing the bed. No obvious next variation is foreclosed.

Security reviewer: This change is shaped correctly: it keeps audio-bed propagation at the existing head-slice truncation seam in both scene loading and scene navigation, without adding a new trust boundary, repository access path, request parser, or dynamic execution surface. The cross-cutting concern is runtime media behavior across loop and scrub modes; carrying an already-resolved composition field through the truncated composition object does not weaken validation or scoping, and it leaves the obvious next variation, more composition-level metadata surviving head slicing, at the same seam.

Blocking findings: 0 (clean run)

[Brad-Edwards]

gc_test_quality_review cycle 1 of 1 — issue #147

Reviewer: test-quality (claude-sonnet-4-6 via gc_test_quality_review)
Branch: 147-preserve-audio-beds
Cycle: 1 / 1
Findings: 1

Finding 1 — [critical] tests/runtime/scene-loader-audio.test.ts::scene loader — audio service wiring (PUL-F024 / ADR-004)::does not build an audio service under mode=prompter (no scene mounts):379

Problem: The head scene is declared with assets: ['/audio/head.mp3'] but without audio: ['/audio/head.mp3'], so scene.audio defaults to []. If a regression caused mode=prompter to take the non-prompter branch and call buildLoad, createAudioService would be called with allowedSources: [], the scene's create() would call audio.load('head-bed', { src: '/audio/head.mp3' }), the allowlist check would reject it, no sound would be created, and expect(audio.created).toHaveLength(0) would still pass. The test cannot distinguish between 'lifecycle was suppressed by the prompter path' and 'lifecycle ran but every audio load was blocked by the empty allowlist'.
Why it matters: A regression removing the effectiveMode(target) === 'prompter' branch guard at line 1711 of scene-loader.ts — causing the scene resolver lifecycle to mount under mode=prompter — would go entirely undetected. The scene's create() hook would execute (and the test comment 'Would register a sound — but prompter never mounts the scene' is incorrect: even if it ran, no sound would be created because scene.audio is empty).
Fix: Add audio: ['/audio/head.mp3'] to the buildScene call (alongside the existing assets entry). Then collectAudioSources would include /audio/head.mp3 in allowedSources; if the lifecycle ran, audio.load('head-bed', { src: '/audio/head.mp3' }) would succeed, audio.created would have length 1, and the assertion would correctly fail — detecting the regression.

[Brad-Edwards]

gc_test_quality_review cycle 1 of 1 complete for issue #147 on branch '147-preserve-audio-beds'. Posted by the MCP server to enforce the gc_test_quality_review hard-cap-1 contract (issue #884 follow-up, default lowered in #906). Do not edit or delete — used by the next gc_test_quality_review invocation to count cycles.

[Brad-Edwards]

Review decision record — test-quality cycle 1 (issue #147)

Reviewer: test-quality
Cycle: 1

Architectural read:

This file pins the scene loader's audio-wiring seam — the four behavioral contracts PUL-F024/ADR-004 record — through a recording audio engine that is a real in-process implementation, not a mock; it captures every createSound call and every handle method call so assertions can be made on both engine-level state (muted flag, created count) and the AudioService's own state machine (isDisposed, isMuted). The three describe blocks (core wiring, rehearsal, unlock-gate bypass matrix, composition bed) map cleanly onto the PUL-Fxxx clause groups being pinned. The recording unlock adapter and recording timeline adapter follow the same pattern and are correctly factored. The issue #99 regression nail (single stop-group for a repeated scene id) uses the cue log as the observable signal — the right observable for that specific invariant. The shape is correct and the overall coverage is strong. The one seam gap is in the prompter-bypass test, where the scene's empty scene.audio allowlist accidentally masks the regression signal: the assertion audio.created.toHaveLength(0) passes whether the lifecycle was suppressed (expected) or accidentally ran but the load was rejected by the allowlist (regression).

Blocking findings: 1

Finding 1 — one-off

• ID: F1
• Title: (no title)
• Decision: fix
• Rationale: Addressed by next cycle

[Brad-Edwards]

Proceeding without an over-cap test-quality review cycle per user instruction after fixing the cycle-1 finding and passing local verification (

@keplerops/pulsar@0.1.0 lint /home/atomik/src/pulsar
biome check .

Checked 191 files in 196ms. No fixes applied.

@keplerops/pulsar@0.1.0 typecheck /home/atomik/src/pulsar
tsc --noEmit

@keplerops/pulsar@0.1.0 test /home/atomik/src/pulsar
vitest run

RUN v3.2.4 /home/atomik/src/pulsar

✓ tests/runtime/rng.test.ts (8 tests) 559ms
✓ createSeededRng > yields every draw in the half-open interval [0, 1) 551ms
✓ tests/runtime/timeline.test.ts (98 tests) 382ms
✓ tests/runtime/policy-q004-resource-cleanup.test.ts (145 tests) 304ms
✓ tests/system/helpers.test.ts (20 tests) 492ms
✓ tests/runtime/policy-a008-mode-dispatch.test.ts (80 tests) 396ms
✓ tests/runtime/scrub-cue-gating.test.ts (1 test) 1067ms
✓ PUL-F017 — monotonic-forward audio cue gating > fires a cue on a forward crossing of its master time and suppresses it on a reverse crossing 1066ms
✓ tests/system/chrome-slots.test.ts (10 tests) 330ms
✓ tests/runtime/policy-q008-dom-css-accessibility.test.ts (85 tests) 1040ms
✓ PUL-Q008 — DOM/CSS accessibility (source scan) > runtime tree (current code revision) > contains no Q008 violations across src/runtime/**/*.ts 779ms
✓ tests/runtime/policy-a004-export-pipeline.test.ts (12 tests) 1396ms
✓ PUL-A004 — live runtime independent of export pipeline (source scan) > runtime tree (current code revision) > contains no A004 violations across the runtime-core file set 1362ms
✓ tests/runtime/policy-a003-rendering-libraries.test.ts (15 tests) 1250ms
✓ PUL-A003 — optional rendering libraries are scene-local (source scan) > runtime tree (current code revision) > contains no A003 violations across the runtime-core file set 1178ms
✓ tests/runtime/policy-a006-slide-frameworks.test.ts (14 tests) 1444ms
✓ PUL-A006 — live runtime independent of slide frameworks (source scan) > runtime tree (current code revision) > contains no A006 violations across the runtime-core file set 1390ms
✓ tests/system/chrome-extras.test.ts (5 tests) 310ms
✓ tests/runtime/policy-a010-export-metadata-share.test.ts (117 tests) 1678ms
✓ PUL-A010 — live and export share scene metadata (source scan) > runtime tree (current code revision) > rule 2: zero forbidden parallel scene/composition declarations across src/**/*.ts 1232ms
✓ tests/runtime/navigation.test.ts (111 tests) 163ms
✓ tests/runtime/policy-a009-captions-single-source.test.ts (145 tests) 1714ms
✓ PUL-A009 — captions / prompter single source (source scan) > runtime tree (current code revision) > rule 2: zero forbidden parallel caption-schema declarations across src/**/*.ts 1367ms
✓ tests/runtime/screenshot-determinism-source.test.ts (80 tests) 1976ms
✓ PUL-Q001 — screenshot determinism source scan > runtime tree (current code revision) > contains no non-deterministic primitives across src/**/*.ts 1861ms
✓ tests/runtime/audio.test.ts (114 tests) 107ms
✓ tests/runtime/policy-a002-audio-encapsulation.test.ts (47 tests) 224ms
✓ tests/runtime/policy-q007-remote-code-execution.test.ts (68 tests) 2336ms
✓ PUL-Q007 — no remote code execution (source scan) > runtime tree (current code revision) > contains no Q007 violations across src/**/*.ts 2197ms
✓ tests/runtime/policy-a001-timeline-encapsulation.test.ts (27 tests) 140ms
✓ tests/system/presenter-bridge.test.ts (10 tests) 126ms
✓ tests/runtime/composition.test.ts (116 tests) 33ms
✓ tests/runtime/policy-q003-url-state-determinism.test.ts (130 tests) 2322ms
✓ PUL-Q003 — URL state determinism (source scan) > runtime tree (current code revision) > contains no Q003 violations across src/**/*.ts 2093ms
✓ tests/runtime/audio-engine.test.ts (15 tests) 98ms
✓ tests/runtime/scene-loader-present.test.ts (47 tests) 54ms
✓ tests/runtime/policy-a005-declarative-composition.test.ts (38 tests) 87ms
✓ tests/runtime/scene-loader.test.ts (29 tests) 49ms
✓ tests/runtime/scene-loader-audio.test.ts (45 tests) 71ms
✓ tests/runtime/composition-resolver.test.ts (42 tests) 76ms
✓ tests/runtime/scene-loader-screenshot-prompter.test.ts (33 tests) 88ms
✓ tests/runtime/policy-scene-trust-model-doc.test.ts (2 tests) 164ms
✓ tests/runtime/policy-q002-browser-support.test.ts (21 tests) 83ms
✓ tests/runtime/asset-preloader.test.ts (34 tests) 148ms
✓ tests/runtime/scene.test.ts (125 tests) 38ms
✓ tests/runtime/scene-loader-beat-mode.test.ts (28 tests) 39ms
✓ tests/runtime/scene-loader-paused-scrub.test.ts (34 tests) 68ms
✓ tests/runtime/id-registry.test.ts (5 tests) 10ms
✓ tests/system/audio-helpers.test.ts (13 tests) 23ms
✓ tests/system/scrub-controls.test.ts (13 tests) 38ms
✓ tests/runtime/validation.test.ts (52 tests) 38ms
✓ tests/runtime/scene-navigation.test.ts (63 tests) 41ms
✓ tests/runtime/pul-q006-error-context.test.ts (16 tests) 44ms
✓ tests/runtime/scene-loader-standalone-loop.test.ts (26 tests) 59ms
✓ tests/system/standalone-audio-bed.test.ts (3 tests) 21ms
✓ tests/system/presenter-driven.test.ts (11 tests) 39ms
✓ tests/runtime/error.test.ts (16 tests) 10ms
✓ tests/runtime/composition-registry.test.ts (24 tests) 39ms
✓ tests/runtime/scene-loader-chrome.test.ts (19 tests) 37ms
✓ tests/runtime/policy-biome-complexity-gate.test.ts (5 tests) 46ms
✓ tests/runtime/registry.test.ts (35 tests) 23ms
✓ tests/runtime/presenter.test.ts (26 tests) 34ms
✓ tests/scenes/browser-support-fixture.test.ts (19 tests) 47ms
✓ tests/scenes/loop-fixture.test.ts (20 tests) 23ms
✓ tests/scenes/screenshot-rng-fixture.test.ts (8 tests) 16ms
✓ tests/scenes/paused-fixture.test.ts (19 tests) 26ms
✓ tests/runtime/scene-loader-screenshot-seed.test.ts (9 tests) 21ms
✓ tests/system/presenter-keyboard.test.ts (11 tests) 55ms
✓ tests/scenes/dom-css-accessibility-fixture.test.ts (22 tests) 55ms
✓ tests/runtime/version.test.ts (2 tests) 7ms
✓ tests/runtime/prompter.test.ts (12 tests) 26ms
✓ tests/scenes/placeholder.test.ts (15 tests) 28ms
✓ tests/runtime/workbench-chrome.test.ts (22 tests) 29ms
✓ tests/runtime/audio-unlock-dom.test.ts (7 tests) 12ms
✓ tests/system/register-tokens.test.ts (2 tests) 5ms
✓ tests/scenes/scrub-fixture.test.ts (18 tests) 38ms
✓ tests/system/counter.test.ts (12 tests) 11ms
✓ tests/system/transitions.test.ts (13 tests) 20ms
✓ tests/system/presenter-renderers.test.ts (6 tests) 10ms
✓ tests/system/prompter-window-url.test.ts (4 tests) 13ms
✓ tests/system/terminal-audio.test.ts (8 tests) 11ms
✓ tests/system/templates-options.test.ts (20 tests) 37ms
✓ tests/system/templates.test.ts (58 tests) 27ms
✓ tests/system/pulsar-intro-deck.test.ts (7 tests) 9ms
✓ tests/runtime/workbench-graph.test.ts (6 tests) 9ms

Test Files 74 passed (74)
Tests 2598 passed (2598)
Start at 09:05:50
Duration 7.67s (transform 9.40s, setup 1.27s, collect 44.26s, tests 21.88s, environment 19ms, prepare 11.10s), plus pre-commit).