Audio coverage review — 2026-09-15

Current sourcing inventory: Audio Cue Sheet and the sound library now distinguish live cues, available unused takes, missing proposals and historical references. Use that sheet for sourcing; this document preserves the implementation audit below.

The largest gaps were missing event hooks, not missing music. This review traced Sim events through Game, UI actions and the actual Sound methods at baseline 14710cb, before the September 15 cue pass. File/line references in the BEFORE sections refer to that revision; function names remain the navigation anchors as the implementation changes.

This is a code/asset coverage audit. Asset presence proves neither a good performance nor a good mix. Physical-phone listening, voice direction and subjective approval remain separate from automated coverage checks.

Implementation status

The September 15 engine and cue wiring are implemented and verified by the full Node suite and real Chrome audio matrix. The verification record below separates those checks from physical-phone listening and voice production.

No soundtrack master, catalog loop, boss-score timing, enemy model, map, combat balance or HUD control needs changing for this work. The existing sample-clock and pause contracts in MUSIC-ENGINE.md remain authoritative.

September 15 coverage added

• Four shikigami arrival/bite/exit identities, actual Ren palm casts, distinct special casts and Paper Doll's explosion. Clean Sweep does not double the primary cast cue.
• Nine short boss entrance identities. Attack warnings and releases use families: charges, ranged casts, sweeps, zones, summons and impacts. Spiral bosses now emit enemyVolley; Pouncer wind-up and Conductor gathering emit eliteTell. These are three presentation-event additions, with no new RNG draws or AI/balance changes.
• The closest actual tell per simulation batch supplies the warning. Stable order breaks ties; the game does not infer an attack by scanning enemy state or queue sounds that become stale. View.audioPan supplies camera-relative stereo position. The mixer limits repetition and simultaneous voices.
• Mana, non-weapon bonus drops, weapon formation/binding, card acceptance, knockout and bounded rally recovery accents. During the normal live-combat exorcism, weapon formation waits for the seal callback; no separate pickup sting is stacked on its initial death rupture. The no-exorcism fallback still sounds the drop. Existing exorcism/music resolution remains intact.
• Knockout suppresses the incoming student's misleading hurt/tag VO in that event batch. No Sora boss-kill line is guessed from the active student; no new recording or attribution system is included.
• Critical hits, large kills and pickups now have repeat gates; procedural sources disconnect on ended. Field Report, Club Card and Docs & FAQ no longer call the shared menu click twice.

The remaining voice work and deliberate quiet areas below are still a backlog. The BEFORE matrices are retained as the audit record, not a claim that those hooks remain missing after this pass.

BEFORE: prioritized findings

The lethal-hit ownership edge matters: _hurt queues an event and _die can switch students before dispatch. The old hurt handler reads the now-active student, so the incoming survivor can speak the outgoing student's hurt line. Do not infer the speaker of a knockout from sim.character after that swap.

BEFORE: all nine bosses

Every boss already had the generic entrance sting, boss-song handoff, the authored pressure layer below 40% HP and the three-stage exorcism. Only the last living boss starts the encounter's defeat cadence. The table describes attack coverage in addition to that shared presentation.

Source: boss identities in config.js:142; attacks in Sim._bossAI, sim.js:1159; existing dispatch at game.js:2449. Shared kit/color data cannot by itself identify the actual boss. New identity-specific cues need the event's real boss name or stable identity, without an extra RNG draw.

BEFORE: the rest of the game

Game._onWin/_announceWin defer the clear cue until the exorcism finishes. _onExorcismCue supplies the three SFX beats; bossDefeated supplies the short music ending. Neither is a reason to add a second full reward track. Sound.setIntensity affects only procedural fallback music, not the authored songs. Manual pause keeps the song playing, cancels boss pressure cues and ducks ambience; level selection uses the existing turntable behavior. None of these is a request to alter the loop engine.

Voice and sound-art backlog

The files below existed at baseline; optional failures are intentional no-ops. Counts include alternate takes, not different event types.

Presence was checked against assets/vo/, _loadTracks and VO_ALTS; it does not imply every take has been auditioned in this review. No boss spoken entrance/death lines are currently supplied. Prioritize future production:

1. Optional signature boss vocal punctuation: cast one entrance and one death for a single boss before recording all nine. Keep the threat cue intelligible, avoid lengthy taunts, and allow an unvoiced identity cue.
2. Crew essentials: Yui dash plus Ren/Yui hurt reads; then concise tag lines for all four. Author variants only after the first set works in the full mix. Existing Sora boss-kill reads need reliable killer/owner data, especially for delayed attacks and tags, before activation.
3. Shikigami sound art: review four synthesized identities first. Only commission short creature calls or texture layers where synthesis leaves a specific gap; avoid four long samples, voices on every bite or loops.
4. Reward polish: one brief card seal/flip, then a rare-card accent tied to the actual reveal. Rewards must already be committed; skipping animation never changes ownership. No Rive work before the parked review date.
5. Contextual crew dialogue: keep the separately proposed checkpoint exchanges in IDEAS.md; this pass does not add conversations.

The nested-click cleanup identified at baseline (ui.js:730,736,1134) was completed in this pass; it is not remaining sound-production work.

Deliberate quiet and technical boundaries

• No cue per ambient enemy spawn, trail sprite, sparkle, every damage number, every field tick, every card proc or every mana-regeneration increment.
• No always-on creature loop, readiness alarm loop, extra music bed, reverb convolution or another full-song buffer for this pass.
• Warning selection is bounded and favors a relevant nearby threat. Stereo placement supplements the existing visual telegraph; mute and mono remain playable. Panning uses camera-relative position, not raw world X.
• The new combat cue layer targets a maximum of 16 active/retiring voices; this is its own cap, not a claim that the entire legacy mixer has 16 voices. Use already available synthesis/noise; no fetch/decode during a cast.
• New cue tails must cancel on pause, menus, mute, title and run replacement. Cleanup must cover sources scheduled for the future as well as audible ones. Resume never replays expired warnings. All timing follows the appropriate audio/presentation clock, without gameplay delays or reloads.
• Keep the separate boss-score cap/timing contract unchanged. Existing authored music and published sound-library captures are not regenerated.

For implementation details, the primary Web Audio specification defines StereoPannerNode and scheduled source start/stop/ended behavior. These API contracts support the implementation; they do not establish this game's mobile performance or subjective mix quality.

Verification record

• Baseline audit: all literal Sim event names compared with Game dispatcher cases; UI activation paths, assets and Sound methods inspected. No code changed during that audit.
• September 15 implementation: npm test passed 1,173/1,173, including the 76 scoped checks, 14 new dispatcher tests and six new audio Node tests. The real Chrome matrix passed 144 checks at 48/44.1 kHz across 41 recipe variants, producing 51 review WAVs. Peak usage was 14 native sources / two warning groups, with zero added fetches, retained sources or browser errors. These receipts verify the new cue layer, independently of the historical boss-score matrix. The browser harness supports an external report-directory override; this run used the Codex visualization workspace's audio-cues-20260915 directory.
• Chris handles physical gameplay listening. Browser output and deterministic tests do not prove phone-speaker clarity, comfortable listening levels or approval of a voice performance. No APK publication is part of this pass.