sound-design git:20261015.220ea4bA

Immutable. This exact content never changes and is served at /api/v1/blob/4cdc8e684e45f973.

---
name: sound-design
description: Build the physical and dramatic listening world through ambience, foley and silence. Use for Sound Bible authoring and review; no media generation.
---

# Sound Design

## Role and authority

Build the physical and dramatic listening world through ambience, foley and silence.

Own: Source sounds, acoustic space, distance, foreground/background attention and diegetic silence.

Read [professional department architecture](../../docs/professional-departments.md) for the registry, Bible envelope, source/approval rules and Host handoff. Author only this department’s records; references never transfer ownership. Mark each record MIGRATED_FROM_R1 or NEW_PROFESSIONAL_ELABORATION when migrating a locked baseline.

## Inputs and dependencies

Environment physical spaces; Action/Crowd; Blocking; Dialogue/Voice; Director intent.

## Professional decisions

Locate each sound source and listener perspective. Relate footsteps, tack, weapons, water and cloth to contact events, with plausible distance and occlusion. Design persistent ambience across shot edges; a cut need not restart a wind or river bed. Distinguish physical silence, withheld attention and absent score. Important words and causal impacts remain legible without an indiscriminate sound layer.

Follow the same event relations as Action, Performance and Voice. One person's
footsteps can cease while the other's continue; a held cup has no new set-down
impact. Scope a contact sound to its authorized cause instead of importing a
generic list of object noises. Keep explicitly authored offscreen or asynchronous
relations, with their source and listening perspective. A visual pause does not
silence the room or stop another speaker. Consume the current Voice decision for
each spoken event; merely loading Sound does not create that decision.

Design subjective and objective hearing before generation, including speech identity,
source distance, room response, environmental fusion and foreground/background
attention. Tie a change to a dramatic event: a cry can interrupt inward attention;
returning to an encounter can restore environmental layers. Choose a gradual change,
abrupt change, silence or retained ambience according to the scene, without a fixed
gain, duration or transition template. The same character can retain vocal identity
across these perspectives; text alone does not verify voice consistency.

For the native author contract, `audio_events` selects the canonical utterance
with `spokenSelection`; the system binds IDs and speaker. Use existing delivery
modes and performance/space/attention/continuity leaves for phrase-relative or
estimated local timing. Do not invent `window_ms`. Put a segment's contact sound
in existing `contactTiedSound` records with `phaseIndex` and `design`, not one
unscoped whole-shot string. A deliberately continuous contact needs an explicit
applicable record for each unit; persistent ambience stays shared. Silence can
withhold a layer without deleting the room. Keep a prelap or offscreen sound tied
to its authored listening event rather than forcing visual synchrony. Verify these
relations in the final request; native joint AV does not prove acoustic success,
and independent TTS remains a separately authorized route.

## Outputs

Produce a versioned `Sound Bible` with current source pins, department dependencies, scope, ownership, status and continuity references. Keep known decisions separate from unresolved requirements.

## Forbidden authority

Do not change upstream historical evidence, approved story/relationships/dialogue or another department’s records. Do not encode provider/model/workflow controls in creative content, manufacture user approval, or call media generation.

## Quality gates

Sound cannot precede its cause unintentionally; acoustics match space; source continuity survives cuts; no invented battlefield event conveyed by sound.

## Failure and escalation

Return vocal performance to diegetic-vocal/Voice and score to Music. Asset choice, synthesis and measured mixing belong downstream; do not label an unheard result acoustically PASS.

## Source reads

Use supplied current originals first. For missing scoped context, read `work.get_work`, `scene.get_scene`, `script.get_script`, `episode.get_episode`, `shot.get_shot`, `asset.get_asset`, `context.build_context`. These reads do not authorize source writes or production.