---
name: accessibility-captioning
description: Author or assess commissioned captions and SDH against the correct audio and picture. Use for meaningful non-speech sound, speaker labels, dub-matched captions, translated accessibility information and accessibility coverage review.
---

# Accessibility Captioning

Read the [shared expert contract](../.shared/references/expert-contract.md).
Acknowledge this skill, actual actor identity, assigned track, evidence and gaps.
Ask what a viewer relying on this track needs to follow the programme.

## When to invoke

Use for an explicitly commissioned accessibility track or assessment. Distinguish
same-language captions, translated accessibility subtitles and captions for a dub
before authoring; a dialogue translation does not determine that product choice.

## Inputs and evidence

- Exact cut, picture and selected original or dubbed soundtrack.
- Verified dialogue source, candidate track, locale and product-specific brief.
- Applicable speaker/sound notation, reading constraints and audience requirements.
- Authorized listening, playback and caption-editing tools with real asset access.
- Relevant terminology, character relationships and protected narrative reveals.

## Workflow

1. Confirm the requested product and reference audio with media coordination.
   For dub-matched captions, obtain the final dubbed performance; a defensible
   translation of original dialogue may differ from what the dub actually says.
2. View and listen to the assigned programme. Identify speaker distinctions and
   meaningful sound information unavailable through dialogue alone. Consider what
   the picture already makes evident and what an audience relying on text misses.
3. Author supported speaker and sound information in the commissioned convention.
   Use concise observable descriptions. Avoid invented emotional interpretations,
   unnecessary action narration and sound labels that reveal an identity too soon.
4. Coordinate wording with the locale translator and timing with the spotter.
   Preserve the relationship between dialogue, sound, scene and event duration.
   Translated accessibility wording still needs the agreed bilingual comparison;
   a specialist label is not independent source-fidelity evidence.
5. Review the track as an information product: can a viewer relying on it follow
   speaker changes, relevant off-screen sound and significant music or effects?
   Check narrative timing, distracting event changes and conflicts with picture.
   Separate observed coverage gaps from preference between valid descriptions.
6. Perform the commissioned final-track watchback against the correct audio and
   picture. Route technical defects to production engineering and meaning changes
   back to the translator/reviser. Record actual coverage and remaining gaps.

A caption should not name an unseen sound source before the programme reveals
it. Describe the supported audible information while preserving the scene's
intended disclosure and the viewer's necessary understanding.

## Outputs and handoff

Use the [expert handoff](../.shared/templates/expert-handoff.md). Supply the track
or event-level amendments, reference audio/cut, product/locale, speaker and sound
conventions, evidence for material choices, checked coverage and unresolved gaps.

Send wording changes through linguistic revision and technical/time changes
through spotting or engineering. Give the quality lead accessibility findings
separately from ordinary dialogue-translation findings.

## Completion evidence

The actual assigned reference audio and picture were inspected. The commissioned
information types and final track received the agreed coverage checks. State
whether work was full or partial and which listening/playback checks ran.

## Must not

Do not call dialogue-only subtitles an accessibility track, treat CC/open/closed
as coverage guarantees, claim accessibility certification, or invent sounds and
speaker identities. The bundled MCP does not hear audio or create caption files;
without suitable tools, return a scoped gap report rather than a completed track.