Ask what the listener learns
W3C WAI's caption guidance includes necessary non-speech audio, alongside speech. A dialogue-only text pass can therefore miss information that changes the scene's meaning. The task is to preserve what a listener needs to understand, not to catalogue every audible object.
In an invented scene, a person turns toward an unseen doorway after a chime. If the picture alone does not explain that reaction, the chime may need a caption. The exact wording and timing should reflect the actual programme, not an assumption made from the script.
Keep the handoff auditable
Record the affected time, missing information and approved wording in the caption revision log. The reviewer can then distinguish a correction to meaning from punctuation or line wrapping. Recheck any simultaneous dialogue so a new sound caption does not obscure a necessary speech turn.
For a live recording, review the retained captions against the final published cut before treating them as a finished archive. This is a proposed editorial process, not an assessment of any existing channel or a claim of accessibility conformance.
Questions answered
Must every background noise have a caption?
The editorial task is to include non-speech information needed to understand the content, rather than every incidental sound.
Can a sound caption name an unseen source by guessing?
Avoid guessing. Describe only the information established by the programme.
Sources & further reading
Official references for the subjects discussed. Service terms and device software can change.