I think the best design would be to have the ffmpeg producer parse 608/708 to events (that could also be expanded to external subtitle files supported by ffmpeg). These events could be exposed so that the HTML producer could be used for open captions/subtitles (burn-in). If the events are in OSC they could also be used by anything external that can listen to OSC. A command would choose which layer(s) is emitting events. Another command could send custom events to allow for live captioning. Following that we could work on final composition to embed as closed captions in SDI or other output modules (if an external service isn’t developed for the insertion).
At least I think it’s important not to confine this to parsing/inserting 608/708 which is of no use to most of us.