In the case of captioned wordplay, the difference between writing and speaking, text and sound, is obvious. What works in speech doesn’t work — or works differently — on the caption layer.
Dialogue that wasn’t intended to be read
Speakers don’t need to spell things out for caption viewers when these viewers can read it for themselves at the bottom of the screen. Speakers only need to spell it out for those audio-only viewers who don’t have the added benefit of reading.