LOADING
WHAT IT DOES
Not guessed from the script and not evenly divided across the shot. The recording knows when each word happened, and that is what the captions are built from.

Every shot stores the start and end of its narration in milliseconds. Captions are cut to those, so they cannot drift out of step however the edit changes.
The same film can be narrated in any of twelve languages, and the captions follow the recording that was actually made.