Underlord tells me that it can't correct the incorrect capitalization in the middle of sentences that transcription produces, nor can it correct extraneous punctuation also added by the transcription process (why does transcription add punctuation between repeated words and stammering? even removing repeated filler words leaves the commas all over the script still because it doesn't remove them after the ums and uhs). It claims the only way to correct this is to change the audio too, by regenerating it.

That's ridiculous. People don't speak in uppercase or lowercase letters, so nothing related to the video should need to be changed in order to correct the script because a change of case would be said exactly the same by the speaker. Please look into allowing Underlord to make script-only changes in order to fix errors transcribing leaves behind.

Or, you know, work harder on getting transcriptions to stop making errors in the first place.

I use Descript primarily for editing transcripts for podcasts & video that are final - meaning only the transcript needs edited, but not the multimedia. Editing in the Descript app and browser is cumbersome, not intuitive, and inaccurate. I have difficulty labeling speakers correctly in the program, editing incorrect transcriptions or spellings, and creating line breaks, for example. Often I use Descript to generate a transcript then simply export the document to a Word file and edit there, as it's easier to do.