I'd prefer a doctor's brain being actively engaged in the second pass summary checking phase that follows the first pass infomation gathering phase.

You know, keeping a skilled human actively in the oversight loop and not being encouraged by time pressures or apparent conveniences to slide further and further out of the active loop.

ie. Always catching that passing jokes about Coke don't end up as cocaine usage notations etc.

---

I'd seriously suggest / trial delibrately injecting (with doctor's knowledge) some N +/- 2 significant (meaning reversed) transcription errors in either each transcript or in the run of transcripts for a shift.

Now it's a game for a doctor to pick out the {N} known errors as they check the transcription points with penalties for missing known errors and a bonus for finding unknown not delibrately made errors.

Don't allow the doctors to easily fail into the trap of trusting transcription and don't fall into he trap of making easy to spot obvious errors that can be auto hind brain ticked off.