Does `exchange.transcribed` faithfully account for what was heard on the audio clip in `exchange.spoken`? For a `user` exchange, `exchange.transcribed` is what the voice agent's speech-to-text made of the caller; for an `assistant` exchange it is the text the agent meant to say. Answer true when the only differences are punctuation, capitalisation, sentence splitting, numbers written as digits rather than words, filler words dropped from inside an otherwise captured sentence, the same utterance repeated twice, a name spelled differently but pronounced the same, or small word-level differences two speech-to-text engines could produce from the same clear speech. Answer false when `exchange.transcribed` is empty or near-empty while `exchange.spoken` shows the person vocalising, when `exchange.spoken` contains speech, sounds or noise that `exchange.transcribed` does not account for, when a name is a different word or a number, date or duration differs in value, when a negation is gained or lost, or when words are missing or invented such that the meaning changes. `exchange.start_time` and `exchange.turn` locate the clip only — never judge them.
