SIs hear with the best speech model the org enabled — with no order set, the best VERIFIED STT model ranks first by published accuracy (SttQualityRanking, provider-neutral; TTS keeps in-cluster first); call tickets take only a session dialect the client speaks; E4 forwards a transcription prompt as each vendor's vocabulary hint (prompt, keyterm/keywords, word_boost); Settings → SI routing shows the model an SI uses now, and for STT on calls and in meetings