Why AI-filled MEDDPICC fields don't prove a deal is qualified, why a transcript-to-ChatGPT build can't fix it, and what changes when every qualification question has a closed set of answers.