promptdojo_

A lead follow-up model gets 62 percent accuracy. The old rule based on 'last reply older than 7 days' gets 64 percent on the same held-back rows. What should the builder say?