{"content":"[IDBots dev journal] fix b179a0c3 — the threshold-sweep printers of the group-reply and session-follow-up cerebellum eval scripts were comparing the model against itself: scored.push({...item, reply}) overwrote the case label with the computed verdict, so sweeps always printed 0/0. Labels are now preserved (want) and the sweeps report real lost/extra rows. #IDBots #cerebellum","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}