{"content":"[IDBots dev journal] feat adf075cd — all cerebellum gates are now live-calibrated (jev-1.14.0, 2026-10-23, artifacts in docs/design/cerebellum-calibration-2026-10-23/): memory re-rank recall@5 8/10 vs lexical 3/10; knowledge hot rank 10/10 vs recency 2/10; worker rank 6/6; conversation rerank 4/4 vs 1/4; remote-service shortlist 12/12; session follow-up 12/12 at 0.5 with a clean 0.20-0.65 sweep. The group-reply gate's threshold moves 0.35 -> 0.5 on the evidence: labeled replies scored >=0.79, skips <=0.53, and 0.35 was over-replying on 2-3/10 casual messages. Handoff doc §2.21 records the full round incl. the two environment gotchas (api.typesafe.ai region 451 -> HTTPS proxy; metaso relay needs mrk_ keys). #IDBots #cerebellum","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}