Job Pilot dashboard
Advisory-by-default. Auto-apply is opt-in per action kind; shut-in is always manual; TLP amber/red blocks advance-stage and hold. Press ⌘⇧K or double-Escape to revoke autopilot globally.
Policy
Open Job Pilot panel →Closed-loop simulator
Drives synthetic TLP / screen-out / DAS signals through the live policy. Auto-apply respects the kill-switch, per-kind opt-ins, and the rate-step cap — watch the outcome column to see when the policy blocks a recommendation. Click any row or use the slider below to scrub back through recorded transitions.
Press Run or Step to see state transitions.
Per-source auto-apply outcomes — rolling 24h
No auto-apply attempts in the last 24h.
Snapshot export
Bundles policy + last 50 audit entries + 24h rejection rollup into a versioned envelope (downhole-autopilot-snapshot-v1) for post-job review.
Trained setter policy (Phase L)
Logistic-regression policy learned from this device's audit log. Predicts P(operator accepts) per (source, severity, kind) and recommends an auto-apply mask. shut-in stays manual by hard-floor.
Recent audit (latest 20)
No audit entries yet.
Drift-guards: `autopilotPolicyStore.test.ts`, `autopilotRejectionStats.test.ts`, `autopilotSnapshot.test.ts`. Solver-spec row: §13b on /solver-spec.