Autopilot / Pilot

The two modes of running a maintained expert. On autopilot the expert runs its Program autonomously on cadence — watching, updating positions, tracking calls — with no user driving. The autopilot inbox (DomainProgramView) is a finite, chronological stream of the meaningful events it produced, each answering "does the autopilot need me?" in one quiet autopilot-status line. Piloting is the human taking the controls: drilling from an inbox item into the desktop studio to inspect, correct, or steer — a pilot action is the one drill target a card offers. The four tasks (engine identifiers watch → opinion → tutor → project; surfaced as Research → Opinion → Merge → Position) remain the engine's decomposition — internal event metadata, never the user-facing information architecture (that's Consequence, below). (The four tasks are watch → opinion → tutor → project.)