Once a rendezvous plan is committed it now runs to completion untouched: an ordinary new setpoint arriving while the group is 'working' is remembered (latest wins) and dispatched sequentially when the group reaches 'ready', instead of aborting + re-planning. A re-plan mid-flight dropped the in-flight schedule and re-deferred a pump that was mid-sequence, parking starting pumps at minimum flow. Only an EMERGENCY pre-empts the lock: a stop (≤0) or a pressure excursion. _isUrgentDemand (which pre-empted on any large step) is replaced by _isEmergencyDemand; the large-step pre-emption is gone — large operator steps now defer like any other setpoint. _pressureEmergency() reads planner.emergencyPressurePa and is INERT until that threshold is configured; handlePressureChange fires a latched bypass dispatch when it breaches. Verified live on the E2E Isolated MGC rig: a 1→2 pump staging transition ramps the added pump straight through (no wait-at-minimum, no start-then-stop) and the group total climbs monotonically. (The Pump-tab node's hunting is a separate demand-feedback-loop issue in that flow's wiring, not the rendezvous.) Integration tests now settle to 'ready' between demands (waitReady) since the lock defers setpoints arriving mid-move. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3.9 KiB
3.9 KiB