End-to-End Timing Model

The nested timescales the stack must respect, from sub-millisecond plasma control to multi-year campaigns, and how each loop is budgeted.

Timescales are nested

Fusion plants are governed by physics that spans more than twelve orders of magnitude in time. Resiliency requires that each control loop closes fast enough for the phenomenon it governs, and that slower loops never block faster ones. The stack is organized as a hierarchy of loops, each with a deadline and a fallback if the deadline is missed.

The budget principle

A loop's latency budget is the sum of sensing, transport, inference, decision, and actuation. The rule is that the worst-case path (WCET) plus jitter must be less than the phenomenon's characteristic time with margin. For the breeder, vertical stability of an elongated, negative-triangularity plasma is the tightest hard-real-time constraint; for the burner, plug-field and RF stability dominate.

def loop_ok(sense_ms, xport_ms, infer_ms, act_ms, jitter_ms, deadline_ms, margin=2.0):
    wcet = sense_ms + xport_ms + infer_ms + act_ms + jitter_ms
    return wcet, wcet * margin <= deadline_ms

# vertical control example (must beat ~1 ms with 2x margin)
print(loop_ok(0.05, 0.05, 0.20, 0.10, 0.05, deadline_ms=1.0))
# (0.45, True)  -> 0.45 ms wcet, 0.90 ms with margin < 1 ms

Fast loops run on deterministic hardware (FPGA / real-time control nodes); slow loops run on the twin and fleet layers. A missed slow-loop deadline degrades optimization quality; a missed fast-loop deadline triggers a protective action. See Control Loop Latency Budget and Quench Detection.