Disruption-Mitigation Workflow

When the twin predicts a breeder disruption, orchestration triggers a bounded mitigation sequence before the plasma terminates uncontrolled.

Predict, then act within bounds

A disruption is a sudden loss of the breeder plasma's confinement that can damage the machine. The twin (L3) forecasts disruption risk from real-time diagnostics; when risk crosses a threshold, orchestration triggers a pre-planned mitigation sequence. Because this is a protective action it runs at high priority, but it still passes the envelope and rate checks, since a mitigation that itself violates magnet limits helps nothing.

The mitigation decision

if twin.disruption_risk > RISK_TRIGGER and twin.uncertainty < UNC_MAX:
    action = mitigation_plan(state)          # e.g. controlled termination / injection
    if envelope_ok(action) and rate_ok(action):
        execute(action, token=EMERGENCY_MITIGATION)
    else:
        fallback_safe_termination()          # always-valid last resort
elif twin.uncertainty >= UNC_MAX:
    conservative_rampdown()                  # do not trust an uncertain forecast

Uncertainty gates the response

A high-confidence forecast can trigger a targeted mitigation; a low-confidence one falls back to a conservative controlled ramp-down rather than a specific aggressive intervention. This encodes the principle that the machine should not take a strong action on a prediction it cannot trust, consistent with twin/copilot coordination.

Relationship to hardware protection

• The predictive workflow aims to act BEFORE the fast hardware interlocks must fire.
• If a disruption occurs anyway, L1 hardware protection and interlocks handle the immediate event.
• A disruption can trigger or accompany a quench; the two workflows coordinate on a shared safe-state path.

Simulation and replay

Disruption mitigation is developed and tuned against replayed and synthetic disruption scenarios in the twin, since the breeder is not built. Every trigger, forecast, and action is journalled so a mitigation can be replayed to ask whether an updated twin would have acted sooner or better.