What-If Scenario Console

An offline console where operators and engineers explore how the machine would respond to a change — driven by the twin, never touching the live plant.

Rehearsing decisions safely

Operators need to ask 'what if' — what if we push end-plug density, what if we take the breeder shape closer to its stability edge, what if a diagnostic drops mid-shot. The what-if console lets them explore these against the KRONOS-CTRL twin in a strictly offline sandbox. It shares the twin's physics and the live interface's look, but it is firewalled from actuation: nothing done here can reach the real machine.

How a scenario runs

A scenario starts from either the current live state (snapshotted) or a saved/recorded state, then applies proposed changes and runs the twin forward. The operator sees the projected trajectory with its uncertainty, the envelope margins it would consume, and any precursors it would raise — the same surfaces they use live, so intuition transfers directly. Because the twin exposes confidence, the console also shows where a scenario pushes the model out of its trained distribution and therefore where the projection is least trustworthy.

scn = twin.sandbox(from_state=live.snapshot())   # isolated copy, no actuation
scn.set('plug_density_setpoint', +5e18)            # burner what-if
traj = scn.rollout(horizon_ms=500)
report(traj.margins, traj.precursors, traj.confidence, traj.ood_regions)
# firewall invariant: sandbox has no path to L1 actuators

Where scenarios come from and go

Scenarios feed pre-operation briefings (rehearse the plan before running it), training (present trainees with situations that would be unsafe to create live), and counterfactual analysis after an incident (would a different action have avoided it?). Useful scenarios are saved to a shared library so the team accumulates a rehearsed playbook for both machines rather than improvising each upset.

The console is the interactive face of the counterfactual replay described in incident replay, and it is a core tool in training-simulator mode. Its firewall from live actuation is the same discipline that keeps observability ubiquitous but authority concentrated across all of L6.