Predictive Shadow Visualization

Drawing the twin's 50-100 ms lookahead as a labeled forecast beside the live state so operators see where the machine is going, with its uncertainty.

Showing the near future honestly

The KRONOS-CTRL twin runs a predictive shadow 50-100 ms ahead of the real plant. L6 renders that lookahead as a distinct, clearly-labeled forecast alongside the live state — never blended into it. The operator sees two things at once: where the machine is, and where the twin expects it to be shortly, with the uncertainty of that expectation drawn explicitly.

Forecast, not fact

The forecast is visually differentiated from measured state — a lighter weight, a leading marker, an explicit horizon label — so it can never be mistaken for data. As the horizon extends, a fan chart widens to show growing uncertainty, honest about the fact that a 100 ms projection is less certain than a 20 ms one. When the forecast's epistemic uncertainty grows large, the display says the twin is extrapolating beyond confident ground.

def draw_shadow(live, shadow):
    plot(live.t, live.y, style='measured')          # solid, current
    plot(shadow.t, shadow.mu, style='forecast')      # distinct, leading
    fan(shadow.t, shadow.mu, shadow.sigma_of_h)      # widening band vs horizon
    for c in shadow.envelope_limits:
        mark_crossing(shadow, c)                      # highlight projected breaches

Projecting toward the envelope

The shadow's value is in projecting toward constraints. For the breeder it forecasts the equilibrium and shape trajectory and flags a projected approach to a stability limit before it happens — the earliest actionable disruption cue. For the burner it projects end-plug density and the ambipolar potential forward, flagging a plug excursion in formation. In both cases the operator gets lead time, which is the whole point of situational-awareness Level 3 (projection).

The shadow underpins projection-level awareness and disruption avoidance — see operator situational awareness and the disruption countdown. Its forecasts carry the same uncertainty encodings as everything else — see uncertainty visualization — because a forecast without its error band would be the most misleading display in the room.