Diagnostics Constellation Timing
Thomson, interferometry, Mirnov, flux loops, ECE, fast-ion, and neutron channels are time-aligned so the control state is a single coherent snapshot.
A constellation, not a sensor
Control of either machine draws on many diagnostics at once. The breeder equilibrium needs magnetics (Mirnov coils, flux loops) plus profile diagnostics (Thomson scattering, interferometry, ECE); disruption avoidance adds fast-ion and neutron-flux cues. The burner needs density and potential diagnostics across its mirror cells. L1 treats these as one constellation whose members must be sampled coherently.
Why alignment matters
Equilibrium reconstruction solves for a self-consistent state; feeding it magnetics from one instant and profiles from another produces a physically inconsistent solution. Kronos samples every constellation member on the shared clock, so the reconstruction operates on a true snapshot. Time-alignment error directly becomes state-estimate error, so it is budgeted in nanoseconds.
Per-channel character
- Mirnov coils / flux loops: fast magnetics for equilibrium and MHD.
- Thomson / interferometry / ECE: density and temperature profiles.
- Fast-ion trackers: energetic-particle behavior, disruption cues.
- Neutron flux: fusion-rate proxy, coherent with the magnetics.
Channels differ in native rate and latency, so L1 aligns them to a common cadence: fast magnetics may sample far above the control rate and be decimated coherently, while slower profile diagnostics are timestamped and interpolated to the control instant. The shared clock makes both operations exact rather than approximate.
Handling dropouts
Any channel can drop. The ingress path flags the gap and marks the state vector's validity mask; L3's GNN can impute the missing value from the sensor topology while L1 continues on the reduced-but-known state. Because the failsafe path never relies on imputed data, a dropout degrades quality, not safety, consistent with the determinism boundary.