A moving target is tracked from 9 noisy sensors. Spoof or degrade some of them and watch a small Boltzmann-machine fuser hold the truth while naive averaging breaks. Everything runs live in your browser — no quantum computer, the pattern deployed on edge robots (Infleqtion/SAPIENT on NVIDIA Jetson).
The model is a conditional Boltzmann machine: a continuous state x and one binary trust unit per sensor. Its energy charges the data mismatch of every trusted sensor; mean-field inference gives each sensor a soft trust weight w = σ(β(λ − r)) and fuses x as the trust-weighted estimate. The energy landscape rejects sensors that disagree — robustness by construction, in ~1.5 µs on a laptop-class core. Honest limit: once the corrupted sensors are the majority, no fuser recovers.