Adaptive resonance tracking
Compare static and adaptive coupling under identical process noise. 1,200 samples · Δt 0.01 s · target 0.297.
Measured tracking error
Root mean square error relative to the target; lower is better.
| Window | Static RMSE | Adaptive RMSE |
|---|
Model and interpretation
xᵢ = xᵢ₋₁ − kᵢ(xᵢ₋₁ − 0.297)Δt + σᵢ√Δt zᵢ. Both paths share zᵢ.
Adaptive gain = clip(0.12 + 1.2|x − 0.297|² − 0.25σ, 0.04, 0.80). The scenario supplies σ exactly; it is not estimated. Lowering gain does not reduce the additive noise input.
The defined coherence score, 1 − |x − target|, may be negative. It is not a phase-synchronisation measurement. The 0.75 threshold creates no nodes. State is scalar, without circular wrapping.
Mulberry32 / Box–Muller random draws reproduce this browser implementation, not NumPy’s random stream. Initial gain is evaluated from the law. Results run locally and are saved only through exports. This experiment does not actuate KEX engines or BRAINK optimisers. Independent assessment report number: unassigned.