The Evidence Layer for Scientific AI

 Quick Answer Copied

 Long Answer

What the evidence supports

The proposed CZ or shared injection qubit is a valid mechanistic test: it could expose products or interactions among θ, α, and 1/f states before regression rather than asking ridge regression to recover them afterward. However, no effect size is currently identifiable: the supplied paper contains no entangling ablation, no shared-qubit condition, and no decomposition of gains from quadratic features versus Pareto hyperparameter selection. Architecture and missing ablations

How large would “useful” be?

For the reported simulated EEG task, quantum NMSE was 0.4208 versus 0.0330 classically, while 1-PLV was 0.1880 versus 0.0110; the multivariate difference was substantial, MANOVA F(3,46)=474.25, p≤0.001, Pillai’s V=0.969. Thus, a modest relative improvement—say 5–15%—would be scientifically interesting but would not demonstrate parity; it would reduce NMSE only to approximately 0.400–0.358, and this numerical illustration is a calculation from the reported baseline, not an observed result. Reported EEG effect sizes

Decisive experiment

Use identical seeds, windows, qubit counts, polynomial degree, ridge grid, noise model, and Pareto rule in a preregistered factorial ablation: independent reservoirs; one CZ applied once per evolution step; CZ at different representative-qubit pairs; and a shared injection qubit. Compare each against linear and quadratic readouts. Report held-out subject-independent or time-blocked NMSE, 1-PLV, DTW, confidence intervals across repetitions, and compute incremental improvement over quadratic augmentation: ΔNMSE_CZ|quadratic and Δ(1-PLV)_CZ|quadratic. The entangling mechanism is supported only if its confidence interval excludes zero and the gain replicates across component pairings and noise settings.

Prediction: cross-subreservoir coupling may help chiefly when the target contains genuine cross-frequency phase–amplitude or nonlinear dependencies; otherwise it may add parameters, noise sensitivity, and variance without improving generalization. The present evidence cannot determine which outcome occurs.