Docs/Build/Quantum response regulation
CLIENT 0.5.1PYTHON 3.8+

Quantum response regulation

Regulate per-channel response profiles under drift and shot noise.

The supported quantum contract is a complete vector of measured channel responses and a feasible vector target. Each shot batch must provide a resolved response before drift makes it stale.

Controlled response-profile tests show four-to-six-acquisition floors and scale-flat behavior through 16,384 controls inside the qualified drift, noise, target, and coupling regime.

Where the loop must live: the accelerator stays in the workload's scaling class exactly when maintenance information crosses the host boundary at O(N) per cycle. Three closures exist: match the widths so n = O(N), terminate the maintenance path beside the substrate, or aggregate to an O(N) interface before the host. A hosted session at n much greater than N crosses the boundary by definition, so the hosted tier is for evaluation and the offline build is the deployment artifact.

Synthetic homeostasis, defined architecturally: local measurement converted into local restorative action, at one parallel acquisition per cycle, such that the maintenance burden never re-crosses the host boundary faster than the workload interface itself. A non-writable plant remains a permanent decline, so the phrase never describes maintaining a biological system.

Under an active workload the layer can maintain the substrate interface rather than taking ownership of the workload variables. Evidence tier SIMULATED: in the registered 64-channel cell, the gap between the reported objective and the commanded-circuit objective fell from 0.0772 to 0.0218. Failed comparison arms remain registered beside the positive result.

Edge-coupled QAOA profiles, extremal targets, and scalar variational energies are outside this evidence. Keep a conventional variational optimizer for those objectives.

PYTHON
opt = SWCOptimizer(key, n=len(x0), mode="regulation", target=target)
x = opt.start(x0)
for _ in range(rounds):
    x = opt.step(measure(x), target=target)
proof = opt.end()
Need the answer on your plant?

Run preflight, inspect the evidence scope, then compare the active default and your incumbent under one measured budget.

Start evaluation