One line in your loop.

You give the runtime a measured statistic and a score each round; it returns the next configuration. The update law runs on our endpoint — you never receive it, so there is nothing to leak in either direction.

Integrate

copy# pip install nothing — the client is standard-library Python
from swc import SWCOptimizer

opt = SWCOptimizer(license_key="EVAL-...", n=len(x0))
x = opt.start(x0)
for _ in range(rounds):
    p_hat = measure(x)              # one measurement → a vector
    x = opt.step(p_hat, score=objective(x))
opt.end()

Two modes

optimization — pass score= each round; the runtime forms its own target. regulation — pass target= to drive a measured output to a setpoint under drift. Both are shown in the examples/ folder of the download.

It tells you where it wins

Call opt.check_envelope(...) and it returns USE, MARGINAL, or DECLINE. It declines static-ample-budget problems, hand-tuned PID, and coupled plants with no calibration — the regimes where it only ties your current method.