Measurements per round

Choose enough reads to make the statistic useful, then stop.

The runtime needs one statistic per update, but that statistic may itself aggregate a shot batch or repeated sensor reads.

Use the smallest batch that resolves the plant response reliably. Larger batches improve each acquisition but reduce how many updates fit in the deadline.

Report acquisitions and raw reads separately so a parallel shot batch is never confused with a free measurement.

In registered sweeps, averaging 16× past the best tested setting produced 1.8–2.5× more error because the plant moved while the read was being averaged. When uncertain, begin with a smaller batch and verify stability.

Inside the tested envelope, shorter acquisition cycles reduced steady error. Treat the correct batch size as a measured property of the deployment, not a universal constant.

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)
opt.end()