Model-guided experiment cycle

Fit a quadratic response model from local observations, propose a new experiment batch, enforce parameter limits, execute the proposed experiments, and verify that the measured optimum improves on the original observations. This is a numerical closed loop, not a hardware experiment.