Design of Experiments

What it does

The inverse question to prediction: not "what will this formulation do?" but "what should we measure next?" DoE generates experiment plans that maximize the information earned per lab run.

Design families

• Full factorial — every combination of levels; complete but expensive.
• Box–Behnken — efficient second-order designs for 3+ factors.
• Central composite — response-surface designs with star points.
• Latin Hypercube — space-filling, ideal when the response is non-linear.
• Sobol — quasi-random sequences, excellent for many factors.

How to use it

Define the factor ranges, pick a design, and download the plan as a CSV ready for the bench. Run the planned experiments, then feed the results back into Prediction — your next model gets exactly the coverage the plan was designed to buy.