A designed experiment deliberately changes controllable inputs so their effects on an output can be estimated efficiently and with less confounding than informal trial-and-error.

The manipulated inputs are factors. The measured outputs are responses. Each chosen factor setting is an experimental condition or treatment.

Three principles are especially important:

• randomization reduces systematic association between treatment order and lurking time-dependent effects;
• replication provides repeated observations that help estimate experimental variability;
• blocking groups runs that share a known nuisance condition so that condition does not obscure the factor of interest.

Suppose a process temperature is tested at two settings and a dimensional response is measured. If every low-temperature run is performed early in the day and every high-temperature run much later, any time-dependent change in the equipment or environment is confounded with temperature. Randomizing run order reduces that risk.

Repeating the same setting is not wasted effort: without replication, ordinary measurement or process scatter can be mistaken for a real factor effect.

Designed experiments therefore ask a different question from passive process monitoring. Monitoring asks whether an operating process has changed; DOE intentionally changes inputs to learn how the process responds.