Reduce variation by tightening decision criteria, not by asking inspectors to "be more consistent." The practical levers are clearer repair standards, better visual exemplars, calibrated measurement methods, structured escalation paths, and traceable feedback on borderline calls. Results depend on data quality, training discipline, gage capability, and how well shop-floor tools stay aligned with controlled documents.

There is no single universal minimum. In practice, about 25 to 30 rational subgroup averages, or roughly 100 to 125 individual observations, is a common starting point for a preliminary capability study. More data is usually needed when the process is unstable, mixed across setups or lots, measured poorly, or run at low volume.