Control
Sustain the gain
Control makes the improvement the new standard. The control plan anchors this: what is monitored, the metric and specification, sample size and frequency, the chart or method, who responds, and the reaction plan. Add documented procedures, training, named ownership, mistake-proofing, and a monitoring or audit schedule. Close the project with verification of realized benefits and captured lessons learned.
Statistical process control
Choose the chart by data type and subgroup size: X̄-R for continuous data with n of roughly 2–9; X̄-s for larger subgroups; I-MR for individual values; p or np for defective proportion (p when sample size varies); c or u for defect counts (u when the opportunity varies). Use rational subgroups so within-subgroup variation reflects common cause only.
Out-of-control signals include a point beyond 3σ, a run of eight on one side of the centerline, six consecutive trending points, and two of three beyond 2σ. Any signal triggers the reaction plan - investigate the process, never adjust the specification.
Traps
Using specification limits as control limits; recalculating limits to erase a signal; reporting capability for an unstable process; ending the project with no process owner or no reaction plan.
Rule of thumb: control limits come from the process (roughly the average ±3σ), never from the customer's specification. Control charts detect sustained mean shifts much faster than they detect an increase in variation, so pair them with other monitoring for dispersion.