Corrective action

Fix the verified root cause and watch the defect rate fall.

What it measures

D5-D6 is a chain with three links, each with a pass/fail condition:

• A verified cause first. The corrective action targets the root cause confirmed in D4, with evidence, ideally including turning the defect on and off through the suspected mechanism. An action against an unverified cause is an experiment, and should be labeled as one.
• The action selection. Candidate actions are judged on whether they remove the cause (not just reduce the symptom), what side effects they introduce, and whether they survive contact with production, an action that depends on permanent extra vigilance is weaker than one that changes the process physics or adds an interlock.
• Effectiveness verification (D6). A defined window, lots, days, or cycles, over which the defect rate must drop to the target and stay there, measured with enough volume that the improvement is statistically real and not a quiet week.

In CAPA language: the corrective action record carries the cause linkage, the implementation evidence, and the effectiveness verification plan, the three things an auditor and a customer both look for.

How to read the output

Read the defect-rate trend against the action date. A clean step down at implementation, sustained through the verification window, is the signature of a correct cause and an effective action. A slow drift down is suspicious, it usually means the improvement came from heightened attention rather than the action itself, and it will drift back when attention moves on. No change means the verified cause wasn’t the cause, or wasn't the only one; the honest move is back to D4, not a second action stacked on faith. And distinguish correction from corrective action: scrapping and reworking the bad lot (D4’s disposition) cleans up the past, but only the D5 action changes the future.

A real use case

A coating line’s streak defects are traced, with on/off evidence, to slot-die lip damage from an undocumented cleaning practice. The corrective action set is layered: a validated cleaning procedure with the abrasive tool physically removed from the station, lip inspection added at die installation, and the die-care steps written into the control plan. Verification is defined before launch: streak rate below 0.2 defects per thousand meters across twenty consecutive lots on both shifts. The rate steps down immediately and holds through the window, and because the verification covered both shifts, it also catches that one crew needed the new procedure re-trained, which a single-shift check would have missed.

Common mistakes

• Acting on an unverified cause. Half of all “corrective actions that didn’t work” were aimed at the wrong cause from the start.
• Calling rework a corrective action. Disposition of bad parts is D4; if nothing upstream changed, nothing was corrected.
• Skipping or shrinking the verification window, declaring victory on three good lots after a defect that arrived in intermittent bursts.
• Choosing vigilance-based actions ("retrain operators, add a checklist") when an error-proofing or physical change is available. Vigilance decays; interlocks don't.
• Not screening the action's side effects, a parameter change that fixes one defect and seeds another swaps a known problem for an unknown one.