Pull ten closed nonconformance records from almost any quality system and read the closure field. A significant fraction will say some version of: discussed with supplier, operator retrained, part reworked and accepted, root cause understood.
None of those are closures. They are the reasons the same NCR will be written again.
The problem
A nonconformance record has two jobs, and organizations reliably do the first one. The first job is to deal with the affected product: identify it, contain it, decide whether it is reworked, repaired, accepted by deviation or scrapped. That is a disposition, and quality systems are generally good at it because the product is sitting there being expensive.
The second job is to make sure the process that produced the nonconformance can no longer produce it undetected. That job has no physical prompt. Nothing is sitting on the floor reminding anyone it is unfinished, and the metric most organizations track — NCR closure rate, or average days to close — actively rewards skipping it. Closing an NCR quickly is easy if closure means the product has been dispositioned.
So the recurring failure mode is not carelessness. It is that the word “closed” is doing two different jobs and only one of them gets audited.
A framework: four gates
I use four gates, in order. An NCR is not closed until all four are satisfied, and each one requires a different kind of evidence.
1. Containment — is the affected population known and controlled?
Not “is the defective unit segregated,” but: what is the population that could share this condition, and how was that population defined? The scoping rule is where containment usually fails. If a defect originated in a purchase order, scoping by build date will miss stock. If it originated in a tooling change, scoping by part number will miss other parts made with the same tool. Write down the rule, not just the list.
2. Cause — does the stated cause account for the observed effect?
This is the gate that separates a root cause from a plausible story, and it is almost never applied. A cause should predict the magnitude of what was measured, not merely be consistent with its existence.
A concrete version: in a synthetic cooling-skid investigation I wrote up, a skid delivered 195 kPa of differential pressure against a 250 kPa requirement. The cause found was a pump impeller one trim size too small — 190 mm where the design called for 216 mm. Pump head varies roughly with the square of impeller diameter, so the cause predicts (190 ÷ 216)² × 250 ≈ 194 kPa. Observed: 195 kPa.
That agreement does two things. It confirms the cause, and — more importantly — it closes off the possibility of a second cause, because there is no unexplained residual left to hide in. If the impeller had accounted for only 20 kPa of a 55 kPa shortfall, the investigation would not have been closable at all, no matter how real the impeller error was.
3. Action — does the action attack the systemic cause?
There are almost always two causes: the thing that went wrong, and the reason nothing caught it. Replacing the impeller addresses the first. It does nothing about the second, which in that case was that two functionally different impellers shared a catalogue description and no control anywhere between ordering and assembly verified the trim.
A useful test: if the same mistake were made again tomorrow, at what point would it be caught? If the honest answer is “at the performance test,” the corrective action is incomplete — you have not changed the process, you have only fixed one unit and confirmed that your last line of defence works.
4. Effectiveness — is there evidence that the action worked?
This gate needs a defined sample and a defined interval, decided when the action is approved rather than when someone wants to close the record. Three consecutive conforming units, or thirty days of production, or the next two receipts — the specific rule matters less than the fact that it was set in advance.
One conforming unit after a repair proves the repair worked on that unit. It says nothing about whether the process now reliably produces conforming units, which is the entire claim being made by closing the record.
The mechanism that makes this hold
Defining the evidence type at the moment the NCR is raised, not at the moment someone wants to close it.
This is a small procedural change with a large effect, and the reason is about incentives rather than rigour. When the evidence requirement is set at closing time, closure becomes a negotiation — and the party arguing for a lower standard is always the party under schedule pressure, which is to say everyone. When it is set at raising time, nobody yet knows what the evidence will cost, so the standard gets set on technical grounds. Closure then becomes a factual question: does this evidence exist, yes or no.
It is exactly the same reason acceptance criteria are negotiated before a run-off rather than during one.
Engineering implication
If you audit one thing in a nonconformance process, do not audit closure rate. Audit whether an effectiveness verification step exists at all, and then sample closed records for whether it was actually performed with a pre-defined sample.
An organization with a 95% on-time NCR closure rate and no effectiveness gate is not performing well. It is closing records. Those are different activities, and the second one produces a quality system that looks healthy in every report right up until a defect escapes to a customer who then asks how a known condition was allowed to recur.
This is also the reason I treat “NCR closure effectiveness not demonstrated” as a major finding in a supplier assessment rather than a minor one. A supplier that raises and disposition nonconformances well but cannot show that its corrective actions work has a quality system that catches defects and never stops producing them.
Takeaway
An NCR is closed when the affected population is controlled, the cause accounts for the observed effect, the action addresses why nothing caught it, and there is evidence — against a sample defined in advance — that the action worked.
Everything else is a record that has been marked closed.