Here is an uncomfortable observation from two decades of running quality in automotive plants: when the same failure family walks back through the door, you do not have a new problem. You have the report card on the last one. Volkswagen is recalling thousands of electric vehicles over battery-fire risk again. I have no visibility into their paperwork and I will not diagnose a root cause from a headline. But I know what a repeat escape means: somewhere in the first corrective action, effectiveness was asserted rather than verified. That is not bad luck. That is a closure process failing quietly until the fleet calls in the result.

A repeat recall is not a new problem

Recurrence is the only honest audit of a corrective action. Everything else – the sign-offs, the closure review, the relieved customer e-mail – is opinion until the same parts have survived the same process at volume. So when a failure family returns, the first question is not what new defect appeared. It is what the first corrective action actually verified.

The classic leak is containment scoped to the visible build window, not the window that shared the cause. Field data lags: complaints cluster where usage is severe and weather unkind, not necessarily where the cause lives. If a supplier's sub-component drifted over five weeks and incidents only surfaced from vehicles built in week three, a boundary drawn at week three is a boundary drawn by noise. The cause does not read your VIN list.

I once watched a customer escape get contained to a single week of production, because that was where the complaints clustered. Traceability later showed the suspect supplier lot had been running for three weeks before anyone noticed. The other two weeks were already in the field, unfiltered. We reopened containment at our own cost, before the customer did it for us – still the cheap option, because a battery campaign starts at four figures per vehicle before you count logistics, loaners and a second engineering analysis.

Where verification goes to die

Three mechanisms do most of the damage, and none of them looks like negligence from inside the building.

  • Symptom removal mistaken for cause removal. A containment filter at end of line catches the defect, PPM drops, the dashboard goes green, someone closes the 8D. The failure mode never left – it was being caught. Then the filter is quietly retired at a milestone, or a new shift never learns it, and the symptom walks back in. A filter is a tourniquet. Tourniquets come off.
  • The supplier's fix parked in a requalification queue. I have heard "we adjusted the parameter" in enough VDA 6.3 supplier audits to translate it: usually it means "we ran the process until it passed". The durable fix – new tooling, a requalified sub-component, a revised PPAP – sits in a queue for months while production runs on interim controls nobody owns after the first fortnight. The requalification report lands long after closure. Nobody reopens anything.
  • Effectiveness signed off on three good parts and a deadline. A sample of three. Thirty days without recurrence, measured across a low-volume month. A closure date circled on someone's calendar. Nothing closes an 8D faster than a quarter-end.
Containment buys time; verified effectiveness pays it back. Everything else is hope with a signature on it.

What verified effectiveness actually looks like

When I built the QA and QC department of a 900-plus employee greenfield plant from the ground up, one rule went into the procedures on day one: no 8D closes on opinion, only on recurrence data. Every corrective action carried a verification window, and the exit criteria were written before the window opened – so many units of production or so many days, whichever came last. Thresholds came from baseline data. The PFMEA and control plan were updated; for supplier fixes, requalification was signed with data attached, not promised.

At SNOP those rules paid out in numbers: defect costs down 70%, and a full quarter with zero critical customer escalations. The discipline underneath was unglamorous. Containment stayed open until recurrence data cleared, not until the customer stopped calling – customers stop calling long before your process stops failing, and silence is latency, not evidence. We ran a daily QRQC cadence on every open item, escalated to A3 when a cause crossed departmental lines, and kept both alive until the data cleared, never until the meeting got shorter. Recurrence and escape rates sat on the dashboard beside PPM, reviewed at the same cadence, because a repeat escape is a defect in your closure process. It belongs on your KPI wall, not in a fresh 8D with a fresh team and amnesia about the first one.

Key takeaways

  • Scope containment by cause, not by complaint. Traceability defines every build window that shared the root cause – the visible window is where usage spoke first, not where the cause lived.
  • Write exit criteria before the verification window opens: units produced or days elapsed, whichever comes last, thresholds derived from baseline data.
  • Keep the QRQC cadence running until the data clears, not until the meeting does. A shorter meeting signals fading attention, not a solved problem.
  • Treat a supplier's "we fixed it" as a claim, not evidence. Requalification signed with data is the exit gate, and interim controls need a named owner and an end date.

The second recall costs more than the first. Always. Euros are the smallest part of it – doubled logistics, campaigns on vehicles you already touched once. A regulator forgives a failure but audits a pattern. And nobody puts a faulty window switch on the evening news, while everybody remembers the words battery and fire in one sentence. So treat every closed 8D as what it is: a hypothesis the fleet is still testing, at speed, in customers' driveways. This second recall is not the failure mode returning. It is the first closure, presenting its invoice.