You add an approval gate to reduce production defects. Six months later defect rates are higher. Explain.

What the interviewer is testing

Whether you can reason about second-order effects and feedback loops rather than about the direct effect of a control.

The mechanism

The gate added latency to every change. Teams responded rationally: they batched changes to reduce the number of approvals.

Larger batches are:

• Harder to review meaningfully, so the approval itself becomes shallower
• Harder to diagnose when something breaks, because there are many suspects
• More likely to contain an interaction defect between changes
• Riskier to roll back, since reverting one change means reverting all of them

So the intervention intended to improve quality increased batch size, which reduced quality. The gate is working exactly as designed and the system's response defeated its purpose.

A secondary loop: the approver, seeing large batches they cannot meaningfully assess, approves on trust — so the control's actual detection value falls towards zero while its latency cost remains.

The general pattern

Interventions change behaviour, and the behavioural response frequently dominates the direct effect. Other examples of the same shape: alert thresholds lowered to catch more issues cause muting, which causes missed incidents. A coverage target produces assertion-free tests. Chargeback for platform usage causes teams to build shadow alternatives.

What to do instead

Automate the checks the gate was performing — tests, scanning, policy — so verification costs nothing per change and does not create the batching incentive.

Reduce batch size deliberately and measure change failure rate, which the DORA research consistently shows improves with smaller batches.

Reserve human approval for the small set of changes where judgement genuinely changes the outcome.

What a strong answer adds

The habit that catches this class of problem in design: after proposing any intervention, ask what people will do in response, and check whether that response undermines the goal. Governance, incentives and process are part of the system, and an architect who reasons only about the technical half will be repeatedly surprised by the other one.

Common weak answers

Concluding the gate needs to be stricter. Attributing the increase to unrelated causes.