Card testing looks like a good day

1 September 2026

If you run a marketplace, the first sign of a card-testing run is usually a metric going up. Signups, or checkout attempts, or transaction count. Something a dashboard renders in green.

This is why it so often runs for days.

What it is

Someone has a list of card numbers of unknown validity and needs to learn which ones work. Your platform is a convenient oracle: attempt a small charge, observe whether it succeeds, discard the failures. The value of the exercise is the answer, not your goods — which is why the attempts are small, plausible, and spread across many accounts.

Why the obvious controls do not fire

Rate limiting by IP assumes concentration, and this is deliberately distributed across residential addresses. Velocity checks on an account assume repetition on one account, and this is one or two attempts each across hundreds. Fraud scoring on the transaction assumes the transaction is the goal, and here the transaction is the probe.

Each control is reasonable. Each is defeated by a pattern that only exists at the population level, and each looks at a single subject.

The signals that do work

The cost people forget

The chargebacks are not the main expense. The main expense is your processor's view of you. Decline ratios feed risk scoring at the acquirer, and a sustained run can move you into a review that affects your rates, your reserve, or your account — long after the run stopped, and with a much slower path back than the one in.

Which is the argument for detecting it in hours rather than at the end of the month. Not because the fraud loss is large. Because the consequence outlives it.