A new customer paid, then got refunded. Does the buyer disappear?

Keep the first positive payment event and the later refund event. Then decide which business question the report answers: customers who first paid, customers whose payment was retained, or net cash received. Those are different measures.

For a channel comparison, write down the observation window. A recent cohort has had less time to refund than an older one. Comparing them without that context can flatter the newest campaign.

The useful table has a stable acquisition definition plus explicit refunds and retained revenue. You can revisit the decision when the cohort matures without pretending the original payment never happened.