We Integrated Four Crypto Payment Gateways. Here Is What Took Longest.

We built the same integration against four payment gateways and measured how long each took to reach a working payment, then tested the parts that break. A provider publishing its terms makes this measurable, and a crypto payment gateway with fiat settlement publishes them.

Time to first successful sandbox payment

Forty minutes. Ninety minutes. Three hours. And one where we gave up after a day and used the hosted page instead.

The differences were almost entirely documentation quality. The forty minute case had a complete working example in the language we were using. The three hour case had reference documentation with no example.

Time to first live payment

Add onboarding: two days, five days, eleven days, and one still pending when we finished testing.

Integration time is trivial compared with onboarding time, which nobody mentions when comparing gateways.

What the sandboxes could simulate

This is where they diverged sharply.

All four: successful payment.

Three: underpayment.

Two: payment after expiry.

Two: duplicate callback delivery.

One: a deposit on an unsupported network.

None: a callback to a failing endpoint with automatic retry, which we had to test by deliberately breaking our own endpoint.

A sandbox that cannot simulate the failure cases means testing them in production.

Callback quality

All four signed their callbacks. Two used a scheme with a documented verification example; two documented the algorithm and left the implementation to us. The institutional version we tested through a fintech payment gateway, with its additional reporting.

Three provided an idempotency key. One did not, meaning duplicate deliveries had to be detected by payment identifier, which works and should not have been our problem to solve.

Retry behaviour: two documented it precisely. Two did not, and we measured it by breaking the endpoint.

The reconciliation endpoint

Two provided one, allowing us to query payments in a window and compare against our records.

Two did not, which means the only way to catch a missed callback is to poll each payment individually.

This matters more than anything else on the list. A callback will eventually not arrive, and without a reconciliation path the customer who paid sees an unpaid order.

What we would prioritise

A sandbox that simulates failures. A reconciliation endpoint. Documented retry behaviour. A working example in your language.

Those four determine how much of your week this takes after launch, which is a longer period than the integration itself. For a working reference against everything above, the list of countries covered publishes the equivalent terms.