ATDD

Acceptance Test-Driven Development: before the feature is implemented, the team writes the test that will say whether it is finished — and agrees on it.

“Done” then stops being a judgement call. It is a test that was red and is now green, and it was red for the right reason.

The cycle

1. Agree the test. From the examples produced in example mapping, written as a BDD scenario or as a plain test in the team’s language.
2. Watch it fail. For the right reason — a missing behaviour, not a missing import. A test that has never failed has proved nothing.
3. Drive inwards. The failing acceptance test is the outer loop. Inside it, the usual unit-level red-green-refactor cycle builds what it needs.
4. Green, then refactor. The third step is the one that gets skipped, and skipping it is what makes people believe test-first slows them down.

Outside-in, and why

Starting from the acceptance test means the first design decision is what the system does for someone, and the internals are discovered by working inwards from that. Starting from the inside produces components that are individually tested and collectively wrong.

It also means the test is written against a port, not against an implementation detail. That is what lets the implementation be replaced later without rewriting the suite.

What it does to estimation

A story whose acceptance test cannot be agreed is a story nobody understands well enough to estimate. Trying to write the test is the readiness check — which is why this practice belongs upstream of the sprint board rather than downstream of it, and why grooming refuses to size an item whose scenarios do not exist yet.

What runs the tests

Cucumber at the use case port, with the ports stubbed from the contract by Microcks, and Playwright for the handful of journeys that genuinely need a browser. See testing tools.

What it does not replace

Unit tests still exist, and there are far more of them. The acceptance test says the feature works; the unit tests say why each piece behaves as it does and make the failure diagnosable. See all-in-one testing for how the levels fit together without duplicating intent.