Test centre
The Test centre is where a flow is exercised before anything depends on it. It holds three things that work together: the Datasets a test runs over, the tests themselves, and the template tests that check what a run produces.
A test is an expectation, written down
Section titled “A test is an expectation, written down”A test names a flow, the records to run it over, and what should be true afterwards. Each expectation is a condition and the outcome it should produce, so a test is a claim rather than a demonstration: it passes or it fails, and a failure names the record that caused it.
That is the difference between a test and an Experiment. A test asserts something and fails when it is not true. An Experiment asserts nothing and shows you a difference. Use a test for what must stay true, and an Experiment for a change you are still deciding about.
Results accumulate
Section titled “Results accumulate”Each run of a test is recorded against it, in the order the runs were dispatched, and they build up rather than replacing each other. So a test that started failing has a history showing when, and the full trace of any run is reachable from the result rather than summarised into a pass or a fail.
A run that fans out over many records is pinned as one batch, so a result set is attributable to the run that produced it even when several are in flight.
Turning a policy counterexample into a test
Section titled “Turning a policy counterexample into a test”When a policy check finds a case where a flow does not do what a policy says it should, that counterexample can be made into a test in one step: the conditions that produced it become the test’s conditions, and the outcome the policy expected becomes the expectation.
This is the loop worth knowing about. A problem found once becomes a thing that cannot come back silently, without anybody retyping the case that found it.
Template tests
Section titled “Template tests”A flow that produces a document has a second question to answer: not only did it decide correctly, but does what it produced look right.
Template tests run a template against known values and render the result, either as the document itself or as the PDF, so a change to a layout or a style can be checked against real output rather than by opening something and squinting at it.
What belongs here and what does not
Section titled “What belongs here and what does not”The Test centre is for things that are not real. Datasets are fixtures, tests are claims about behaviour, and none of it touches a live case.
The moment you want a list your organisation actually maintains, you want a register, and that is not here. The moment you want to compare two published versions over a set of records, you want an Experiment, which is in Operate, because by then you are making a decision rather than checking your work.