Experiments
An Experiment runs two versions of a flow side by side over the same records and shows you which ones come out differently.
It is the answer to the question that stops most rule changes: not “will this work” but “what will it do to the cases we already have”.
What you are comparing
Section titled “What you are comparing”Two sides, and each side is any published version of a flow. Usually that is today’s version against the one you are proposing, but nothing stops you comparing two candidates against each other, or a current version against one from six months ago.
What it runs over
Section titled “What it runs over”A Dataset. You choose which one, and it is the same set of records for both sides unless you deliberately give the challenger its own.
That second option exists for the case where the change includes a change to the shape of the input. When you use it, the records pair up in order, so the two Datasets have to hold the same number of records for the comparison to mean anything.
This is worth being precise about. An Experiment tells you what two versions do to a set of records. It is exactly as representative as the records you gave it. A Dataset assembled from real past cases gives you a real answer about those cases; a Dataset of ten invented rows gives you an answer about ten invented rows. The tool does not assemble it for you, and a comparison over unrepresentative records is confidently wrong rather than merely unhelpful.
What you get back
Section titled “What you get back”A run, held permanently, with a result per record: which side did what, and where they differed.
Runs accumulate rather than replacing each other, so the history of a decision is on the Experiment rather than in somebody’s memory of last Tuesday. A run in progress is visible as such, and survives you closing the page.
What it is not
Section titled “What it is not”It is not a rehearsal against production. Nothing it does touches a live case, and nothing it runs changes anything outside the Experiment.
It is not a test. A test asserts that a flow does a particular thing and fails when it does not; that is the Test centre. An Experiment asserts nothing. It shows you a difference and leaves the judgement to you, which is why it is the thing to put in front of whoever has to approve the change.