Experimentation
Test runs before launch
Rehearse the experiment as a marked test, then throw it away without touching the numbers.
Before an experiment goes to a real audience it is worth proving that the whole chain works: the variant applies, the targeting matches, the conversion event arrives. A test run does that against the real runtime rather than a simulation.
A test experiment is a duplicate flagged as a test. It behaves like any other experiment while it runs, which is the point, and it is listed separately so a rehearsal is never mistaken for live work.
When you are done, the test is deleted outright rather than ended. Nothing it collected has to be reasoned about later, and no report has to carry a footnote explaining that one row was a rehearsal.
For checking a specific variant by eye rather than checking the pipeline, forcing a variant by URL is usually faster, and the two are complementary.