Experiments fail quietly when instrumentation drifts mid-test or when the primary metric was never locked. Before arguing about lift, confirm the event still fires the same way it did on day one.
We separate integrity checks from interpretation. First: did the test measure what we claimed? Second: given that, what should we do?
A readout clinic ends with three options — ship, iterate, stop — each with a sentence of evidence. Ambiguous “interesting learnings” without a decision waste the next sprint.
Secondary metrics can inform nuance but should not overturn a pre-registered primary unless the team explicitly re-scopes the question.