The Question You Didn't Ask Gets Asked at the Worst Possible Time

Confidence: hypothesis. The failure modes below are drawn from published regulatory guidance and widely reported industry patterns, not firsthand deployment. The buildable pattern at the end is unbuilt: I have not shipped it. Argue with it.

Every engineer has sat in some version of this meeting. The results are in. Two reasonable people look at the same dashboard and disagree about what it means, because the system encountered a case nobody specified in advance (a user who cancelled and re-subscribed, a request that partially succeeded, a sensor reading that arrived out of order), and each person is filling the specification gap with their own retroactive judgment. Both are arguing in good faith. Both are also secretly arguing for whichever interpretation makes the number they wanted come out true. Nobody can prove the other one wrong, because the rule was never written down before the data existed to bias the writing of it.

Clinical trials have a name for this exact failure, a regulatory framework built specifically to prevent it, and a bruising institutional history of learning why the framework was necessary.

The clinical why

A trial exists to answer one precise question, and for a long time that question was left dangerously underspecified. Consider a readout meeting, years after a trial's design, where the primary result hinges on how to treat patients who took rescue medication partway through the study: one interpretation keeps the effect strong, the other shrinks it toward noise. Both camps can point to the protocol. The protocol is silent, because nobody asked the question when asking was free.