When an AI output is wrong, the first question is whether the error reached a clinician or stopped at the review layer. The second question is whether the review layer should have caught it.
We triage every clinically-significant error within twenty-four hours and we document the root cause in a publicly-readable internal report.
We do not characterize errors as edge cases. Every error tells us something about the system.
