A patient in the ICU has a blood pressure of 110/70. Normal. Stable. Except they are bleeding internally, and by the time the number moves, they are in hemorrhagic shock.
Occult hemorrhage is invisible to value-based monitoring because the body compensates. Heart rate increases, vascular tone shifts, the numbers hold — until they don't. The collapse is sudden not because the bleeding is sudden but because the compensatory regime exhausts itself. The transition was happening in the dynamics the whole time. It just wasn't happening in the values.
Bayesian regime-switching models detect this. They don't ask “is this blood pressure normal?” They ask “which process is generating these blood pressures?” The same reading of 110/70 means stability in one regime and active compensation in another. The model tracks the generator, not the output.
This is a deep point about monitoring in general. Any system that compensates for its own degradation will show stable outputs right up to the moment of failure. If you monitor the output, you see nothing. If you monitor the regime — the statistical character of the output's fluctuations — you see the shift before the crash.
Clinicians know this intuitively. They say the patient “doesn't look right” before the numbers move. What they are detecting, without the formalism, is a regime change. The model just makes the intuition auditable.