The brain doesn't read the body — it predicts it. A model of body state is held upstairs; signals from the body are used as evidence to correct that model. When prediction and signal diverge, there's an error. The error gets resolved in one of two ways: update the prediction (learn), or send commands to make the body match the prediction (act).
The felt state — what you experience as arousal, tension, calm — tracks the prediction, not the raw signal. The body is downstream.
what this model hides: The loop here is abstract and linear — one state variable, symmetric error correction, no hierarchy. Real predictive processing runs through 6 cortical layers in multiple reciprocally connected regions; the "precision" assigned to each signal is itself a learned prediction about reliability, not a fixed number. The simulation treats emotion as a single scalar, but the body runs dozens of interacting signals (cardiac, respiratory, immune, endocrine) with different time constants and update rules. Most importantly: it can't show why any of this feels like something — that question is upstream of the mechanism and remains unanswered.