The brain is a guessing machine. Higher layers PREDICT what the senses should report, send the guess DOWN, and the senses pass UP only what the guess got wrong — the error. Predict well and the channel goes quiet; get surprised and a loud signal races up. It’s why you tune out a steady hum and snap to a sudden change. Slide how well the prediction matches.
Predictive coding (Rao & Ballard 1999): each cortical level sends a top-down PREDICTION down, and the level below returns only the residual PREDICTION ERROR (input − prediction) — the brain transmits surprise, not raw data, an efficient code since the world is largely predictable. Match → error → 0 and the channel is quiet (adaptation, sensory suppression); mismatch flares (a mismatch-negativity). A fail-loud self-check throws unless a well-predicted input yields near-zero error and a surprising one yields a large error. ◆ real theoretical neuroscience, node-verified.
A single-level input−prediction residual; full predictive coding is hierarchical and precision-weighted, and ‘the brain minimises free energy’ is a framework, not settled fact — flagged. The error-goes-quiet-when-predicted behaviour is the exact core mechanism.