Two maps of the same model

The most current version of the architecture. Four levels of prior-based refinement with neuroanatomy labeled (brainstem → subcortical → insula/ACC → hippocampus/limbic), the concept capture mechanism, the DMN as constructed simulation, and the predictive stream flowing back down. The tank's allostatic report enters at Level 3. Updates flow downward only to the level where the intervention reaches.

Prediction error enters the tank. Burners add heat. Temperature determines what the signal can do — protect or travel. Four compression gates carry it downstream from pressure to narrative. The fork between model-fidelity and data-fidelity. The complete mechanistic architecture in one diagram.

Seven parallelogram gates — each one denser than the last, each one further from the source. Pressure, reflex, premises, concepts, emotion/strategy, narrative, symbolic. Updates travel one gate at a time backward, which is why insight at the narrative layer cannot reach the procedural and somatic priors. The visual argument for working upstream.