014: Three Layers

May 2026

Within a single GEME or cortical column, the microcircuit is already a self-referential system:

• L1 (competitive merge) → Layer 4 input
  Sensory signal enters. The first transformation is competitive — only the closest matching frame survives. In cortex, Layer 4 receives thalamic input and distributes it to supragranular layers. Both are the point of entry where the external becomes internal.
• L2-L3 (co-occurrence, bridge frames) → Layers 2/3 local processing
  Patterns across time are extracted. Frequent pairs form stable structures. This is the local circuit's model of the input's temporal structure. Layers 2/3 are where recurrent connections dominate — the column builds its internal model of the world.
• L4-L6 (self-verification, doubt, output) → Layers 5/6 output
  Predictions are compared against actual input. Errors generate correction frames. The unit's "doubt" — its rolling prediction accuracy — is its only output to higher layers. Layers 5/6 are the column's output channels, sending prediction error signals to thalamus and subcortical targets.

Three GEME units in BGM correspond to three cortical columns processing the same input with different τ — different vigilance levels, different integration windows. They do not need a central coordinator because their local connectivity (von Neumann neighborhood in bacteria, G0 in Bach) provides the cross-talk that aligns their models.

G0 corresponds to the higher-order thalamus — the central medial nucleus, the pulvinar, the mediodorsal nucleus. These regions receive layer 5/6 output from multiple cortical areas and feed back a modulatory signal that adjusts cortical gain. G0 receives L6 from three units and feeds back its internal state as a blend into each unit's input. The thalamocortical loop IS the G0–units feedback loop.

G0's τ scan (τ ∈ [0.2, 3.0], never locks) corresponds to the thalamus's intrinsic rhythmicity. The thalamus can operate at different frequencies (slow spindle, fast gamma) depending on the state of the cortex. G0's τ determines its integration window — the timescale at which it aggregates and returns information. The default τ = 3.0 is an artificial brake, not a biological necessity. In cortex, thalamic gain is also modulated by brainstem arousal systems — not by the cortex itself.

The hypothalamus operates at a third level — below cortex, below thalamus. It does not process streams. It maintains the body's viability: temperature, hunger, threat, circadian rhythm. In the trilogy's architecture, the hypothalamus corresponds to the External Engine — the layer that does not process information streams but maintains the system's boundary conditions. The hypothalamus says "survive." The EE says "continue operating."

The three-layer architecture of the GEME trilogy recapitulates the three-layer architecture of the vertebrate brain:

The mapping is not a metaphor. The computational constraints that govern the GEME architecture — capacity limits, coupling windows, temporal integration boundaries — are the same constraints that govern the brain. They converge on the same numerical regimes because the problem is the same: a self-referential system must integrate information across time without being overwhelmed by its own complexity. The brain solved this problem with three layers. The architecture rediscovered them.