Status: pre-launch. This page describes target design; this document does not establish a deployed service.
McGilchrist Mapping — Left, Right, Callosum
Current reading — 2026-09-27
The hemispheric comparison is an interpretive routing lens. Its compute distribution, biological equivalences and performance claims remain unmeasured hypotheses. [I]
This is an editorial boundary, not a new runtime contract, upstream repeal or proof of adoption. See current status and the product roles. Original revision dates remain as recorded below.
Historical research draft — retained on 2026-09-27; not current terms or implementation instructions
Iain McGilchrist's hemispheric hypothesis maps precisely onto the Rosetta levels and has direct implications for how Skyzai allocates compute. This is not a loose analogy. The mapping is structural and produces testable predictions about AI model selection.
The Hemispheric Mapping
| Mode | Rosetta Levels | Attention Type | Cognitive Style | AI Implication |
|---|---|---|---|---|
| LEFT | L1, L3, L6 | Narrow, focused | Analytic, sequential, categorical | Cheap models (small, fast, deterministic) |
| RIGHT | L2, L5 | Broad, contextual | Holistic, relational, embodied | Capable models (large, slow, probabilistic) |
| CALLOSUM | L4, L7 | Integrated | Synthetic, paradox-holding | Best models (frontier, expensive, integrative) |
Why These Assignments
LEFT hemisphere (L1, L3, L6): These levels share a property — they isolate. L1 (point/distinction) draws hard boundaries. L3 (ego/rationalism) categorizes and sequences. L6 (dissolution/mysticism) strips away context entirely. All three modes narrow attention to extract signal. A small language model does this well: classification, extraction, rule-following.
RIGHT hemisphere (L2, L5): These levels share a property — they connect. L2 (field/empiricism) senses the whole environment. L5 (witness/idealism) holds the whole picture without reducing it. Both modes require broad contextual awareness. A large language model with wide context windows excels here: synthesis, pattern recognition, anomaly detection.
CALLOSUM (L4, L7): These levels integrate opposites. L4 (equator/pragmatism) balances phi and nu — it holds both hemispheres in productive tension. L7 (north pole/axiomatic) transcends the distinction entirely. Integration requires the best models because it requires holding contradictory frames simultaneously without collapsing either.
The Boltzmann Distribution Applied to Compute
Compute allocation follows a Boltzmann distribution across the seven levels:
P(level) = exp(-E(level) / kT) / Z
Where:
- E(level) = energy cost of the cognitive operation
- kT = the organism's compute budget (temperature)
- Z = partition function (normalization)
Low-energy operations (LEFT): L1 classification, L3 rule-checking, L6 filtering. These are high-frequency, low-cost. Run them on cheap models (BitNet ternary, small language models, deterministic scripts). They constitute 80%+ of all compute cycles.
Medium-energy operations (RIGHT): L2 environmental scanning, L5 comprehensive assessment. Medium-frequency, medium-cost. Run them on capable models (GPT-4 class, Claude Sonnet class). They constitute 15% of compute cycles.
High-energy operations (CALLOSUM): L4 integration (the Court Deliberation in APU), L7 transcendent synthesis (organism-level P-score assessment). Low-frequency, high-cost. Run them on frontier models (Claude Opus class, o3 class). They constitute <5% of compute cycles but produce the highest-value decisions.
Skyzai Implementation
The organs naturally distribute across the hemispheric map:
- Circle (IS): Primarily LEFT (L1 classification of sources, L3 reliability scoring) with RIGHT support (L2 environmental scanning). Cheap models dominate. EigenTrust is pure L1.
- ReFu (COULD): Primarily RIGHT (L2 market sensing, L5 crowd wisdom aggregation) with LEFT support (L3 LMSR pricing mathematics). Capable models dominate.
- Agentz (SHOULD): Primarily CALLOSUM (L4 Court Deliberation integrating all directorate inputs). Frontier models required. This is where compute cost concentrates.
- Skyzai execution: Primarily LEFT (L1 transaction execution, L3 settlement logic). Cheap models or deterministic scripts. Execution should be mechanical.
The Economic Consequence
eta = 0 means the organism cannot afford to waste compute. The McGilchrist mapping ensures that 80% of operations run on models that cost fractions of a cent, 15% on models costing cents, and <5% on models costing dollars. The Boltzmann distribution is not aesthetic — it is the only allocation that sustains the organism economically under zero extraction.
LEFT for the many. RIGHT for the few. CALLOSUM for the decisive. Boltzmann allocates.