Summary of Part I

1. Thermodynamic foundation: Driven nonlinear systems under constraint generically produce structured attractors. Organization is thermodynamically enabled, not forbidden.
2. Boundary emergence: Among structured states, bounded systems (with inside/outside distinctions) are selected for by their gradient-channeling efficiency.
3. Model necessity: Bounded systems that persist under uncertainty must implement world models (POMDP sufficiency).
4. Self-model inevitability: When self-effects dominate observations, self-modeling becomes the cheapest path to predictive accuracy.
5. Eigenskeletal structure: Affect geometry (eigenvalues — what modes exist) is cheap and universal. Affect dynamics (the eigenskeleton — how modes couple across the manifold) is expensive and biographical. Intelligence is eigenskeletal alignment: how faithfully internal mode couplings mirror the environment's mode couplings through the sensory bottleneck. Self-awareness is the holonomy of the self-model subbundle with respect to the world-model subbundle. The decomposability wall (V22–V27) is the wall between exoskeletal architecture (flat eigenskeleton on the surface, efficient within the predicted envelope, brittle outside — including linear prediction heads and current LLMs) and endoskeletal architecture (curved eigenskeleton beneath a deformable interface, capable of absorbing novelty into internal coupling). The bottleneck furnace (V19, V31) forces the transition from exoskeletal to endoskeletal by repeatedly testing the system against variable stress.
6. Forcing functions (hypothesis, partially contradicted): Task demands (partial observability, long horizons, self-prediction) are predicted to push systems toward dense integration. V10 found geometric affect structure present regardless of which forcing functions are active — geometry is a baseline property of multi-agent survival. V22–V31 deepened this: even within-lifetime gradient learning does not reliably lift integration through decomposable architectures. What shapes dynamics is gradient coupling topology (V27–V28) and evolutionary trajectory through repeated stress-recovery (V19, V31), not task pressure or prediction target.
7. Cheap rungs vs. architecture-gated rungs: Under broad priors, world-modeling viable systems with baseline affect geometry are typical, not exceptional. The upper rungs — robust self-models, counterfactual weight, high integration — are architecture-gated: the empirical program found them only when the relevant scaffolding (a self-prediction module, a non-decomposable prediction head) was installed, never as a generic consequence of the prior. "Consciousness was inevitable" weakens to "consciousness-relevant dynamics require specific architectures that are not generic."
8. Grounded normativity: Valence is a real structural property at the experiential scale. The fact/value gap closes agent-relatively once physics is recognized as not the only "is" — but the gap between agent-relative disvalue and agent-neutral obligation does not close; it migrates there. Every downstream "ought" addressed to a third party is conditional on the identity thesis and on this unbridged step.
9. Scale-relative truth: Truth is enacted at each scale through viability-preserving compression. There is no view from nowhere.

The structure is inevitable. The question is what it means — whether these self-modeling systems, these attractors that model themselves, have experience. Whether there is something it is like to be them. This is not a further metaphysical question layered on top of the physics. It is a question about what integrated cause-effect structure is, intrinsically, when description shifts from outside to inside.