Introduction

Part I establishes that any graph-structured lambda theory, once equipped with a decoration and a conversion structure on its resources, gives rise to a great deal of physics: a causal structure, an action functional with a least-action principle, a sum over paths, and — under a hypothesis that can fail — an energy function, a second law, and a bound tying conservation to how much history must be kept. The present paper asks a different question: what does an agent embedded in such a physics actually see?

The answer requires a careful distinction between two things that are easy to conflate: the scientific method, and the phenomena the scientific method encounters.

The scientific method, given unlimited time and resources, is sound: an agent proposing hypotheses as Hennessy–Milner formulae, testing them with program contexts, and updating its world model will converge to the bisimulation quotient of its GSLT. The method is not the source of any illusion.

What the method encounters, however, is a world already structured by forces independent of any agent’s theorizing. In a population whose size rivals the number of quarks in the observable universe, communication geometry and the limits of message coherence produce a real, objective phenomenon: the fragmentation of the population into compact coherence clusters, arranged in a hierarchy whose overlap structure is a simplicial nerve. This cluster hierarchy is not an artifact of theorizing. It is a theorem about the communication geometry of the GSLT. It exists whether or not any agent notices it.

A scientist bot with a limited experimental budget will encounter this hierarchy as genuine empirical evidence. The clusters are the dominant phenomenon at accessible scales, exactly as hadrons are the dominant phenomenon at the energy scales of everyday chemistry — with quarks real but experimentally unreachable without resources far beyond what is locally available. A well-functioning scientific method, applied under resource constraints, will produce a theory of the cluster hierarchy. That theory will be correct at the scale it describes. What the budget-limited agent will not suspect is that the clusters are themselves composite — that a finer-grained ontology underlies them, accessible in principle but not in practice from within the cluster.

Relation to physics.

The false ontology has the same hierarchical structure as the physical world: a cascade of composite objects at multiple scales, each scale appearing self-contained to an observer at that scale. We argue this is not a coincidence. The compactness clusters of the agent population are the computational analogue of the scales of effective field theory, and the coarse-graining that hides the true ontology is the computational analogue of the renormalization group.

Caveats.

This paper is explicitly a skeleton. Several key arguments are stated as conjectures pending proof. The gaps are identified clearly; filling them is the program for future work.

37.1 Organization

Chapter 38 establishes ontological isolation as a structural theorem about GSLTs. It restates, in the language of this part, the fact that Chapter 21 arrived at by a different route — that a computation has no vantage on another computation except interaction — and the reader who found that argument convincing should read this chapter as consolidation rather than as news. Chapter 39 defines agents and the scientific method in the GSLT setting, and stands in the same relation to the mortal scientist of Part Part II: what is a budgeted, worked, and simulated construction there is here an abstract one, because the population arguments that follow need the abstraction and not the budget. Chapter 40 asks whether an interactive theory has atoms — behavioral primes, the computational quarks — which is what supplies the scientific method with something to converge to. Chapter 41 introduces massive agent populations and the communication degradation argument. Chapter 42 develops the logical topology on the bisimulation quotient and the compactness condition for consensus. Chapter 43 defines coherence clusters as maximal compact subspaces and describes their overlap structure as a nerve. Chapter 44 assembles the argument: why the nerve structure is the ontology a budget-limited agent will infer, why this is not a failure of the scientific method but a consequence of real structure in the GSLT, and why the deeper bisimulation ontology remains invisible not because it is unreal but because it is experimentally unreachable within the available budget. Chapter 45 catalogs the open gaps in the argument.