Emergent Spacetime from Self-Referential Computation: A Hierarchical Cellular Automaton Framework

This paper proposes a cosmological model — the Singularity-Bounded Holographic Class 4 Automaton (SB-HC4A) — derived from the convergence of four independently motivated frameworks: a five-class computational taxonomy that refines Wolfram's (2002) classification by separating fractal from random dynamics, a theoretical framework for self-referential computation in self-modeling systems (Gruber, 2015, 2026a, 2026b) which identifies self-referential simulation at criticality as a universal computational pattern, and 't Hooft's (1993, 2016) holographic automaton interpretation of quantum mechanics. The model proceeds by elimination: Classes 1–3 cannot sustain the universal computation the universe demonstrably supports; Class 5 (genuine randomness) makes physics fundamentally impossible; therefore the universe operates at Class 4 — the edge of chaos. Combined with the information-theoretic observation that singularities at every physical scale (Planck regime, particle interiors, event horizons, cosmological horizons, temporal endpoints) share the property of information impermeability and Bekenstein saturation, the model proposes that these singularities are structurally identical — scale-invariant instances of the same information boundary. The resulting architecture is a self-referential holographic Class 4 automaton bounded at every scale by singularity surfaces, where the observable interior is the "simulation" and the singularity boundary is the "substrate." All singularities — including temporal endpoints — are shown to be asymptotically unreachable from within the computational domain, strengthening the unification claim. Because singularities transform rather than destroy information, heat death constitutes a singularity transition that triggers cyclic renewal, with potential CPT signature alternation across cycles — connecting to Penrose's Conformal Cyclic Cosmology and Boyle and Turok's CPT-symmetric universe. All three cosmological endgames — heat death, Big Crunch, and Big Rip (Caldwell, 2002) — drive the computational domain to Bekenstein saturation, with the Big Rip uniquely producing a branching tree of daughter universes rather than a linear successor. This architecture is structurally identical to self-referential computational systems that operate at criticality, where implicit knowledge (substrate) is separated from explicit representation (simulation) by an information-opaque boundary. Self-modeling cognitive systems are thus local, scale-reduced instances of the same computational pattern the universe implements globally. Six weak points are identified, including the fundamental epistemological objection that Class 4 observers may be constitutionally incapable of determining whether this model describes the universe or merely the ceiling of their own computational capacity. Changelog v6 **v6 (2026-08-27) — one symbol was carrying two different claims** The fixed-point condition Φ(U) = U appears throughout the paper, and it has two readings. Under one, Φ is whatever the dynamics produce, and the condition holds of every deterministic system and asserts nothing. Under the other, Φ composes holographic encoding with decompression, and the condition says that encoding the universe onto its boundary and decompressing it again returns the universe — a losslessness commitment whose denial is information destruction at a singularity boundary. Section 6.3 already stated the second reading and rejected the first. Three other places had not caught up. *The incompleteness argument no longer rests on the fixed point.* Section 6.4 argued that a self-referential system of sufficient complexity contains truths no formal subsystem can derive, and applied Gödel's theorems on that basis. Gödel's theorems govern formal systems that represent arithmetic and represent statements about their own derivations, which is a condition on representation; the fixed point of Section 6.3 is syntactic and requires nothing to be about anything. The argument now proceeds from universality together with embedding instead: a Class 4 automaton is computationally universal, so it implements formal systems of the kind the theorems govern, and every physical theory is such a system implemented inside the universe it describes and a proper part of its own subject matter. The conclusion is unchanged and no longer depends on a premise that cannot support it. *The cross-scale mapping names which operator is which.* The Section 7.1 correspondence table set Φ(U) = U beside the cognitive Φ(m\*) = m\* without distinguishing them, so a reader consulting the table alone would take the shared symbol for a shared operator. Each side now names its own sense. Section 7.2 asserted that "the laws of physics are the universe's dynamics applied to itself," which is the reading Section 6.3 rejects, and now states the losslessness condition; the same section's inexpressibility item cites the argument of Section 6.4 rather than a bare appeal to Gödel. *The companion formalization is revised in the same pass and is not separately deposited.* Its Section 5 defined Φ as the "compute the output" operator, applied Lawvere's theorem in a form that used the existence of a fixed point to license identity with the actual universe, and inferred from closure that the universe is a formal system containing itself. The fixed point is now defined as the composition; Lawvere's theorem is reported as supporting existence, with uniqueness identified as what the claim still needs and the final-coalgebra route named as the candidate; point-surjectivity is stated as an open task rather than as satisfied; and the incompleteness argument runs from universality and embedding as in the parent paper. Its functor table also carried a row asserting information conservation across the implicit-explicit boundary, which conflicts with the selective projection the parent paper describes; both sides now state conservation of the total with a selective projection. No numerical result, prediction, reference or falsification criterion changed. The paper is 72 pages and its 147 references resolve as before.