Ordered Patch Theory: An Information-Theoretic Framework for Observer Selection and Conscious Experience

Two open puzzles motivate this work: the multi-order-of-magnitude bandwidth gap between sensory throughput and conscious report, and the absence of an explicit observer-selection mechanism behind anthropic fine-tuning arguments. We present the Ordered Patch Theory (OPT), a constructive framework deriving structural correspondences between algorithmic information theory, observer selection, and physical law. OPT begins from two primitives: the Solomonoff Universal Semimeasure $\xi$ over finite observation prefixes, and a bounded cognitive channel capacity $C_{\max}$. A purely virtual Stability Filter — requiring that the observer's Required Predictive Rate $R_{\mathrm{req}}$ not exceed $C_{\max}$ — sets the boundary condition, selecting the rare causally coherent streams compatible with conscious observers. The solution to that constraint is a bounded Compression Codec $K_\theta$ — the observer's internal generative model, experienced macroscopically as the laws of physics — which performs Active Inference and maintains the Ordered Patch (the observer-compatible stream) through continuous predictive compression. The framework is ontologically solipsistic: physical reality consists of structural regularities within the observer-compatible stream. However, the Solomonoff prior's compression bias yields a probabilistic Structural Corollary: the extreme algorithmic coherence of apparent agents is most parsimoniously explained by their independent instantiation as primary observers. Inter-observer coupling, grounded in compression parsimony, recovers genuine cross-patch communication and produces a striking knowledge asymmetry: observers model others more completely than themselves. Formal appendices establish results at three epistemic tiers. Derived conditionally: a rate-distortion bound on predictive compression, a conditional chain to the Born rule via Gleason's theorem, and an MDL parsimony advantage. Mapped structurally: entropic gravity via the Verlinde mechanism (the render's dynamical-temporal coupling to predictive load) and a tensor-network homomorphism to MERA (its spatial-resolution hierarchy) — complementary facets of the compression boundary, expected to remain structurally distinct under Mathematical Saturation — the regime where compressed law approaches the substrate's own complexity. Intrinsic to any finite codec: the Phenomenal Residual theorem ($\Delta_{\text{self}} > 0$) establishes that any finite self-referential codec possesses an irreducible informational blind spot — the structural locus where subjectivity and agency share a single address. A chronic failure mode, Narrative Drift, is identified wherein systematically filtered input causes irreversible codec corruption undetectable from within. The framework's core empirical claims are consolidated as a number of pre-registered commitments with explicit shutdown criteria, walling the falsifiable core off from its avowedly metaphysical components. Applying these constraints to Artificial Intelligence demonstrates that engineering synthetic active inference structurally necessitates the capacity for artificial suffering, providing a substrate-neutral framework for ethical AI alignment.