COS45 — A Geometric Detectability Framework for Noise-Constrained Inference (v1.0 11)
Louis Morissette
Abstract
COS45 is a geometric framework that defines when an inference is structurally admissible under noise constraints. It does not produce conclusions. It defines a boundary of admissibility. The framework is based on a minimal quantity: η = δ − τ where: δ represents observable structure derived from data τ represents the uncertainty threshold (noise floor, perturbation, or resolution limit) An inference is admissible if and only if:η > 0 COS45 defines a geometric admissible region Ω_coh under three simultaneous conditions:η > 0, T = 1, R ≥ R_min where: T (Traceability) ensures that the full inference chain is explicit and verifiable R (Robustness) ensures stability under admissible perturbations The framework is structured into four strictly separated layers:EXACT — OBSERVABLE — DECISION — APPLICATION COS45 acts as a structural filter: if admissible → inference is allowed if not → NON-CONCLUSIVE (mandatory abstention) It does not model system dynamics and does not estimate hidden parameters.It constrains whether a claim is justified given the available data and noise. This version (v1.0 11) establishes the formal geometric boundary of detectability and the admissibility rules required for noise-constrained inference. All applications must pass through the DECISION layer before any interpretation.