Litcius/Paper detail

COS45 — A Geometric Detectability Framework for Noise-Constrained Inference (v1.0 11)

Louis Morissette

2026Zenodo (CERN European Organization for Nuclear Research)6 citationsDOIOpen Access PDF

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.

Topics & Concepts

InferenceVerifiable secret sharingObservableMathematicsStability (learning theory)Boundary (topology)AlgorithmNoise (video)Computer scienceSequence (biology)Decision boundaryResolution (logic)Rule of inferenceTheoretical computer scienceStatistical inferenceDecision theoryFiducial inferenceImage (mathematics)Discrete mathematicsApplied mathematicsDiscrete time and continuous timeArtificial intelligenceControl Systems and IdentificationBayesian Modeling and Causal InferenceProbabilistic and Robust Engineering Design