Eyes Wide Sharp: A Biaxial Fresnel Framework for Passive Polarimetric Preprocessing with Applications to Sovereign Physical AI Coordination

A passive optical preprocessor based on biaxial Fresnel walk-off is developed as a substrate for material-aware perception in heterogeneous physical multi-agent systems. We complete the derivation of the walk-off displacement d from the indicatrix-form Fresnel equation without uniaxial approximation, with full Poynting-vector geometry; the paraxial small-angle form d=ttan is used in numerical evaluation with documented 0.7% deviation from the exact oblique-incidence form (§2.2). We scale the supporting Monte Carlo from N=5104 to N=5106 rays under bootstrap CI 95%, and sweep a four-dimensional parameter envelope (CT,,T,material) across eight biaxial crystals. The thermal sensitivity =d/T — typically treated as a tolerance budget — is reframed as a passive information channel through the operator O:d⇌T (with T=d/|T0); we designate this subsystem OSARIS and characterize its thermal resolution for the eight-material library. The framework is then specified as the perceptual layer of a sovereign coordination ecosystem (ROCH3): not only as agent-borne preprocessing (extending the multi-agent claims of the second provisional application), but as a candidate for distributed perceptual infrastructure in which fixed installations contribute polarimetric and OSARIS-thermal context to a shared minimum viable reality (MVR) [Roche 2026b §3.2] under the conservative composition operator [Roche 2026b §3.3]. A companion software demonstrator (software/odin/) provides reference implementations of the walk-off, OSARIS, and distributed-perception components. The contributions are documentary and architectural: every numerical result is traceable to an executable function with no proprietary dependencies, and every claim that exceeds first-principles physics is flagged with falsification criteria.