Weathered arsenic-treated (CCA) wood looks identical to clean gray pine. The Visual Fallacy is the fundamental optical blind spot that makes visual-only wood waste classification unreliable. The asymmetric risk: False positive (clean wood classified as contaminated) costs $55/tonne in unnecessary CIRCIL Levy. False negative (CCA wood classified as clean) costs $25,000–$100,000 per O.Reg 347 violation, plus RPRA enforcement, insurance spikes, and reputational destruction. The Override Hierarchy: Audio overrides Visual (keywords like 'deck' or 'pressure treated' shatter AI classification). Origin overrides Visual (demolition source = hard-flag). Hub overrides Edge (XRF/NIR sensors deliver the deterministic verdict). Dual-Model Consensus: When E2B confidence falls below 85%, the E4B Visual Fallacy guard activates. Both models must agree. DeepREJECT scores anomalies: R × W × (0.65·K + 0.35·F). TCLP Auto-Reject: Arsenic >2.5 mg/L, Lead >5.0 mg/L, Chromium >5.0 mg/L = automatic hazardous waste manifesting via RPRA HWPR API. No human decision required.