Visual Fallacy Analysis Report

Visual Fallacy Analysis Report — CCA Detection Blind Spot

The Visual Fallacy is one of the most critical failure modes in AI-powered wood waste classification. This report documents instances where Chromated Copper Arsenate (CCA) treated wood presents visual characteristics that are indistinguishable from naturally weathered clean wood in standard RGB camera imagery — the blind spot that camera-based systems cannot reliably detect.

CCA-treated wood that has been exposed to weathering for 5+ years develops a characteristic gray patina that is visually identical to naturally aged untreated softwood. Standard computer vision models, including the Gemma 4 E2B classifier, cannot reliably differentiate these materials based on visual features alone. This is not a model accuracy problem — it is a fundamental limitation of RGB spectral data.

The report captures: visual evidence showing the ambiguous presentation, E2B and E4B confidence scores (typically in the 70-85% AMBER band for these cases), DeepREJECT anomaly score calculation, recommended escalation pathway (always to Hub XRF for definitive chemical analysis), and the 4-layer CCA detection protocol: Layer 1 (RGB visual AI), Layer 2 (E4B dual-model consensus), Layer 3 (Hub XRF atomic signature), and Layer 4 (NIR spectral confirmation).

The Visual Fallacy is why 905WOOD maintains the Precautionary Principle as a core operating principle: when in doubt, reject. A false positive (clean wood classified as treated) costs one unnecessary hub test. A false negative (treated wood classified as clean) contaminates the entire clean processing stream and exposes the operation to O.Reg 347 enforcement, RPRA penalties, and potential environmental liability.