Error Attribution and Capability Boundaries in Object-Routed Power Equipment Defect Recognition with a Simulated PQC

Power equipment defect recognition is challenged by long-tailed labels, subtle state differences, scarce fault images, and variable acquisition conditions. Fixed-crop evaluation can obscure whether errors arise from localization, routing, crop quality, or state recognition. We developed an object-routed framework that links equipment localization to object-specific classifiers while retaining a classical decision path. A simulated low-qubit parameterized quantum circuit (PQC) was evaluated as an auxiliary representation for gated residual fusion. Matched analyses separated encoding, representation capacity, and fusion effects. Under parameter-matched branch substitution, the amplitude-encoded PQC achieved the highest descriptive means for accuracy (0.8870±0.0596), balanced accuracy (0.8648±0.0821), and macro-F1 (0.8539±0.1026). In a separate provenance-supported analysis, risk-aware residual fusion increased macro-F1 by 0.1584 relative to ordinary residual fusion (95% confidence interval, 0.0584–0.2641; p=0.0322). This inference supports the gating strategy rather than any specific residual representation. End-to-end experiments traced failures across the pipeline and identified errors that residual correction could not address. They were not designed to demonstrate industrial deployment benefit. The resulting contribution is an auditable fusion architecture with explicit capability boundaries. Its conclusions do not depend on PQC superiority.