Abductive reasoning—the capacity to infer the most plausible explanation from incomplete or noisy observations—remains a significant hurdle for language models that often rely on simple associative patterns. In this paper, we present our framework for the SemEval 2026 Task 12. We propose an Option-Aware Retrieval and Cross-Encoder Reasoning Framework designed to bridge the gap between evidence acquisition and causal inference. Our architecture utilizes a dual-track retrieval strategy that gathers both global background and option-specific evidence, ensuring high recall of decisive clues. This is coupled with a 4-pass independent cross-encoder validation using DeBERTa-v3-large, which isolates individual hypotheses to prevent attention dispersion. Experimental results on the official dataset show that our system achieves a robust test score of 0.8358.