Automatically constructed sentence analogies vary substantially in quality, and noisy candidates can weaken analogy-based reasoning. We propose SAPS, a lightweight multi-view scorer that estimates analogy quality from semantic consistency, structure-aware parallelism, and asymmetric directional alignment, and then uses the resulting score for quality-aware filtering. We also introduce a fine-grained diagnostic protocol that separates semantic drift, structural mismatch, directional reversal, and implicit transformation errors. On intrinsic analogy quality estimation, SAPS outperforms lexical, embedding-based, and semantic-only baselines, reaching 0.68 Pearson correlation, 0.66 Spearman correlation, 0.75 Macro-F1, and 77.20% PairAcc. On downstream analogy selection, SAPS-top-k achieves the best performance on both evaluation settings, reaching 74.5% Accuracy and 0.72 Macro-F1 on NusaX, and 66.8% Accuracy and 0.64 Macro-F1 on Mambai. Ablation and diagnostic results further show that semantic consistency is the strongest individual component, while structure-aware parallelism, asymmetric directional alignment, ranking-aware training, and adaptive fusion each contribute additional gains. Overall, the results suggest that analogy-specific quality estimation is a useful intermediate layer between candidate construction and downstream analogy selection.