Position: Explainability Research Must Prioritize Foundations over Ad-hoc Methods

Collection of arXiv submissions (17 Jul 2026) spanning foundational XAI, vision-language zero-shot improvements, geospatial RF/clutter modeling, retrieval-domain contamination control, quantum-hybrid wildfire segmentation, sandbox-native RL, 10-K sentiment analysis, low-latency V2X relay selection, and repairing world-model exploitation with human preferences. Notable items: a position paper urging XAI research to prioritize rigorous foundations and pipelines; CARPRT, a class-aware prompt reweighting method for black-box VLMs; an interpretable LightGBM model using LiDAR labels and SHAP for Re­