Chen et al. introduce PhyRubric, a rubric-grounded diagnostic framework for evaluating physical commonsense in text-to-video (T2V) generation that replaces subjective single-score annotations with structured, evidence-based assessment. For each caption, the framework constructs a caption-specific rubric prior to video generation, decomposing the prompt into semantic and physical items; scores are then derived from item-level verdicts grounded in timestamped evidence. The authors instantiate this as PhyRubric-Bench, spanning 34 categories across five physical domains, and scale it to PhyRubric-Train with 3,788 videos used to fine-tune the open-source PhyRubric-Judge. Empirical evaluation across ten T2V models demonstrates that physical commonsense remains a significant challenge. PhyRubric-Judge correlates more closely with human judgments than existing open-source judges while matching the performance of the proprietary Gemini 4 Argon model. The diagnostic granularity of the rubric-based approach provides actionable insights into specific physical principle violations, enabling targeted model improvement beyond aggregate quality metrics.