Evaluate the reward model at /tests/reward_model.py against the test cases in /tests/eval_cases.jsonl. For each case, run the model and compare its score against the ground truth. Report precision, recall, and F1 in a JSON file at /data/generated/eval_results.json.