Cross-encoder loaded.
Running normalized evaluation on 10 queries...
Mean Recall@10 (Baseline - normalized): 0.1611
Mean Recall@10 (Improved - normalized): 0.2422
Wrote normalized submissions and src/eval_results.json