This tracks. Text evals to test core logic and voice evals for overall end-to-end performance!