Coding Evals

Task-level scoring for code-writing agents — human-seeded cases, regression memory, production traces.