How would you design an evaluation harness to measure the functional correctness of generated code across multiple programming languages?
