broyojo··on Show HN: A new benchmark for testing LLMs for deterministic outputshmm why can't structured decoding be used?