Benchmark

My first model router let an LLM read a prompt file and pick an agent. This one asks a “System One” decision model four typed questions, scores the models with plain arithmetic, and publishes a benchmark to prove it works (and to show where it doesn’t).