Which coding-agent harness should you use for your model?
Fast System 1 decisions with calibrated probabilities