dashboard
Prediction calibration
How well does the LLM's stated conviction track reality? Target is >55% overall hit rate (PRD §11), with each conviction bucket landing near its stated level — e.g. 70% conviction should hit ~70% of the time.
How well does the LLM's stated conviction track reality? Target is >55% overall hit rate (PRD §11), with each conviction bucket landing near its stated level — e.g. 70% conviction should hit ~70% of the time.