Add iofold's evals ↔ optimization loop in 10 minutes. Code + LLM-as-judge evals, written by a deepagent with data science tools and backtesting. Meta-prompting for better instruction following, higher tool calling accuracy, and reduced hallucinations.