From Stochastic Answers to Verifiable Reasoning: Interpretable Decision-Making with LLM-Generated Code
This addresses the need for verifiable and interpretable decision-making in high-stakes domains like venture capital, though it is an incremental improvement over existing LLM-based rule systems.
The paper tackles the problem of reconciling scalability, interpretability, and reproducibility in LLM-based high-stakes decision-making by reframing LLMs as code generators that produce executable decision logic, applied to venture capital founder screening. It achieves 37.5% precision and an F0.5 score of 25.0% on VCBench, outperforming GPT-4o while maintaining full interpretability.
Large language models (LLMs) are increasingly used for high-stakes decision-making, yet existing approaches struggle to reconcile scalability, interpretability, and reproducibility. Black-box models obscure their reasoning, while recent LLM-based rule systems rely on per-sample evaluation, causing costs to scale with dataset size and introducing stochastic, hallucination-prone outputs. We propose reframing LLMs as code generators rather than per-instance evaluators. A single LLM call generates executable, human-readable decision logic that runs deterministically over structured data, eliminating per-sample LLM queries while enabling reproducible and auditable predictions. We combine code generation with automated statistical validation using precision lift, binomial significance testing, and coverage filtering, and apply cluster-based gap analysis to iteratively refine decision logic without human annotation. We instantiate this framework in venture capital founder screening, a rare-event prediction task with strong interpretability requirements. On VCBench, a benchmark of 4,500 founders with a 9% base success rate, our approach achieves 37.5% precision and an F0.5 score of 25.0%, outperforming GPT-4o (at 30.0% precision and an F0.5 score of 25.7%) while maintaining full interpretability. Each prediction traces to executable rules over human-readable attributes, demonstrating verifiable and interpretable LLM-based decision-making in practice.