Researchers have developed FIDES, a protocol to evaluate the reliability of trading strategies generated by large language models (LLMs). The protocol treats three artifacts - natural-language rationale, executable implementation, and track record - as separate views that need to be reconciled. In experiments with four LLMs and 40 trading strategies, the study found that concordance between these views does not predict profit, self-assessment is poorly calibrated, and switching models can significantly change results. The authors frame FIDES as a tool for measuring measurement fidelity, rather than making claims about market performance. --- Why it matters: This research matters because it highlights the limitations of LLM-generated trading strategies and the need for more robust evaluation methods. It has implications for investors and researchers who rely on these models to inform investment decisions, and can help improve the accuracy and reliability of AI-driven financial tools. Source: https://arxiv.org/abs/2608.23308