Reproducibility Audit Finds LLM Trading Research Lacks Execution Realism

A review of 30 LLM-based trading studies finds that architecture details are consistently reported but evaluation assumptions — point-in-time controls, transaction costs, turnover, execution timing — are frequently missing or inconsistent, making performance claims economically uninterpretable. The paper provides a coded evidence matrix and a 10-equity worked example as a methodological scaffold.