Princeton, Ant Group, and Stanford Propose AQuA Framework for Autonomous Quant Research

Quantitative research agents that write their own experiments can corrupt the evidence they later learn from. [1] A leaky feature that scores well gets stored as a successful precedent and propagated through later iterations. [1] Prompt-level instructions and reviewer agents do not close this, because author and reviewer share the same blind spots. [1] A team of researchers from Princeton University, Ant Group and Stanford University propose AQuA to address these vulnerabilities. [1]

AQuA is a pair of language-model-driven research systems that improve their own research process across iterations while the thing judging them stays frozen. [1] One discovers symbolic alpha factors on crypto, and the other develops time-series models on US equities. [1] They share no agents, memories, candidate spaces or research state. [1] Each part fixes its splits, feature and label definitions and evaluator before any iteration starts, and the agent emits only a constrained factor expression or a single config diff. [1]

By separating the iteration process from a frozen judging mechanism, AQuA attempts to solve the recursive amplification of undetected bugs in automated quantitative research. The architectural choice to physically restrict the agent to emitting a constrained factor expression or a single config diff highlights a shift away from pure prompt-based guardrails toward structural system constraints. What the sources don't address: How the frozen judging mechanism is initially calibrated or whether its static nature limits the discovery of novel factor classes over extended market cycles.