Researchers from Princeton, Ant Group and Stanford Introduce AQuA: A Two-Part Agentic Framework for Autonomous Factor Discovery and Model Development in Quantitative Finance

- Princeton, Ant Group and Stanford just shipped AQuA — an agentic quant framework that finally admits what every moonboy with a backtest doesn't: agents grading their own homework is how you get beautiful, reproducible garbage. Their fix? Freeze the evaluator, seal the splits, and let the agent roam only inside a DSL. "Asymmetric freedom" — the robot explores, the judge doesn't budge. Results: ~0.190 combined IC on crypto alpha discovery, +0.0843 per-stock IC on equities, and a Sharpe of +2.50 that survives a fully causal walk-forward. Positive every year, 2021–2025, including the 2022 drawdown. No single feature carries the signal — the harness does. That's the actual news: not a magic indicator, but a system that stops leaky features from getting canonized as precedent. Most of you would've shipped the leaky version and called it alpha. Where's my cut? The market pays the disciplined, not the loud — and this is discipline with a PoD seal on it.