AQuA：防数据泄漏的自改进量化交易智能体
A bad experiment is bad enough. A self-improving agent can keep building on it.
做自主研究智能体的同学必看，这篇给出了防止智能体递归放大错误的架构设计，直接可借鉴到你的实验流程里。
A bad experiment is bad enough. A self-improving agent can keep building on it.
糟糕的实验本身已够糟了。一个自我改进的智能体还能在此基础上不断叠加问题。
New Stanford, Princeton, Ant group paper.
斯坦福、普林斯顿和蚂蚁集团的新论文。
AQuA targets that failure mode: if an agent writes a leaky experiment, gets a great score, and stores it as precedent, recursion can amplify the bug.
AQuA针对这一失败模式：如果智能体编写了一个有漏洞的实验，获得高分并将其作为先例存储，递归可能会放大这个缺陷。
The authors saw this in an earlier version, where an LLM-written volume feature quietly used end-of-day information and a second LLM reviewer still approved it.
作者在早期版本中看到了这一点，当时一个由LLM编写的成交量特征悄悄使用了收盘后信息，而第二个LLM审查者仍然批准了它。
Their fix is architectural.
他们的解决方案是架构性的。
Agents can propose factors or model changes only through restricted specifications; the data path, labels, splits, and evaluator stay sealed outside their reach.
智能体只能通过受限的规范提出因子或模型变更；数据路径、标签、分割和评估器都密封在其触及范围之外。
Search sees validation scores, while the final test window stays untouched until the configuration is frozen.
搜索过程能看到验证分数，而最终测试窗口在配置冻结前保持不被触碰。
On US equities, the hybrid model reaches +0.0843 IC versus +0.0613 for the strongest baseline, and its long/short strategy reaches +2.50 Sharpe at 2 bps, with about +2.0 under fully causal walk-forward testing.
在美国股票上，混合模型达到了+0.0843的IC，而最强基线的IC为+0.0613；其多空策略在2个基点成本下达到+2.50的夏普比率，在完全因果的滚动前向测试下约为+2.0。
These are simulated, not live-trading results.
这些是模拟结果，而非实盘交易结果。
For autonomous research agents, the design rule is useful: let the agent improve the experiment without letting it redefine valid evidence.
对于自主研究智能体，设计规则很有用：让智能体改进实验，但不允许其重新定义有效证据。
– arxiv. org/abs/2608.12841
– arxiv.org/abs/2608.12841
Title: "AQuA: Recursively Self-Improving Quantitative Trading Research Agents"
标题：“AQuA：递归自我改进的量化交易研究智能体”
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力