Stanford University
CoDICoordinated Decision
Intelligence Center

Projects

Coordination in practice.

Publicly documented work across healthcare, markets, education, optimization, and scientific understanding.

Financial markets

The AI monoculture paradox in financial markets

When many traders adopt the same AI forecast, their individual choices can crowd the same trades and reduce the forecast’s value to its adopters. Monoculture Lab combines a market model with an interactive simulation platform to explore how shared information, communication, and feedback shape adoption and trading outcomes.

↗

Financial markets

Settlement manipulation in prediction markets

How do settlement rules shape incentives across connected markets? This research combines a model and empirical analysis to study prediction-market contracts whose settlement prices can be influenced by trading the underlying asset, and examines the implications for contract design and price discovery.

↗

Platforms

Policy gradient estimation for platform experiments

In dynamic systems, a decision changes both current outcomes and the conditions for future decisions. The truncated policy gradient estimator uses short outcome trajectories from randomized experiments to estimate treatment effects under nonstationarity, with theoretical analysis and real-world case studies.

↗

Causal inference

CausalSmith: AI-assisted causal research

CausalSmith coordinates AI agents that propose theoretical results in causal inference, formalize statements, and construct proofs checked by Lean. A statement audit compares formal theorems with their intended informal claims, while human inspection remains part of the research process.

↗

Optimization

FLARE: verifying optimization models

FLARE combines language models with formal theorem proving to check whether reformulating a mixed-integer linear program preserves the underlying optimization problem. Its Lean-checked proofs support verification of proposed reformulations, evaluated using the FormulationBench benchmark.

↗

Scientific understanding

Readable Intelligence

Readable Intelligence is a research-reading prototype that starts with mathematics. Readable versions sit alongside original papers, with paper maps, notation tables, and a reading agent that links answers to source material. The platform helps readers connect explanations to the research they are trying to understand.

↗

Education & entrepreneurship

Proxy-Expert Practice: An Agentic AI System for Teaching Entrepreneurial Problem Validation

How can students learn to determine whether a problem is real, important, and worth solving? Proxy-Expert Practice builds AI proxies of specific, consenting Stanford Distinguished Careers Institute fellows, grounded in multiple discovery interviews. Students practice entrepreneurial problem validation with these proxies before meeting the real fellows. The team is testing whether this practice improves live conversations through a randomized study in E145 and MS&E 272. PhD students Yikai Cao and Xilan Zhang develop the project with Chuck Eesley and Itai Ashlagi, with funding from the Stanford Accelerator for Learning.

↗

Education & public systems

Optimization meets participation

This work brings optimization and stakeholder feedback into an iterative school-zone design process. Interactive tools help participants explore candidate zones and tradeoffs, including diversity and travel distance, so that mathematical models can inform practical assignment decisions.

↗

Connect with CoDI

Put coordination
at the center.

Explore the research, meet our faculty, and find the right way to engage.

Explore ways to engage