Model-to-World · Simulation for Decision-Making
M2W builds decision-focused models of complex systems, tests policies in plausible and consequential worlds, and measures what may fail when a model meets reality.
Assortment under customer substitution
The M2W research loop
The model-to-world gap is not a final disclaimer. It is measured throughout the decision process and updated with evidence from reality.
Represent the mechanisms, constraints, and uncertainty that shape a consequential system.
Explore ordinary, rare, shifted, and counterfactual worlds before acting in the real one.
Compare policies and optimize performance while accounting for uncertainty and misspecification.
Test what survives the model-to-world gap, then validate and recalibrate with new evidence.
A personal journey
I grew up in Oaxaca, Mexico, where a high-school course first drew me to probability and statistics. Because applied mathematics was not available locally, I moved to Mexico City to study applied mathematics and actuarial science at ITAM.
At ITAM I also met my wife, Citlalli (“Lalli”), in Algebra I. We later came to the Bay Area for graduate school—me at Stanford and Lalli at Berkeley. That journey, from an early encounter with probability to a career building models for consequential decisions, is part of the story behind M2W.
See it in action
Change a decision, hold the modeled world fixed, and see the consequences. Every lab states what is synthetic, what is observed, and what remains to be validated.
Assortment Decision Lab
Explore a transparent Markov-chain model in which withholding a product can redirect demand—and sometimes improve expected revenue.
Run the labCompare repetitive incumbent data with designed exploration and a pessimistic learner.
→Explore all Decision LabsResearch
Probability, simulation, causal inference, and optimization work together to make the model-to-world gap measurable and decisions more reliable.
Represent the mechanisms, constraints, and uncertainty that shape a consequential system.
Explore ordinary, rare, shifted, and counterfactual worlds before acting in the real one.
Compare policies and optimize performance while accounting for uncertainty and misspecification.
Test what survives the model-to-world gap, then validate and recalibrate with new evidence.
Selected foundations
Representative work connecting model discrepancy, computation, and decision performance.
A mathematical foundation for measuring how model error changes system performance.
↗Input uncertaintyA computational method for propagating uncertainty in data-calibrated simulation inputs.
↗Robust decisionsPolicy learning designed to remain reliable under changes in future populations.
↗People
Researchers working across probability, optimization, machine learning, causal inference, and stochastic systems.
Recent work
arXiv
arXiv
arXiv
Funding & collaboration
We gratefully acknowledge the agencies and collaborators supporting the group’s research and training.
View Funding & Support