The Stanford Computational Antitrust Project announces the publication of “Ulysses: A Case Outcome Predictor for Computational Antitrust” by Piero Alexis Malca Vilchez, César Humberto Quiñones Costa, and Enzo Rodrigo Gomez Rojas. The article appears in Volume 6 of Stanford Computational Antitrust (pp. 140–166).

Can a large language model predict whether a business practice will be found anticompetitive? The authors build a tool that answers the question for one jurisdiction. They take Peruvian competition law as their case study and construct a case outcome predictor. The design is meant to be replicated. An enforcer or a practitioner elsewhere can follow the same framework and adapt it to a given practice under a given legal regime.

The method matters as much as the result. The authors do not retrain a model. They guide it with legal expertise through prompting and retrieval-augmented generation. The system retrieves past case law and structures the legal reasoning around it. This keeps current assessments closer to prior administrative precedent and improves consistency across decisions. It also lowers the barrier to building such a tool, since no model retraining is required.

The authors are careful about scope. Their results are preliminary. The benchmark is a small proof of concept rather than a full validation. They present Ulysses as a demonstration of what domain-adapted legal workflows can do for specialized legal judgment prediction, not as a finished enforcement instrument.

Thibault Schrepel, founder of the Stanford Computational Antitrust Project, stated:

“This paper makes a useful point. You do not need to retrain a model to build a working legal prediction tool. You need legal experts who know the case law and a careful retrieval design. The authors prove it on Peruvian competition law, and they hand others a framework to follow. That is how computational antitrust reaches jurisdictions without large engineering teams. I value the honesty about the limits as much as the result.”

The article is available for download on the Stanford Computational Antitrust project’s page.