Examples

Real-world applications of PrefGraph.

Detecting Inconsistency in AI Agents

Do LLMs keep a stable action ranking across menus? We build preference graphs from model choices and test for cycles (SARP, IIA), then quantify minimal edits (HM) to restore consistency.

Descriptive Study of Classifieds Choice

How consistent are 70,000 users on Norway’s largest classifieds platform? We map 1.3M unique listings to 290 category-geography groups, run SARP, HM, and RUM tests, and compare search versus recommendation behavior.

Predicting Customer Spend & Engagement

Do RP features improve predictive models? We benchmark GARP, CCEI, MPI, HM, and VEI features against spend/engagement baselines on churn, high-spender, engagement, and LTV tasks across 10+ datasets.

Performance Benchmarks

Throughput and scaling metrics for the Rust engine across dataset sizes, user counts, and choice dimensions.