Case Studies

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.

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.