RESEARCH DASHBOARD
Is PAM-E better than PAM?
Run the study. Read the evidence.
The full experiment: tuning on practice cases, then every method on unseen cases, with paired statistics. It runs on this computer with simulated agents.
RUN THE STUDY
Run the offline study
Tunes PAM-E on the practice cases, then runs PAM, PAM-E and the controls on cases it has never seen, at several handoff sizes. Takes about five minutes. Offline and free: simulated agents, no API calls.
RESULTS
Study results
Simulated agents: these numbers check the method, not real AI models.Real-model results come from the OpenRouter study (configs/openrouter-study.yaml).