๐Ÿ† Leaderboard

How well do methods catch procedural hallucination, when an LLM agent stays fluent and factually correct but stops following its workflow?

๐Ÿ“„ Paper ยท ๐Ÿค— Dataset ยท ๐Ÿ’ป Code ยท โž• Add your method

๐Ÿ“„ from the DialFork paper ยท ๐ŸŒ community submission ยท click a column to sort

Built something that beats these? Or that fails in an interesting way?

Score it with evaluate.py and open a pull request on GitHub: CI re-scores your prediction file, and your row shows up here once it is merged. Negative results are welcome too.