Compare is Chat A and Chat B on the same collection. It is not a comparison of two catalogs.

Compare answers

Chat A and Chat B side by side on one collection
Compare mode puts two chats on the same collection. Each side has its own model, question, and Run. This is not a comparison of two catalogs.

Compare is Chat A and Chat B on the same collection. It is not a comparison of two catalogs.

Set the two sides

Choose Compare mode. Chat A and Chat B each have a question and a policy. Copy Chat A settings to Chat B when you want the same start. Then change one setting, or change the question on purpose. If you leave Chat B’s question empty, it reuses Chat A’s question.

Run

Run, stop, or retry each side on its own. One side can finish while the other fails. A prior-result mark means that side is showing an earlier run, not a new one.

How to read the difference

Highlighted words are text that differs. They are not a difference in evidence, and they are not an Eval Studio score. Evidence differences are which passages each side cited. If a side has no citation, that side did not ground its answer. Do not treat the prettier paragraph as the better one.

Related documentation