Run the same analysis across multiple models in parallel and compare results. Confidence is scored by the gateway. To try two wordings of one prompt, open Compare prompts tab.
Ok means that model wrote its own answer. Fallback means the first model could not answer, so a different model filled in. That fill-in is not a second opinion.
The confidence score says how much the answers agree with each other. It does not say the answer is right. High means at least two models agree. They can use different words in a summary and still count as agreeing. Medium means they mostly agree. Low means they disagree, or only one model was able to answer.