World-Building GPT: Trying to achieve response/lore consistency

I have been working on a pretty expansive world in a gpt. I currently have 13 files uploaded that cover various topics such as empires, characters, timelines, religion, etc. It generally works very well but it is often inaccurate on the first response.

To help with this I created a master response guide and instructed the gpt through the configuration instructions to follow this guide first. This guide specifically outlines which documents the gpt should reference and instructs it on how to formulate responses based on subject matter and to follow up with related questions.

Despite this it still generates responses that are incorrect. If pressed with a sort of “Are you sure that’s right?” or “that is wrong. check again” it will always then land on the right response.

Is there a way to make sure things are absolutely accurate. 100% of the time?

The process has been really interesting and fun but these consistent errors are a l little confusing.