Chat logs need to render the same as the original chat.

When inspecting chat histories, much formatting is lost. This makes it impossible to QA end-users’ experiences. Importantly for us, we use models that properly render Maths notation instead of showing the underlying LaTex (which is cumbersome to read). HOWEVER, the chat logs don’t do proper Maths notation. We can’t read the chats properly, and we can’t assess what the original chats actually looked like.