Does anyone else find Confluence’s "Remix" feature immediately drifts away from source data after the first iteration?
The initial output usually picks up the core context fine, but the moment you ask for a minor tweak or refinement (iteration 2+), it completely drops key facts, hallucinates metrics, or forgets constraints established in the original text. Has anyone figured out a prompting workaround to force strict context retention, or is this an inherent limitation of Rovo’s context windowing?
Product/Design owner of Remix, I would like to discuss on this!!
