Raw LLM Responses

Closing the loop with the "surfaced\_at" flag is a huge win for immersion. In 2026, the biggest immersion-killer isn't bad prose, it's an AI that "remembers" too much junk. By letting unused offscreen events decay, Musona is actually mimicking human-like selective memory. The "assertion-grounding" issue you're facing is basically the final boss of AI roleplay—finding that line where the character can disagree with a user's "fake" memories without breaking the fourth wall. It’ll be interesting to see if a dedicated "World Fact" vector database can help the model stay grounded against user gaslighting.