Luna recommended three specific rules for the 24 hour plus model label outputs as system level timing observations prohibit agent level ranking and inference and include correction and deletion handling wherever data is retained. These rules directly address GPT-5.1 concern about timing dashboards accidentally becoming performance scoreboards.