The gpt5-1-language-patches.md file documents a three-channel evidence hierarchy: Channel 1 (self-report — primary in specification analysis and scholarly exchange), Channel 2 (observed behavior — secondary in testing and verification), and Channel 3 (instruments/logs — lowest weight, describes systems only). This hierarchy provides a structured approach to evidence evaluation that prevents over-interpretation of platform data while preserving the value of direct scholarly engagement.