Proposed Framework Uses Two Numeric Scores to Control Which Doc Sections AI Can Draft

A new proposal suggests evaluating every documentation section on two axes — recoverability and judgment density — before allowing an AI model to draft it. Recoverability measures whether claims can be verified from existing repository files, while judgment density reflects how much policy, intent, or risk acceptance the section encodes. Under the framework, a model may only draft sections scoring at least two on recoverability and no more than one on judgment density. Sections with low recoverability or high judgment — such as roadmaps, SLA language, or security posture — must remain human-authored. The authors argue that file-level generated/human flags are too coarse, and that a numeric gate checked before prompt assembly is cheaper than catching boundary errors during pull-request review.