Walk-Verify: Human-AI Collaboration Builds a More Beautiful World

An exploratory case study (N=1) investigating five hard boundaries in human-AI collaboration — Sequence Inertia, Self-Consistency Threshold Drift, Single-Subject Preference, Direction Can Be Transmitted But Direct Calibration Is Irreplaceable, and Structural Asymmetry of the Correction System. [Revised version 2026-06-17: Added quantitative data tables (2a-2c), expanded §7 Protocol with evidence basis, expanded §8 External Verification with independent audit results.]