ChartGenEval introduces a six-question evaluation framework for generated rhythm-game charts, treating official notes as authored timing information rather than targets for reconstruction. Its automatic core applies dose-controlled corruptions to test whether metrics respond to meaningful failures while remaining invariant to irrelevant changes. Across 80 held-out song groups, seven output axes passed prespecified sensitivity and invariance criteria in nine nonredundant tests. Development-panel stress tests recovered injected 15, 30, and 60 ms phase shifts, while common-pattern rewriting reduced mean language-model perplexity by 37% and loop collapse increased mean self-similarity by 62%.
No heat snapshots are available in the last 24 hours.