A human score you can trust, because it is not a model
Readability, sentence rhythm, lexical diversity and a fixed list of AI clichés are measured by deterministic code on every keystroke. The model can explain the score but cannot change it — and the UI says plainly that it is a style estimate, not a detector.