Why this verdict is uncertain — and why that's the honest answer
AI detectors fail constantly, in both directions. The published research is consistent on this:
False positives on human writing hit non-native English speakers, neurodivergent writers, and anyone using formal or academic registers the hardest. Several studies have found misclassification rates above 50% for ESL student essays. False negatives are just as common — light edits to AI text, or a different model than the one in the training set, easily evade detection.
The signals used here (burstiness, phrase fingerprints, punctuation patterns) are statistical tendencies, not proof. A single essay or paragraph cannot be reliably attributed to a human or a machine from text alone. Anything that claims otherwise is selling you confidence it hasn't earned.