No authoritative watermark was found, but multiple calibrated detectors lean AI. Treat this as a lead for review, not proof.
Absence ≠ human. Watermarks are silent on non-watermarked models, short samples, and rewritten text, absence here does not indicate human authorship.
Due process: decision-support only, not a finding of misconduct. Seek the author’s response, drafts, and version history before any adverse action.
▲ If a Tier-A watermark were present, this verdict would read "Provenance confirmed."
| Detector | Raw P(ai) | Measured FPR | Calibrated weight | Read |
|---|---|---|---|---|
| Commercial detector A BYO | 0.92 | 4.8% | 0.31 | High sensitivity, high false-positive rate, down-weighted. |
| Commercial detector B BYO | 0.88 | 2.1% | 0.44 | Lower FP; corroborates, still not proof. |
| Provenote stylometry Inkprint | 0.61 | , | 0.35 | Reads unlike this author’s known Print. |
That same detector call in isolation returns "0.92, likely AI." Provenote shows that at its measured 4.8% false-positive rate, that score is worth a calibrated weight of 0.31, a lead for review, never an accusation. That gap is the product.