The green-list rate and its interval
The headline figure is the proportion of scored word pairs that fell in the green list. Under no watermark, that proportion should sit near the key’s expected fraction, typically half.
The interval beside it is a Wilson score interval: the range of true rates consistent with what was observed, given how many pairs were scored. A short document produces a wide interval because a short document genuinely carries less information. Reporting the point estimate alone would hide exactly that.
z and p
The z score expresses how far the observed count sits from what chance would produce, in standard deviations. Roughly: 2 is unremarkable, 4 is notable, above 6 is very hard to explain by chance.
The p value converts that into a probability: how often chance alone would produce a green count at least this extreme. It is not the probability that AI wrote the document. That is a different quantity and this test does not compute it.
Why per-passage results are corrected
Each passage gets its own test, so a long document runs dozens at once and some will look significant by luck. WatermarkRemoverPro applies a Benjamini-Hochberg false-discovery-rate correction across all passages and reports how many were tested and how many survived.
A per-passage highlighter without that correction will confidently colour in sentences of any document you give it. If a tool shows you highlighted passages without saying how many tests it ran, that is the question to ask.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. WatermarkRemoverPro's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.