WatermarkResult
Models & Types · AiDotNet.Safety.Watermarking
Detailed result from watermark detection across any modality.
For Beginners
a small p-value (say below 0.0001) means the content almost certainly carries the watermark of the key used for detection; a p-value near 0.5 means there is no evidence either way.
How It Works
A keyed detector reports the test statistic behind its confidence: ZScore is approximately standard
normal for content that does not carry the detector’s watermark, and PValue is the probability of a
z at least this large by chance. Confidence maps z to [0, 1] as described in
Confidence.
Constructors
| Constructor | Summary |
|---|---|
WatermarkResult() |
Properties
| Property | Summary |
|---|---|
bool WatermarkDetected { get; set; } |
Whether a watermark was detected. |
double Confidence { get; set; } |
Detection confidence (0.0 = no watermark, 1.0 = certain). |
double PValue { get; set; } |
The one-sided p-value of ZScore: the chance of so much evidence without the watermark. |
double ZScore { get; set; } |
The detection z-score; approximately standard normal for content without this key’s watermark. |
int ScoredUnits { get; set; } |
The number of units the test scored (tokens, substitution sites or transform coefficients). |
string Modality { get; set; } |
The modality of the content checked (text, image, audio). |
string WatermarkType { get; set; } |
The detected watermark type, if identifiable. |