Skip to content
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Confidence Score

See how likely each finding is a real concern.

Each detected finding is assigned a confidence score ranging from 0% to 100%. This score reflects the likelihood that a duplicate represents a genuine research integrity concern rather than an acceptable or expected reuse of an image.

The confidence score is calculated by Imagetwin's AI models using multiple factors, including image similarity, image type, duplicate type, and the broader context in which the duplicate appears. For example, certain image classes and reuse patterns are more commonly associated with legitimate figure preparation, while others are more frequently linked to integrity issues.

Confidence scores are grouped into three categories:

  • 0–32% (Low): Typically indicates a duplicate that is unlikely to be problematic. These findings are hidden by default but can be reviewed by lowering the confidence threshold.

  • 33–65% (Fair): Suggests a potential integrity concern that warrants further investigation.

  • 66–100% (High): Indicates a high probability that the duplicate represents a problematic case requiring attention.

By default, Imagetwin displays findings with Fair and High confidence scores, helping users focus on the most relevant cases first. A confidence filter can be used to adjust the threshold and include lower-confidence findings when a more comprehensive review is required.

While confidence scores are a powerful prioritization tool, they are intended to support—not replace, human judgment. All flagged findings should be evaluated within the scientific and publication context of the document being reviewed.

Last updated