Skip to content

Relevance & review

Relevance scores help prioritize the code previews returned by Search. Use the model selector to choose the available scoring model, then sort the relevance column to inspect higher-scoring previews first.

Scores range from 0 to 100 and describe the supplied preview and its context. They are ranking signals, not calibrated probabilities that a credential is valid. Scoring does not verify a credential with its provider or establish ownership.

The ranking can account for repository and file context and signs of example or noisy content. A token-like substring embedded inside a larger high-entropy string can be ranked lower than a standalone occurrence.

Open a match’s code details and inspect the surrounding preview. Choose Relevant, False positive, or Skip as appropriate. Use Relevant for evidence useful to your investigation, False positive for an irrelevant match, and Skip when you cannot make a confident decision.

Feedback and saved evidence serve different purposes. Use Save to Results when you want to retain a match in your workspace.

Scores arrive asynchronously after previews are received. Available model versions and feedback can affect ordering. A score applies to the preview the model saw; it is not an assessment of the full repository.

If a score is unavailable, continue reviewing the source context. An unavailable score does not mean the match has zero relevance.