Skip to content

Improve evaluation failure reporting #5

Description

@SinghCod3r

Make evaluation results more useful for understanding why a skill succeeds or fails.

Expose useful information such as:

  • task ID
  • task description
  • baseline result
  • skill-enabled result
  • judge result
  • failed dimensions
  • score difference
  • judge type
  • diagnostic feedback

A developer should be able to understand failures without reading application logs.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    evaluationEvaluation, benchmarking, scoring and judgeshelp wantedExtra attention is needed

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions