Skip to content

Clarify benchmark task definition and labeling rubric in README #7

@rafaelsideguide

Description

@rafaelsideguide

Several reviewers assumed truth_text represents full-page content. In reality, contractors labeled snippets from rendered pages: ~100 words of important_content and ~10 words of not_important_content. This needs to be explicit in the README to avoid misinterpretation of drift and scoring.

TODO:

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions