ground-truth

What problem does it solve?

This Skill ensures that evaluation datasets have accurate, consistent, and defensible ground truth labels by deriving them from authoritative guidelines rather than intuition.

Core Features & Use Cases

• Guideline-Driven Labeling: Enforces labeling based on explicit rules, not subjective judgment.
• Reproducible Reasoning: Documents the exact guideline provisions and rationale used for each label.
• Ambiguity Management: Provides a structured way to identify and document unclear or conflicting guidelines.
• Use Case: When training a content moderation AI, use this Skill to ensure all flagged content is labeled according to a predefined policy, minimizing bias and improving model fairness.

Quick Start

Use the ground-truth skill to establish labels for the dataset in the 'evaluation-data/' directory.