evaluation-anchor-checker

What problem does it solve?

This Skill addresses the common issue in technical surveys where strong numeric or performance claims are made without sufficient context, making them difficult for reviewers to verify and potentially leading to rejection.

Core Features & Use Cases

• Numeric Claim Auditing: Reviews sentences containing numbers to ensure they are accompanied by minimal, interpretable evaluation context (task, metric, constraint).
• Contextual Downgrading: Weakens or removes numeric claims if the necessary context is missing, preventing guesswork and maintaining evidence-boundedness.
• Ambiguous Naming Check: Flags and corrects underspecified model names that appear hallucinated.
• Use Case: Before submitting a research paper, use this Skill to automatically scan all claims like "Model X achieved 90% accuracy" and ensure they are rewritten to include the specific task and metric, e.g., "On the ImageNet classification task, Model X achieved ~90% top-1 accuracy".

Quick Start

Use the evaluation anchor checker skill to audit and rewrite numeric claims in the provided workspace.