compare-attempts

What problem does it solve?

This Skill consolidates evaluation reports from brazil-bench attempts to produce a ranked leaderboard and a detailed comparison summary, enabling rapid assessment of performance across attempts.

Core Features & Use Cases

• Leaderboard generation: Build a scoring-based ranking from canonical evaluation data.
• Detailed comparison: Produce per-attempt metrics, trend insights, and patterns across attempts.
• Automated validation: Enforces canonical 16-requirement denominators and normalization where needed.

Quick Start

Run the comparison workflow after generating evaluation reports. Example: set results_dir to the folder containing markdown reports, then execute the comparison to output a Markdown leaderboard to the specified output_file.