Ranktration

Rank/compare algorithms, models, or approaches with weighted multi-criteria analysis.

How It Works

1. Collect trajectories - Gather approaches with measurable metrics
2. Smart sampling - Select representative sample for large datasets
3. Pairwise battles - Compare sample trajectories using weighted scores
4. Tournament ranking - Establish global rankings through competitive analysis
5. Statistical confidence - Measure ranking stability and significance
6. Final scoring - Apply ranking bonuses to create comprehensive evaluation

License

MIT License - see LICENSE file for details.

Credit

This implementation is inspired by and derived from the RULER (Robust Unified Learning Evaluation & Ranking) framework originally developed by OpenPipe for AI evaluation and trajectory analysis in machine learning.

Contact

For questions, issues, or contributions:

• GitHub: https://github.com/V-Sekai-fire/ranktration
• Hex.pm: https://hex.pm/packages/ranktration