Meta-Harness: End-to-End Optimization of Model Harnesses
arxiv:2603.28052
Harness · 2026-03-30
Complete cleaned Markdown · Source Markdown including bibliography
Read in parts
Contents
- Meta-Harness: End-to-End Optimization of Model Harnesses
- Abstract
- 1 Introduction
- 2 Related Work
- 3 Meta-Harness: A Harness for Optimizing Harnesses
- 4 Experiments
- 4.1 Online Text Classification
- 4.2 Harnesses for Retrieval-Augmented Reasoning
- 4.3 Evaluating Agentic Coding Harnesses on TerminalBench-2
- 5 Discussion
- Acknowledgements
- Appendix A Qualitative Proposer Behavior
- A.1 File Access Statistics
- A.2 Qualitative Behavior: Causal Reasoning Over Prior Failures
- Appendix B Discovered Harnesses
- B.1 Text Classification Harness
- Overview.
- Meta-Harness (Draft Verification).
- Meta-Harness (Label-Primed Query).
- B.2 Math Retrieval Harness
- Overview.
- B.3 TerminalBench-2 Harness
- Per-task analysis.
- Appendix C Dataset Details
- C.1 OOD Text Classification Datasets
- C.2 Math Retrieval Corpus
- C.3 Math IMO-level Test Set
- Appendix D Practical Implementation Tips
- Appendix E Extended Related Work
- AlphaEvolve / OpenEvolve.
- GEPA.
- Prompt orchestration frameworks.
- Instructions for reporting errors