metadata
pretty_name: GitGoodBench parity experiments (Harbor adapter)
viewer: false
tags:
- harbor
- parity-experiments
- gitgoodbench
GitGoodBench adapter: parity experiment artifacts
Parity and oracle results of the Harbor adapter for GitGoodBench (merge conflict resolution, GitGoodBench Lite, 60 tasks).
These files belong in harborframework/parity-experiments, in PR 281. That repository currently refuses uploads because it is over the Hub's limit of 1,000,000 files per repository, so the artifacts are published here in the same layout until they can be moved.
- Results, directory structure and notes:
adapters/gitgoodbench/README.md - Adapter: https://github.com/harbor-framework/adapters/pull/78
- Dataset: https://github.com/harbor-framework/harbor-datasets/pull/277
| Side | Run 1 | Run 2 | Run 3 | Mean ± sample SEM |
|---|---|---|---|---|
| Original (reference harness) | 36.67 | 41.67 | 43.33 | 40.56 ± 2.00 |
| Harbor (Daytona) | 40.00 | 43.33 | 41.67 | 41.67 ± 0.96 |
Solve rate in %, mini-swe-agent 2.4.6 with deepseek-v4.1-flash on both sides, 60 tasks per run. Oracle: 60/60 on both sides.
The trajectories quote source code from the repositories of the benchmark, which GitGoodBench selected for their permissive licenses.