SWE-Explore

SWE-Explore isolates repository exploration from patch generation, scoring the ranked code regions an agent returns for a GitHub issue under a fixed line budget.

Also known as: SWE-Explore-Bench

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorycoding
Subcategoryrepository exploration / code localization
Page statusactive
Metricline-level F1 / nDCG@B (with HitFile, HitRegion, context-efficiency and noise-rate as secondary metrics)
Directionhigher_is_better
Dataset size848
Dataset licenceCC BY-NC-ND 4.0

What it measures

Given a repository snapshot and a real GitHub issue, the agent (or retriever) must return a ranked list of code regions it believes are relevant to resolving the issue, subject to a fixed line budget (100, 300 or 500 lines in the released evaluator). The task separates exploration and localization from patch writing, which most issue-resolution benchmarks grade only as a single pass/fail outcome. Ground-truth "core" and "optional context" regions are derived from independent, successful repair trajectories rather than authored by hand.

Task format

Text input (issue description plus repository access); output is a ranked list of code regions (file + line range) within the line budget, not a patch.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub