Research map / Large language models / Agents, code and math
TestExplora: Benchmarking LLMs for Proactive Bug Discovery via Repository-Level Test Generation
Steven Liu, Jane Luo, Xin Zhang, Aofan Liu, Hao Liu, Jie Wu, Ziyang Huang, Yangyu Huang, Yu Kang, Scarlett Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add full code and dependencies (White Box instead of Black Box) | GPT-5-mini · TestExplora | F2P: 5.17 -> 7.54 | Improved |
| add dependency code (White Box instead of Black Box) | GPT-4o · TestExplora-Lite | F2P: 3.61 -> 2.20 | Hurt |
| add full dependency code (White Box instead of Black Box) | o4-mini · TestExplora-Lite | F2P: 3.74 -> 6.45 | Improved |
| replace human-written docs with DocAgent-generated docs (ablation) | GPT-5-mini · TestExplora-Lite (Black Box) | F2P: +6.06 | Improved |
| replace human-written docs with DocAgent-generated docs (ablation) | LLMs on Black Box · TestExplora-Lite | number of tests: -0.16 to +0.70 | No clear effect |
| replace human-written docs with DocAgent-generated docs (ablation) | Gemini-2.5-pro · TestExplora-Lite (Black Box) | F2P: +0.61 | Improved |
| use SWEAgent instead of plain static context | GPT-5-mini · TestExplora (White Box) | F2P: 12.79 -> 17.27 | Improved |
| use SWEAgent instead of plain static context | o4-mini · TestExplora (White Box) | F2P: 5.92 -> 9.42 | Improved |
| use TraeAgent instead of plain static context | o4-mini · TestExplora (White Box) | F2P: 5.92 -> 12.16 | Improved |
| use TraeAgent instead of plain static context | GPT-5-mini · TestExplora (White Box) | F2P: 12.79 -> 14.55 | Improved |
| increase samples from pass@1 to pass@5 | SWEAgent GPT-5-mini · TestExplora-Lite | F2P: 17.3 -> 29.7 | Improved |
| increase samples from pass@1 to pass@5 | SWEAgent o4-mini · TestExplora-Lite | F2P: 9.4 -> 21.2 | Improved |
Closest papers by meaning
- SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents (NeurIPS 2024 · similarity 0.89)
- TestGenEval: A Real World Unit Test Generation and Test Completion Benchmark (ICLR 2025 · similarity 0.88)
- RepoAudit: An Autonomous LLM-Agent for Repository-Level Code Auditing (ICML 2025 · similarity 0.88)
- DPIAgent: Divide, Protocol, Isolate for Agentic Reproduction Test Generation (NeurIPS 2026 · similarity 0.88)
- SWE-bench Goes Live! (NeurIPS 2025 · similarity 0.86)
- SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
