Research map / Large language models / Agents, code and math
MatchFixAgent: Language-Agnostic Autonomous Repository-Level Code Translation Validation and Repair
Ali Reza Ibrahimzada, Brandon Paulsen, Reyhaneh Jabbarvand, Joey Dodds, Daniel Kroening
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 semantic analyzer and test generator agents on top of a standalone coding agent (ablation) | MatchFixAgent · 1091 non-dispute translation pairs | validation accuracy: accuracy drops by 42.3% without them | Improved |
| add the six semantic analyses to test generator and verdict agents (ablation) | MatchFixAgent · 1091 non-dispute translation pairs | validation accuracy: 60.31% -> 100% | Improved |
| add semantic analyses to reduce agent cost (ablation) | MatchFixAgent · 1091 non-dispute translation pairs | turns, tokens, time: turns 66.3K -> 63.7K; input tokens 136K -> 128K; duration 61.28h -> 56.71h | Improved |
| standalone agent versus multi-agent on disputed cases where MatchFixAgent is correct (ablation) | standalone Claude Code agent · 145 manually resolved disputes | accuracy: 47.7% vs 56.1% on prior-tool-correct cases | Improved |
| use MatchFixAgent repair instead of reprompting-based repair | translation repair · 265 buggy translations | repaired (%): 18.5% -> 50.6% | Improved |
| add generated tests from MatchFixAgent | project code coverage · subject projects | coverage (%): 89.6% -> 98.1% | Improved |
| swap Claude 3.7 Sonnet and Claude Code for o4-mini and Codex (ablation) | MatchFixAgent · 96 sampled instances | agreement with existing validation: 73% agreement | No clear effect |
| short-circuit control flow analyzer at CFG similarity threshold 0.7 | MatchFixAgent control flow analyzer · all experiments | LLM invocations skipped: approximately 25% of LLM invocations skipped | Improved |
| short-circuit data flow analyzer at similarity threshold 0.7 | MatchFixAgent data flow analyzer · all experiments | LLM invocations skipped: approximately 35% of LLM invocations skipped | Improved |
| use PL-agnostic Tree-Sitter design instead of per-pair tooling | MatchFixAgent · development cost | lines of code: 1650 vs 10859 / 19052 / 3843 | Improved |
Closest papers by meaning
- Automatically Testing Functional Properties of Code Translation Models (AAAI 2024 · similarity 0.89)
- AlphaVerus: Bootstrapping Formally Verified Code Generation through Self-Improving Translation and Treefinement (ICML 2025 · similarity 0.87)
- SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents (NeurIPS 2024 · similarity 0.87)
- Synthetic Programming Elicitation for Text-to-Code in Very Low-Resource Programming and Formal Languages (NeurIPS 2024 · similarity 0.86)
- Verifiers in the Loop: Decoding Time Verification for Code Translation (NeurIPS 2026 · similarity 0.86)
- Verify0: Can AI Agents Build Formally Verified Software Repositories? (NeurIPS 2026 · 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.
