CausalRepair fixes 3x more code bugs with dual-slicing AI
New AI tool CausalRepair slashes debug costs to 3 cents per bug using DeepSeek-V3...
Deep Dive
CausalRepair is a new conversation-driven automated program repair (APR) framework that uses a dual-slicing strategy—combining static and dynamic code analysis—to build minimal causal context and pinpoint the true root cause of failures. Evaluated on Defects4J V1.2, V2.0, and Defects4J-Trans with DeepSeek-V3, CausalRepair correctly fixes 313 bugs on Defects4J, outperforming state-of-the-art approaches like ReinFix and TSAPR while cutting the average repair cost to $0.029 per bug.
Key Points
- CausalRepair improves automated bug fixing by 3x over prior SOTA tools on Defects4J benchmarks
- Dual-slicing combines static analysis and runtime traces to isolate causal code dependencies
- Repairs cost just $0.029 per bug using DeepSeek-V3, down from ~$0.08+ in existing systems
Why It Matters
Cuts debugging costs 60% while boosting AI-driven code repair accuracy—critical for DevOps and enterprise engineering teams.