JevCoder: When Does Decision Delegation Pay Off in Code Repair?
A coding agent repeatedly makes decisions as it repairs code.
ProofPaper ↗
Key points
- These decisions shape repair trajectories and affect the token cost of successive model requests.
- We ask which repair decisions can be delegated to a low-cost probabilistic judge to reduce token use while maintaining repair quality.
- We evaluate judgment quality across repair stages and test how individual interventions change subsequent repair work.
- Based on these studies, we propose JevCoder, a workflow built around Jev, a low-cost probabilistic judge.
Sources (1)
- [1]JevCoder: When Does Decision Delegation Pay Off in Code Repair?arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 05:11 AM
A coding agent repeatedly makes decisions as it repairs code.
These decisions shape repair trajectories and affect the token cost of successive model requests.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026DecisionRobust: Benchmarking Robustness for JEV Style Decision Models Across Text and Vision
- Oct 13, 2026Show HN: Parsimony – an open-source benchmark for how much code AI models add
- Oct 13, 2026Run fast, private agents on-device with LFM2.5-24B-A2B
- Oct 13, 2026SWEPatchTest: Evaluating Software Tests Beyond Reproduction
- Oct 13, 2026Reading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent Loops
- Oct 12, 2026Deploy decision models with llama.cpp on Amazon SageMaker AI