Saved in:
Bibliographic Details
Main Authors: Li, Jia, Su, Yuxin, Peng, Ting, Huang, Hailiang, Deng, Yuetang, Lyu, Michael R.
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.07276
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910200612519936
author Li, Jia
Su, Yuxin
Peng, Ting
Huang, Hailiang
Deng, Yuetang
Lyu, Michael R.
author_facet Li, Jia
Su, Yuxin
Peng, Ting
Huang, Hailiang
Deng, Yuetang
Lyu, Michael R.
contents Code-agent RL often receives weak feedback: rollout-time signals are reliable and executable, but capture only necessary or surface conditions for task success rather than the target semantic predicate. Using agentic compile-fix as the setting, we study signal reshaping for standard GRPO under such feedback. Our central claim is that GRPO's within-group comparison is meaningful only after three kinds of signals are reshaped: outcome rewards recover semantic ranking, process signals localize intra-trajectory credit, and rollouts from the same prompt remain execution-comparable. We operationalize these conditions with a minimal signal-reshaping construction that leaves GRPO's group-normalized advantage construction unchanged: compile-and-semantic layered rewards reshape trajectory ranking, step-level process scores outside group reward normalization reshape within-trajectory update strength, and failure-cause-aware rollout governance reshapes within-group comparability. Experiments show a clear end-to-end gain: full signal-reshaped GRPO improves strict compile-and-semantic accuracy from the base model's zero-shot $0.385$ to $0.535$. Controlled comparisons further explain the source of this gain: binary rewards remove the compile-only middle tier and degrade trajectory control; on top of layered rewards, process-score weighting further improves accuracy from $0.48$ to $0.53$ and reduces average evaluation steps from $23.50$ to $17.02$. As a boundary comparison, privileged-prompt token-level distillation mainly optimizes local distributional alignment; in long tool-use trajectories, this signal is diluted by non-critical tokens and cannot replace outcome semantics, process credit, or within-group comparability.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07276
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair
Li, Jia
Su, Yuxin
Peng, Ting
Huang, Hailiang
Deng, Yuetang
Lyu, Michael R.
Artificial Intelligence
Code-agent RL often receives weak feedback: rollout-time signals are reliable and executable, but capture only necessary or surface conditions for task success rather than the target semantic predicate. Using agentic compile-fix as the setting, we study signal reshaping for standard GRPO under such feedback. Our central claim is that GRPO's within-group comparison is meaningful only after three kinds of signals are reshaped: outcome rewards recover semantic ranking, process signals localize intra-trajectory credit, and rollouts from the same prompt remain execution-comparable. We operationalize these conditions with a minimal signal-reshaping construction that leaves GRPO's group-normalized advantage construction unchanged: compile-and-semantic layered rewards reshape trajectory ranking, step-level process scores outside group reward normalization reshape within-trajectory update strength, and failure-cause-aware rollout governance reshapes within-group comparability. Experiments show a clear end-to-end gain: full signal-reshaped GRPO improves strict compile-and-semantic accuracy from the base model's zero-shot $0.385$ to $0.535$. Controlled comparisons further explain the source of this gain: binary rewards remove the compile-only middle tier and degrade trajectory control; on top of layered rewards, process-score weighting further improves accuracy from $0.48$ to $0.53$ and reduces average evaluation steps from $23.50$ to $17.02$. As a boundary comparison, privileged-prompt token-level distillation mainly optimizes local distributional alignment; in long tool-use trajectories, this signal is diluted by non-critical tokens and cannot replace outcome semantics, process credit, or within-group comparability.
title Signal Reshaping for GRPO in Weak-Feedback Agentic Code Repair
topic Artificial Intelligence
url https://arxiv.org/abs/2605.07276