AI Paper Atlas

Research map / Large language models / Agents, code and math

PGS: Effective LLM Code Refinement via Property-Oriented and Structurally Minimal Feedback

Lehan He, Zeren Chen, Zhe Zhang, Xiang Gao, Lu Sheng

ICML 2026 · Large language models › Agents, code and math · approach: Execution-based code verification

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 changedWhereMeasuredEffect
reframe the same failure as a property violation instead of I/O feedback (ablation)one-shot refinement · LiveCodeBench-Hardpass@1: 32.0 -> 36.2Improved
select minimal-token-count counterexample instead of maximum (ablation)PGS feedback · LiveCodeBench-Easypass@1 improvement: +9.1 -> +13.2Improved
select minimal-token-count counterexample instead of median (ablation)PGS feedback · LiveCodeBench-Hardpass@1 improvement: +3.3 -> +4.0Improved
use minimal line coverage instead of minimal token count (ablation)PGS feedback · HumanEvalpass@1 improvement: +7.1 vs +7.0No clear effect
use maximal-complexity counterexample (max token count) rather than min: token cost (ablation)PGS feedback · LiveCodeBench (DS-R1-32B)avg tokens per attempt: 4.72k -> 5.68kHurt
use Qwen2.5 as Tester for DS-R1-32B generator (ablation)PGS · LiveCodeBenchpass@1: 76.5 -> 74.9No clear effect
use Qwen3-30B as Tester for DS-R1-32B generator (ablation)PGS · LiveCodeBenchpass@1: 76.5 -> 76.1No clear effect
inject property checks into code before refinement (ablation)PGS · LiveCodeBench (DS-R1-32B)Wrong Answer rate: 25.3% -> 12.6%Improved
full PGS refinement after property injection (ablation)PGS · LiveCodeBench (DS-R1-32B)Pass rate: 64.4% -> 74.5%Improved
filter properties by validation before use (ablation)verification · LiveCodeBench-Hardverification accuracy: 76.5% -> 88.2%Improved
PGS input synthesis vs baseline TDD refinementPGS · HumanEval (DS-V2) failed problemsfix rate: 23.1% -> 53.8%Improved
PGS vs best baseline Self-EditPGS · LiveCodeBench (DS-R1-32B)pass@1: 73.6 -> 76.5Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy