From Code Review to Code Critique: Intent, Drift, and Spotlight for AI-Generated Diffs at Scale
Source: https://arxiv.org/abs/2607.29516
Authors: Chandra Maddila, Mashrur Rashik, Euna Mehnaz Khan, Smriti Jha, James Saindon, Nachi Nagappan, Peter...
Published: 2026-07-31
Categories: cs.SE, cs.AI
Grounding: opencli arxiv metadata + Obsidian paper digest excerpt (OpenClaw定时任务/论文流水线/2026-08-04-论文流水线.md)
摘要（中文）
论文提出 ARCTIC，把 AI 生成 diff 的 review 改成 code critique：预测变更意图检测开发者意图和 Agent 输出之间的 drift标出最需要人工看的 diff 区域摘要报告 intent prediction F1 为 0.86，drift detection 与人工 ordinal 标注的 QWK 为 0.907，spotlight 在 5x 更少 token 下质量估计优于 baseline AI reviewer 2.4x
Abstract
AI coding agents are generating code at volumes that exceed the capacity of traditional peer review. At the same time, existing AI code review tools over-index on low-value suggestions such as style and best practices while under-indexing on the concerns human reviewers prioritize most: correctness, security, and performance. We present ARCTIC, an AI-powered Code Critique system that reframes code review around three capabilities: intent prediction, which infers why a change was made from conversation logs and metadata; drift detection, which measures divergence between the developer's intent and the agent's output via backtranslation; and code spotlight, which ranks the regions of a diff most warranting human scrutiny. We ground these capabilities in a six-theme taxonomy derived from 18,000 code reviews.
Intake rationale
- Category: coding
- Quality score: 4/5
- Why it matters: ARCTIC reframes AI-generated diff review around intent, drift, and human spotlighting.
Obsidian evidence excerpt
论文提出 ARCTIC，把 AI 生成 diff 的 review 改成 code critique：预测变更意图、检测开发者意图和 Agent 输出之间的 drift、标出最需要人工看的 diff 区域。摘要报告 intent prediction F1 为 0.86，drift detection 与人工 ordinal 标注的 QWK 为 0.907，spotlight 在 5x 更少 token 下质量估计优于 baseline AI reviewer 2.4x。