SGCR: A Specification-Grounded Framework for Trustworthy LLM Code Review

2025-12-19

AI Summary

LLM-driven code review suffers from low reliability, weak contextual awareness, and poor controllability. To address these limitations, we propose SGCR, a specification-anchored code review framework featuring a novel explicit–implicit dual-path architecture: the explicit path rigorously derives verifiable rules from human-written specifications, while the implicit path assists in identifying latent defects not explicitly defined. SGCR integrates specification modeling, dual-path prompt engineering, context-aware feedback generation, and industrial-scale LLM deployment. Evaluated in a real-world setting at Haitong Securities Research Institute, SGCR achieved a 42% developer adoption rate—representing a 90.9% relative improvement over the baseline LLM (22% → 42%). This constitutes the first empirical validation that specification anchoring significantly enhances the reliability of LLM-based code review.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Automating code review with Large Language Models (LLMs) shows immense promise, yet practical adoption is hampered by their lack of reliability, context-awareness, and control. To address this, we propose Specification-Grounded Code Review (SGCR), a framework that grounds LLMs in human-authored specifications to produce trustworthy and relevant feedback. SGCR features a novel dual-pathway architecture: an explicit path ensures deterministic compliance with predefined rules derived from these specifications, while an implicit path heuristically discovers and verifies issues beyond those rules. Deployed in a live industrial environment at HiThink Research, SGCR's suggestions achieved a 42% developer adoption rate-a 90.9% relative improvement over a baseline LLM (22%). Our work demonstrates that specification-grounding is a powerful paradigm for bridging the gap between the generative power of LLMs and the rigorous reliability demands of software engineering.

Problem

• Addresses LLMs' unreliability in automated code review
• Enhances context-awareness and control in code feedback
• Bridges generative power with software engineering reliability

Innovation

• Framework grounds LLMs in human-authored specifications for trustworthy feedback
• Dual-pathway architecture ensures rule compliance and heuristically discovers issues
• Specification-grounding bridges LLM generative power with software engineering reliability