Sense and Sensitivity: Examining the Influence of Semantic Recall on Long Context Code Reasoning

TL;DR

This paper investigates whether large language models truly understand code semantics in long contexts or rely on pattern matching, revealing significant semantic recall degradation and positional effects.

Contribution

It introduces semantic recall sensitivity, a novel measurement method, and a new task SemTrace to evaluate true semantic understanding in LLMs.

Findings

Frontier models excel at lexical recall but struggle with semantic recall in long contexts.

Models heavily depend on pattern matching shortcuts, especially when relevant code is centrally positioned.

Semantic recall sensitivity reveals severe positional accuracy drops, indicating underestimated semantic understanding failures.

Abstract

Large language models (LLMs) are increasingly deployed for understanding large codebases, but whether they understand operational semantics of long code context or rely on pattern matching shortcuts remains unclear. We distinguish between lexical recall (retrieving code verbatim) and semantic recall (understanding operational semantics). Evaluating 10 state-of-the-art LLMs, we find that while frontier models achieve near-perfect, position-independent lexical recall, semantic recall degrades severely when code is centrally positioned in long contexts. We introduce semantic recall sensitivity to measure whether tasks require understanding of code's operational semantics vs. permit pattern matching shortcuts. Through a novel counterfactual measurement method, we show that models rely heavily on pattern matching shortcuts to solve existing code understanding benchmarks. We propose a new task SemTrace, which achieves high semantic recall sensitivity through unpredictable operations; LLMs' accuracy exhibits severe positional effects, with median accuracy drops of 92.73% versus CRUXEval's 53.36% as the relevant code snippet approaches the middle of the input code context. Our findings suggest current evaluations substantially underestimate semantic recall failures in long context code understanding.