Efficiency Hallucination: Measuring Behavioral Calibration in LLM-Based Code Optimization

The integration of Large Language Models (LLMs) into automated code optimization introduces a reliability risk we term the Efficiency Hallucination: the generation of plausible code edits accompanied by confident, unverifiable performance claims on already-optimal code. Unlike functional hallucinations, these are invisible to per-commit test suites: the code compiles and passes tests, yet no performance improvement occurs. In production pipelines, such edits cost review time and erode trust in automated tooling.