Specification Grounding Drives Test Effectiveness for LLM Code

Amin Haeri, Mahdi Ghelichi

Large language models frequently generate code that appears correct on typical inputs yet fails on edge cases, invalid inputs, and other specification-defined corner conditions. A…