TraceEval is the first execution-verified multi-language benchmark for recovering runtime call structures from source code, containing 10,583 programs and showing top LLMs reach 72.9% average F1.
main . MyClass . func
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
An Execution-Verified Multi-Language Benchmark for Code Semantic Reasoning
TraceEval is the first execution-verified multi-language benchmark for recovering runtime call structures from source code, containing 10,583 programs and showing top LLMs reach 72.9% average F1.