CodeAssay is a 185-task Python benchmark showing that auditing benchmark ground truth flips 9% of model correctness labels and widens measured model spread, and that a security-focused prompt costs code size and complexity without measurable security gains.
ACM Transactions on Software Engineering and Methodology33(8), 1–79 (2024)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.SE 1years
2026 1verdicts
ACCEPT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
CodeAssay: A Multi-Metric Benchmark with Audited Ground Truth for LLM Code Generation
CodeAssay is a 185-task Python benchmark showing that auditing benchmark ground truth flips 9% of model correctness labels and widens measured model spread, and that a security-focused prompt costs code size and complexity without measurable security gains.