QuSema Finds 40 Quantum Library Bugs

QuSema makes a compelling case for domain-grounded coding agents as a complement to black-box testing, especially where expected outputs or cross-library comparisons do not exist.

• –Outperformed Claude Code and Codex on a benchmark of 20 historical silent bugs
• –Found five bugs that differential and metamorphic testing could not detect
• –Generates executable tests, API-level triggers, and root-cause explanations for maintainers
• –Its reliance on LLM reasoning and documentation still leaves room for semantic hallucinations and incorrect specifications
• –The approach could generalize to other specialized developer ecosystems with formal domain rules