1. Code correctness
AdequateProduces working changes that meet requirements without introducing defects.
An independent 2025 study tested Copilot suggestions on 2,033 LeetCode problems and found at least one correct suggestion for 70% of problems, with correctness varying substantially by language and difficulty. This supports useful code generation, but the algorithmic, suggestion-focused evaluation does not establish the reliability of end-to-end project changes by Copilot’s current agent workflows.
Evidence confidence: medium