AI ยท 2026-10-25 16:09
The paper examines test-graded coding-agent benchmarks and identifies gaps between task specifications and test-based judgments, warning that these gaps can distort conclusions about agent capabilities.
Auto-generated summary for reference only. RSS