Checked this week
- SWE-bench Lite added: all 300 tasks checked, each with a page of quoted evidence and its git timeline. 72 state the fix in the issue or discussion (Tier A), 12 repeat added lines of the gold patch (Tier C), 67 point to the code that changed (Tier B).
- Correction log: adding Lite changed no earlier verdict. The Verified counts are unchanged (500 tasks, 131 Tier A, 18 Tier C); a task in both tables keeps one verdict on both.