eval/scoring.md and docs/scoring.md are both 190 lines and ~identical — but they have already started to drift, and the drift is in the links that matter:
|
eval/scoring.md |
docs/scoring.md |
| Live leaderboard |
huggingface.co/spaces/NAIL-Group/clawbench-leaderboard (superseded — kept only as a historical mirror per the 2026.05.12 news entry) |
huggingface.co/spaces/TIGER-Lab/ClawBench (canonical) |
| HF data card |
datasets/NAIL-Group/ClawBench |
spaces/TIGER-Lab/ClawBench (points at a Space, not a dataset) |
README links both: line 190 sends readers to docs/scoring.md, line 268 sends them to eval/scoring.md. zh-CN links eval/scoring.md twice. So a reader following "the scoring formula" gets one of two documents depending on which link they clicked, and one of them steers them to a deprecated leaderboard.
Fix: pick one canonical location (docs/scoring.md matches where the rest of the prose docs now live after #283), make the other a one-line stub pointing at it, and update the four inbound links. Also fix the spaces/ vs datasets/ mixup on the data-card row while touching it.
Related: this class of bug is what #291's markdown link checker guards against structurally, but it can't catch two files that both exist and disagree.
eval/scoring.mdanddocs/scoring.mdare both 190 lines and ~identical — but they have already started to drift, and the drift is in the links that matter:eval/scoring.mddocs/scoring.mdhuggingface.co/spaces/NAIL-Group/clawbench-leaderboard(superseded — kept only as a historical mirror per the 2026.05.12 news entry)huggingface.co/spaces/TIGER-Lab/ClawBench(canonical)datasets/NAIL-Group/ClawBenchspaces/TIGER-Lab/ClawBench(points at a Space, not a dataset)README links both: line 190 sends readers to
docs/scoring.md, line 268 sends them toeval/scoring.md. zh-CN linkseval/scoring.mdtwice. So a reader following "the scoring formula" gets one of two documents depending on which link they clicked, and one of them steers them to a deprecated leaderboard.Fix: pick one canonical location (
docs/scoring.mdmatches where the rest of the prose docs now live after #283), make the other a one-line stub pointing at it, and update the four inbound links. Also fix thespaces/vsdatasets/mixup on the data-card row while touching it.Related: this class of bug is what #291's markdown link checker guards against structurally, but it can't catch two files that both exist and disagree.