Scaffold modification performance on SpreadsheetBench (250 held-out tasks)
57.4Mean ScoreIG report
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| IG reportCondition=IG report, n=62026.05 | 57.4 | 30.4 | 4 | 256 | 53 | 203 | |
| Claude Code reportCondition=Claude Code report, n=62026.05 | 43.2 | 16.2 | 3.2 | 342 | 56 | 286 | |
| BaselineCondition=Baseline, n=–2026.05 | 27 | — | 0 | 0 | 0 | 0 |