Condition: Naive vs Guided — both conditions, scored identically
Figures shown: mean signed error (model − ground truth) per page, averaged over 10 runs, naive vs guided
Page generated: 2026-08-16
Data (model runs) generated: 2026-04-11
Scoring inputs: ground truth cc2ed8f5, evaluate.py 845b70e9
For each counting question, each model shows the mean signed error (model response − GT) per page across 10 runs. Zero line = ground truth. Bars to the right = over-count; bars to the left = under-count. Naive Guided Both at 50% transparency so overlaps are visible. Page IDs run top to bottom; models read left to right. Hover for details.