Condition: Naive vs Guided — both conditions, scored identically
Figures shown: mean accuracy across all questions and all 10 runs, per page, naive vs guided

Page generated: 2026-08-16
Data (model runs) generated: 2026-04-11
Scoring inputs: ground truth cc2ed8f5, evaluate.py 845b70e9

Per-Image Difficulty: Naive vs Guided

Pages: 39  |  Models: Claude, GPT, Gemini 2.5, Gemini 3.1, Claude (thinking)  |  Runs per condition: 10

Combined mean accuracy is the average across all 5 models × all 22 questions × 10 runs per page. Δ = guided − naive.

Click a sort button below to reorder the x-axis. Click a legend entry to toggle a series. Hover any bar for details.

Sort by: Order:

Per-Image Combined Accuracy Heatmap

Sorted by combined mean accuracy (lowest first). For each model, the left cell is the naive mean and the right cell is the guided mean. Mean N / Mean G = combined means across all models. Δ = guided − naive in percentage points.

Rank▲Page▲ImageClaudeGPTGemini 2.5Gemini 3.1Claude (thinking)CombinedStructural
density
Visual
density
NaiveGuidedNaiveGuidedNaiveGuidedNaiveGuidedNaiveGuidedMean NMean GΔ
121767/8 02 01 n146%47%34%30%48%55%46%53%52%47%45%46%+1.3652,695
241767 MV 02 01 n142%43%46%49%40%56%35%58%46%47%42%51%+9.0461,596
3111768 05 08 n439%37%22%44%65%59%46%68%39%44%42%50%+8.21182,633
4371770 05 28 n248%45%22%35%49%62%64%52%47%47%46%48%+2.31082,502
5231768 MV 01 0941%45%41%33%54%55%47%47%51%60%47%48%+1.5491,989
6221768 MV 01 0239%44%38%35%49%70%54%68%43%51%45%53%+8.9512,366
7241768 MV 01 16 n150%48%35%45%44%54%59%61%47%53%47%52%+5.4321,656
881768 05 08 n146%31%38%43%60%55%72%69%41%45%51%49%−2.6992,691
9101768 05 08 n345%40%28%43%71%67%58%60%50%45%50%51%+0.3972,480
10341770 05 21 n243%54%37%37%58%54%60%51%54%60%50%51%+1.1561,778
11191768 11 0750%48%51%51%42%53%51%60%55%54%50%53%+3.5532,308
12271768 MV 02 20 n154%45%44%46%62%49%58%58%52%48%54%49%−4.8672,629
1361768 03 2857%50%45%47%61%57%55%45%51%55%54%51%−3.1462,597
14161768 05 1642%45%33%35%64%58%75%61%61%59%55%51%−3.8351,935
15281768 MV 02 20 n247%46%48%46%61%73%54%63%46%51%51%56%+4.5361,614
16361770 05 28 n154%61%43%41%54%56%62%57%52%55%53%54%+1.0732,585
17261768 MV 01 16 n343%48%45%43%52%58%60%68%50%68%50%57%+6.8361,818
1811767 06 0150%43%38%38%57%69%73%73%58%61%55%57%+1.5562,136
19211768 12 0560%65%42%43%49%57%67%65%50%67%54%59%+5.9481,857
20121768 05 08 n551%44%37%43%77%77%66%77%63%60%59%60%+1.2721,710
21381770 05 28 n357%54%44%45%79%83%65%62%60%59%61%61%−0.5381,915
2251767 MV 02 01 n267%59%46%55%35%49%79%71%74%82%60%63%+2.8201,218
2391768 05 08 n254%42%48%52%79%76%69%79%62%60%62%62%−0.5511,614
24201768 11 2850%60%41%53%60%68%77%73%69%74%59%66%+6.2392,157
25331770 05 21 n162%58%58%53%57%54%81%82%57%65%63%62%−0.5281,939
26141768 05 09 n263%59%58%60%68%75%71%61%67%67%65%64%−1.1381,838
27351770 05 21 n345%50%48%59%79%75%81%92%69%71%64%69%+5.1191,571
28171768 08 0161%65%52%57%69%57%85%81%69%80%67%68%+0.4431,530
29321770 03 1359%57%60%60%78%77%73%72%81%81%70%70%−0.4311,800
30251768 MV 01 16 n250%54%58%68%78%74%85%86%73%80%69%73%+4.0271,102
31301769 05 08 n262%65%59%55%80%81%79%90%72%79%70%74%+3.8241,305
32291769 05 08 n168%74%67%63%71%89%90%90%81%88%75%81%+5.3251,438
33311770 03 0567%66%71%68%83%79%90%87%85%88%79%78%−1.7191,897
34131768 05 09 n172%67%70%65%73%94%100%95%90%88%81%82%+0.910782
35181768 09 1273%77%66%71%86%83%95%87%91%94%82%82%±0.0201,225
36151768 05 09 n385%76%65%71%74%81%96%99%87%91%82%84%+2.071,431
3731767/8 02 01 n293%84%43%56%85%90%95%100%95%96%82%85%+3.311423
38391770 08 2087%89%64%69%92%97%86%86%94%95%85%87%+2.7151,706
3971768 05 0290%75%78%82%77%86%95%96%95%93%87%86%−0.810805