You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Secondary hygiene assertions excluded from main score: 3
Merged diagnostic + public-real cases: 74
Merged main assertions: 674
Source Mix
Source PDF
Pages
govinfo_cfr_2024_title1_vol1.pdf
4
irs_f1040_2024.pdf
1
irs_f1040_schedule_a_2024.pdf
1
irs_f1040_schedule_c_2024.pdf
2
irs_f1040_schedule_d_2024.pdf
2
nist_ai_rmf_1_0.pdf
5
nist_sp800_53r5.pdf
5
Document Types
Type
Pages
government_form
6
government_legal_text
4
government_technical_report
10
Main Assertion Mix
Assertion type
Count
caption_binding
1
reading_order
40
table_cell_exists
61
table_grid_cell
13
table_shape
1
text_presence
52
Public-Real Only Leaderboard
Parser
Passed
Failed
Score
PyMuPDF4LLM plain
114
54
0.6786
PyMuPDF4LLM bbox
114
54
0.6786
Qwen-VL API
91
77
0.5417
Marker
87
81
0.5179
Docling
77
91
0.4583
MinerU
67
101
0.3988
PaddleOCR
44
124
0.2619
Merged Leaderboard
Parser
Passed
Failed
Score
PyMuPDF4LLM bbox
550
124
0.8160
PyMuPDF4LLM plain
541
133
0.8027
Marker
522
152
0.7745
Docling
465
209
0.6899
Qwen-VL API
443
231
0.6573
MinerU
388
286
0.5757
PaddleOCR
317
357
0.4703
Review Standard
table_cell_exists checks visible form/table fields that should remain cell-like in Markdown.
table_grid_cell and table_shape are limited to the visually unambiguous NIST SP 800-53 revision table.
reading_order uses short page-local anchors and was spot-checked visually on forms, two-column legal text, and technical reports.
caption_binding is used once for a unique figure/caption pair.
Page-furniture text_absence checks are secondary hygiene checks and do not affect the main leaderboard.
Remaining Gap
This RC is stronger than the synthetic-heavy diagnostic release, but it is still government-source heavy. The next community step is to add 20-40 non-government public pages from sources such as PMC OA, OpenStax, PubTables-1M, ACL Anthology, DocLayNet, BLS, and permissive arXiv/OpenReview papers.