mirror of
https://github.com/lllin000/PaperForge.git
synced 2026-07-22 17:00:23 +00:00
3cda942f96
3 distinct bugs found and fixed:
1. TABLE in figure regex (_FIGURE_NUMBER_PATTERN)
Removed TABLE/Table from figure regex — caused table captions to be
extracted as figure numbers and roadmapped into figure inventory.
Added unit tests (table_numeric_caption_is_not_a_figure_number,
table_appendix_caption_marker_has_no_figure_number).
2. int block_id type mismatch in cross-page lookup
CrossPageSettlementPass and PrimarySamePagePass compared block_id
with === but deduped_legends use int keys while ResourceRef stores
str. Fixed by casting both to str. Table continuation lookup
missed page filter, picking wrong page's same block_id. Added
page filter + int_block_id tests.
3. Table appendix support gaps
- Table prefix regex only accepted digits/roman, not 'TABLE A1'.
Added [A-Z]\d+ token and strip-leading-alpha in parse.
- _is_validation_first_table_candidate didn't cover figure_title
raw_label blocks with table_caption_like style. Added second
gate.
- _is_weak_explicit_table_caption only checked table_caption roles.
Extended to include validation-first candidates.
- Same-page tie-break didn't apply for weak-explicit captions.
Ported _bare_table_tie_break into vnext pass.
- figure_caption_candidate not excluded from note attachment.
- Continuation merge stripped leading duplicate table marker.
M84CTEM9 vault verification: 6/6 figures matched (3 main + 3
appendix + assets), 4/4 tables (Table 1,2 + Table A1,A2 + assets),
0 false positives, 0 figure asset leakage.
144 lines
No EOL
4.7 KiB
JSON
144 lines
No EOL
4.7 KiB
JSON
{
|
|
"audit_date": "2026-07-03",
|
|
"vault_papers": 731,
|
|
"papers_with_captions": 712,
|
|
"total_caption_blocks": 5822,
|
|
"caption_format_distribution": {
|
|
"standard_numeric": {
|
|
"count": 4093,
|
|
"pct": 70.3,
|
|
"status": "matched OK"
|
|
},
|
|
"plain_prefix_missing": {
|
|
"count": 1343,
|
|
"pct": 23.1,
|
|
"status": "PARTIAL — some reserved, some body text mislabeled"
|
|
},
|
|
"unclassified_other": {
|
|
"count": 270,
|
|
"pct": 4.6,
|
|
"status": "not matched, needs per-case check"
|
|
},
|
|
"scheme_numeric": {
|
|
"count": 32,
|
|
"pct": 0.5,
|
|
"status": "GAP — Scheme prefix not in figure regex"
|
|
},
|
|
"numeric_alpha_suffix": {
|
|
"count": 27,
|
|
"pct": 0.5,
|
|
"status": "GAP — suffix letter lost, hyphen variants break"
|
|
},
|
|
"chinese_图": {
|
|
"count": 22,
|
|
"pct": 0.4,
|
|
"status": "GAP — Chinese prefix not in figure regex"
|
|
},
|
|
"appendix_alpha": {
|
|
"count": 14,
|
|
"pct": 0.2,
|
|
"status": "GAP — alphabetic number prefix not supported"
|
|
},
|
|
"fig_nospace": {
|
|
"count": 10,
|
|
"pct": 0.2,
|
|
"status": "actually OK — parsed as fig_nospace→matched"
|
|
},
|
|
"suppl_S_numeric": {
|
|
"count": 8,
|
|
"pct": 0.1,
|
|
"status": "OK — S prefix supported"
|
|
},
|
|
"suppl_numeric": {
|
|
"count": 3,
|
|
"pct": 0.1,
|
|
"status": "OK — Supplementary prefix supported"
|
|
}
|
|
},
|
|
"vision_verified_gaps": [
|
|
{
|
|
"gap_id": "GAP-1",
|
|
"name": "Scheme N prefix not supported",
|
|
"affected_papers": 28,
|
|
"affected_captions": 32,
|
|
"root_cause": "regex only handles 'Figure'/'Fig.' prefixes",
|
|
"location": "ocr_figures.py:_FIGURE_NUMBER_PATTERN (line 14)",
|
|
"vision_verified": [
|
|
"BL8ABQ2A p2 — Scheme 1 has diagram, reserved with 0 assets"
|
|
],
|
|
"fix": "add 'Scheme' as recognized prefix option"
|
|
},
|
|
{
|
|
"gap_id": "GAP-2",
|
|
"name": "Alphabetic figure number (Figure A1/A2, Fig. A1)",
|
|
"affected_papers": 12,
|
|
"affected_captions": 14,
|
|
"root_cause": "regex uses \\d+ only for number capture, [A-Z] not recognized",
|
|
"location": "ocr_figures.py:_FIGURE_NUMBER_PATTERN, _extract_figure_marker (lines 14, 128)",
|
|
"vision_verified": [
|
|
"M84CTEM9 p8 — Figure A1 caption→image (block 33) not matched as asset",
|
|
"M84CTEM9 p9 — Figure A2 caption→nomogram image UNMATCHED",
|
|
"AH6Q7DLC — expected appendix pattern, no match"
|
|
],
|
|
"fix": "extend regex to optional ([A-Z])\\d* after 'Figure ' prefix"
|
|
},
|
|
{
|
|
"gap_id": "GAP-3",
|
|
"name": "Subfigure alphabetic suffix with hyphen (Fig. 2-A, Figs. 2-A through 2-D)",
|
|
"affected_papers": 14,
|
|
"affected_captions": 27,
|
|
"root_cause": "hyphen + letter suffix causes partial match or doubles",
|
|
"location": "ocr_figures.py:_FIGURE_NUMBER_PATTERN",
|
|
"vision_verified": [
|
|
"VAMSAZMG p4 — Fig. 2-A through 2-D caption UNMATCHED, Fig. 2-C duplicated"
|
|
],
|
|
"fix": "add hyphen+letter optional suffix as subfigure indicator"
|
|
},
|
|
{
|
|
"gap_id": "GAP-4",
|
|
"name": "Chinese 图 prefix not supported",
|
|
"affected_papers": 2,
|
|
"affected_captions": 22,
|
|
"root_cause": "regex doesn't include Chinese '图' prefix",
|
|
"location": "ocr_figures.py:_FIGURE_NUMBER_PATTERN",
|
|
"fix": "add '图' as recognized prefix"
|
|
},
|
|
{
|
|
"gap_id": "GAP-5",
|
|
"name": "Descriptive/plain-text captions misclassified as figure_caption",
|
|
"affected_papers": 374,
|
|
"affected_captions": 1343,
|
|
"root_cause": "67% (descriptive paragraphs + short_trash) are NOT figures — role classifier over-assigns figure_caption_candidate",
|
|
"subtypes": {
|
|
"descriptive_paragraph_not_figure": {
|
|
"count": 767,
|
|
"pct_of_category": 57.1
|
|
},
|
|
"very_short_trash": {
|
|
"count": 418,
|
|
"pct_of_category": 31.1
|
|
},
|
|
"short_alpha_label_subfigure": {
|
|
"count": 222,
|
|
"pct_of_category": 16.5
|
|
},
|
|
"mislabeled_table": {
|
|
"count": 67,
|
|
"pct_of_category": 5.0
|
|
}
|
|
}
|
|
}
|
|
],
|
|
"baseline_performance": {
|
|
"standard_numeric_papers": {
|
|
"matched": "~679 papers, ~4093 captions",
|
|
"verified": "NC66N4Q3 (26 fig, 52 assets) — all correct"
|
|
},
|
|
"fig_nospace": {
|
|
"matched": "all 5 captions in 5MAW65YD OK",
|
|
"verified": "regex actually handles no-space variant"
|
|
},
|
|
"summary": "Standard Figure N / Fig. N format works correctly for all 4093 cases across 679 papers"
|
|
},
|
|
"total_impact_estimate": "~50-80 figures with real content (caption + image) not matched across 731 papers, primarily from GAP-1 (Scheme, 28 papers) and GAP-2 (appendix alpha, 12 papers)"
|
|
} |