Docling layout block audit

original.pdf docling.json layout_blocks.json layout_blocks.tsv layout_blocks.jsonl excluded_blocks.html excluded_blocks.tsv final_body_blocks.tsv visual_assets.tsv visual_assets.json

Summary

{
  "blocks": 0,
  "included_in_body": 0,
  "gray_background_blocks": 0,
  "framed_blocks": 0,
  "by_role_guess": {},
  "by_docling_label": {},
  "by_body_decision_reason": {},
  "visual_assets": {
    "count": 5,
    "indexable_count": 5,
    "suppressed_count": 0,
    "by_type": {
      "figure": 4,
      "table": 1
    },
    "by_caption_source": {
      "missing_caption": 5
    },
    "by_duplicate_reason": {
      "": 5
    },
    "missing_caption_count": 5
  }
}

Diff Summary

{
  "parsed_text_blocks": 0,
  "final_body_blocks": 0,
  "excluded_blocks": {
    "count": 0,
    "by_reason": {},
    "by_role_guess": {},
    "by_risk_level": {},
    "high_risk_count": 0,
    "medium_risk_count": 0
  },
  "char_counts": {
    "parsed_text_chars": 0,
    "final_body_chars": 0,
    "excluded_chars": 0
  }
}

Truncation

{
  "truncated": false,
  "message": "",
  "truncated_block_count": 0
}

Page Overlays

Green = final body chunks. Orange/purple asset boxes = actual figure/table assets used by ingestion. Cyan dashed text boxes = text blocks consumed by those assets as caption continuations. Red STOP = truncation trigger. Red boxes = blocks after truncation.

Page 1

Page 2

Visual Assets

这里对齐真实图表资产提取链路。caption_source=embedded_table_cell 表示表注来自 Docling table cell,不会出现在 text block 审计差集里;caption_continuation_used_by_asset 表示某个 text block 已被图表 caption 吸收,不应按普通 metadata 解读。

#typelabelpagecaption sourcesuppressedduplicate reasonrescue reasongroupconfidencebboxcaption
1figureDocling Figure 11missing_caption0.55[53.73, 630.89, 234.39, 92.21]
2figureDocling Figure 21missing_caption0.55[317.24, 529.91, 227.44, 163.82]
3figureDocling Figure 32missing_caption0.55[49.75, 81.97, 235.07, 185.22]
4figureDocling Figure 42missing_caption0.55[307.43, 83.21, 241.68, 163.11]
5tableDocling Table 52missing_caption0.55[48.08, 645.52, 240.51, 80.59]

Blocks

pageorderlabelroleincludedriskreasonparser reasonproduction usage truncbody regionregionbgframebboxraw textcleaned text