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": 17,
  "included_in_body": 1,
  "gray_background_blocks": 0,
  "framed_blocks": 0,
  "by_role_guess": {
    "publisher_landing_page": 9,
    "page_header": 2,
    "reference": 1,
    "affiliation": 2,
    "unknown_text": 1,
    "metadata": 1,
    "body": 1
  },
  "by_docling_label": {
    "page_header": 3,
    "text": 6,
    "page_footer": 1,
    "section_header": 2,
    "list_item": 5
  },
  "by_body_decision_reason": {
    "publisher_landing_page": 9,
    "first_page_metadata": 2,
    "docling_page_header": 1,
    "first_page_front_matter_heading": 1,
    "first_page_author_or_affiliation": 2,
    "outside_body_flow_list_item": 1,
    "body_before_non_intro_heading": 1
  },
  "visual_assets": {
    "count": 1,
    "indexable_count": 1,
    "suppressed_count": 0,
    "by_type": {
      "figure": 1
    },
    "by_caption_source": {
      "missing_caption": 1
    },
    "by_duplicate_reason": {
      "": 1
    },
    "missing_caption_count": 1
  }
}

Diff Summary

{
  "parsed_text_blocks": 17,
  "final_body_blocks": 1,
  "excluded_blocks": {
    "count": 16,
    "by_reason": {
      "publisher_landing_page": 9,
      "first_page_author_or_affiliation": 2,
      "first_page_metadata": 2,
      "docling_page_header": 1,
      "first_page_front_matter_heading": 1,
      "outside_body_flow_list_item": 1
    },
    "by_role_guess": {
      "publisher_landing_page": 9,
      "affiliation": 2,
      "page_header": 2,
      "metadata": 1,
      "reference": 1,
      "unknown_text": 1
    },
    "by_risk_level": {
      "low": 15,
      "high": 1
    },
    "high_risk_count": 1,
    "medium_risk_count": 0
  },
  "char_counts": {
    "parsed_text_chars": 1994,
    "final_body_chars": 69,
    "excluded_chars": 1925
  }
}

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。

Blocks