audit: data-driven truth audit of 2HEUD5P9 + 4AG67PBH

Findings:
  - zone_leak_frontmatter_to_body: ORCID/copyright/DOI metadata in body_zone (2 papers)
  - reference_boundary_body_mix: body blocks intermixed with ref zone (2 papers)
  - title_repeat_on_page_2: section heading misclassified as paper_title (2HEUD5P9)
  - 12 empty unknown_structural ghost blocks (2HEUD5P9)
  - backmatter_body in body_zone vs tail zone (4AG67PBH)

audit/2026-06-28-data-audit-findings.json

@@ -0,0 +1,125 @@
+{
+  "audit_date": "2026-06-28",
+  "audit_type": "data_driven",
+  "papers": {
+    "2HEUD5P9": {
+      "total_blocks": 670,
+      "total_pages": 27,
+      "findings": [
+        {
+          "severity": "BUG",
+          "page": 2,
+          "block_role": "paper_title",
+          "text": "2. Main Types of Bone Biomaterials",
+          "issue": "section heading on page 2 misclassified as paper_title (title repeat detection too broad)",
+          "expected": "section_heading or noise",
+          "category": "title_repeat_misclassification"
+        },
+        {
+          "severity": "BUG",
+          "pages": ["1", "23"],
+          "block_role": "frontmatter_noise",
+          "zone": "body_zone",
+          "text": "ORCID + Received/Revised dates",
+          "issue": "frontmatter metadata elements in body_zone instead of frontmatter_main_zone",
+          "expected": "frontmatter_noise in frontmatter_main_zone",
+          "category": "zone_leak"
+        },
+        {
+          "severity": "MINOR",
+          "pages": ["2", "3", "6", "8", "10", "12", "16", "19", "22", "23", "25", "26"],
+          "block_role": "unknown_structural",
+          "zone": "body_zone",
+          "text": "empty or whitespace",
+          "issue": "12 empty unknown_structural blocks spread across body zone — likely OCR ghost blocks",
+          "expected": "dropped or noise in no_zone",
+          "category": "ocr_ghost_blocks"
+        },
+        {
+          "severity": "BUG",
+          "page": 23,
+          "zone_mix": true,
+          "zones": ["body_zone(7)", "reference_zone(45)", "frontmatter_side_zone(1)"],
+          "issue": "body blocks intermixed with reference zone on page 23 — body blocks before reference boundary",
+          "expected": "body blocks should be pre-reference and reference_zone should start cleanly",
+          "category": "reference_boundary_body_leak"
+        },
+        {
+          "severity": "INFO",
+          "pages": ["1"],
+          "note": "Frontmatter clean: paper_title (2 blocks), authors (1 block) in frontmatter_main_zone"
+        },
+        {
+          "severity": "INFO",
+          "pages": ["1"],
+          "note": "Abstract: 1 abstract_body in body_zone — correct"
+        }
+      ]
+    },
+    "4AG67PBH": {
+      "total_blocks": 599,
+      "total_pages": 25,
+      "findings": [
+        {
+          "severity": "BUG",
+          "page": 1,
+          "block_role": "frontmatter_noise",
+          "zone": "body_zone",
+          "text": "ORCID, copyright, DOI",
+          "issue": "3 frontmatter_noise metadata blocks in body_zone instead of frontmatter_main_zone",
+          "expected": "frontmatter_noise in frontmatter_main_zone",
+          "category": "zone_leak"
+        },
+        {
+          "severity": "BUG",
+          "page": 21,
+          "block_role": "backmatter_body",
+          "zone": "body_zone",
+          "text": "Supporting Information, dates, animal models",
+          "issue": "5 backmatter_body blocks in body_zone — post-reference backmatter should be in tail zone",
+          "expected": "backmatter_body in post_reference_backmatter_zone or tail_nonref_hold_zone",
+          "category": "backmatter_zone_leak"
+        },
+        {
+          "severity": "BUG",
+          "page": 21,
+          "zone_mix": true,
+          "zones": ["body_zone(8)", "reference_zone(35)", "frontmatter_side_zone(1)"],
+          "issue": "Mixed body + reference zone on page 21 with Conflict of Interest in frontmatter_side_zone",
+          "expected": "conflict of interest should be in tail zone or body_zone",
+          "category": "reference_boundary_mix"
+        },
+        {
+          "severity": "INFO",
+          "pages": ["1"],
+          "note": "Frontmatter clean: paper_title (1), authors (1) in frontmatter_main_zone"
+        },
+        {
+          "severity": "INFO",
+          "pages": ["1"],
+          "note": "Abstract: 1 abstract_body in body_zone — correct"
+        }
+      ]
+    }
+  },
+  "cross_paper_patterns": {
+    "zone_leak_frontmatter_to_body": {
+      "count": 2,
+      "papers": ["2HEUD5P9", "4AG67PBH"],
+      "description": "ORCID/copyright/DOI/date metadata blocks in body_zone — likely from frontmatter normalization missing these specific text patterns",
+      "fix": "Add ORCID, DOI, and date text patterns to frontmatter_noise normalization in _normalize_frontmatter_blocks"
+    },
+    "reference_boundary_body_mix": {
+      "count": 2,
+      "papers": ["2HEUD5P9", "4AG67PBH"],
+      "description": "Body paragraphs intermixed with reference zone on reference boundary pages",
+      "fix": "P1 backmatter boundary redesign should handle this — verify ref-anchored partition"
+    },
+    "title_repeat_on_page_2": {
+      "count": 1,
+      "papers": ["2HEUD5P9"],
+      "description": "Section heading on page 2 classified as paper_title",
+      "fix": "Tighten doc_title raw_label check for page 2: only accept if text matches page 1 title"
+    }
+  }
+}