PDF Parity Checker Skill

Purpose

Compare the 44 XHTML chapter files against their corresponding POD (print-on-demand) PDF files to ensure visual and structural consistency. This is critical for maintaining brand quality across digital and print editions.

When to Invoke

• User asks "do the PDFs match the EPUB chapters?"
• Before sending POD files to IngramSpark or print vendor
• After making changes to XHTML or CSS
• User mentions "print edition" or "PDF consistency"
• User asks "verify the PDFs are up to date"

Workflow

Run PDF Parity Verification

python3 scripts/pdf_verify.py \
  --root REBRANDED_OUTPUT \
  --targets docs/REBRANDED_VISUAL_AUDIT.json \
  --update-json

What it does:

1. For each of the 44 XHTML files:
   • Locates corresponding PDF in REBRANDED_OUTPUT/pdf-pod/
   • Compares:
     • Page count (XHTML rendered vs PDF pages)
     • Media box dimensions (PDF page size)
     • First-page visual hash (downscaled grayscale comparison)
     • Text extraction and paragraph continuity
2. If PDF is missing:
   • Generates temporary reference PDF via headless browser print-to-PDF
   • Uses this for comparison (but does NOT commit to repo)
   • Flags as "MISSING" in report
3. Updates docs/REBRANDED_VISUAL_AUDIT.json with:
   • pdf_check object for each chapter
   • Fields: page_count_match, bbox_match, image_hash_delta, pdf_status

Comparison Metrics

1. Page Count Match

Compares rendered XHTML page count vs PDF page count.

Example:

Chapter IX: "Unveiling Your Creative Odyssey"
- XHTML rendered: 8 pages (at 6×9" print size)
- PDF actual: 8 pages
- Status: ✅ MATCH

Acceptable variance:

• Exact match: ✅ PASS
• ±1 page: ⚠️ WARN (minor reflow difference)
• ±2+ pages: ❌ FAIL (significant layout mismatch)

2. Media Box (Page Size)

Verifies PDF pages are correct physical dimensions.

Expected for 6×9" POD:

• Width: 432 points (6 inches × 72 DPI)
• Height: 648 points (9 inches × 72 DPI)

Example:

Chapter XV: Media box check
- Expected: 432×648 pt
- Actual: 432×648 pt
- Status: ✅ MATCH

3. Visual Hash Comparison

Computes perceptual hash of first page to detect visual differences.

Process:

1. Render XHTML first page as PNG (grayscale, downscaled to 200×300)
2. Convert PDF first page to PNG (same size)
3. Compute average hash for both
4. Calculate Hamming distance

Scoring:

• Hash delta 0-5: ✅ IDENTICAL (perfect match)
• Hash delta 6-15: ✅ SIMILAR (acceptable variance)
• Hash delta 16-30: ⚠️ DIFFERENT (minor layout shift)
• Hash delta >30: ❌ MISMATCH (significant visual difference)

Example:

Chapter IV: Visual hash comparison
- XHTML hash: d4a3f2c1...
- PDF hash: d4a3f2c1...
- Hamming distance: 3
- Status: ✅ IDENTICAL

4. Text Extraction

Extracts text from PDF and verifies key content is present.

Checks:

• Chapter title appears in first 500 characters
• Heading order matches XHTML heading structure
• Paragraph count is similar (±10%)

Example:

Chapter XII: Text extraction
- Title found: ✅ "Financial Wisdom"
- Headings: 12 in XHTML, 12 in PDF ✅
- Paragraphs: 84 in XHTML, 83 in PDF ✅ (within 10%)
- Status: ✅ PASS

Interpreting Results

JSON Output Structure

{
  "file": "REBRANDED_OUTPUT/xhtml/9-chapter-i-unveiling-your-creative-odyssey.xhtml",
  "basename": "9-chapter-i-unveiling-your-creative-odyssey",
  "pdf_check": {
    "pdf_path": "REBRANDED_OUTPUT/pdf-pod/chapters/9-chapter-i-unveiling-your-creative-odyssey.pdf",
    "pdf_status": "ok",
    "page_count_match": true,
    "page_count_xhtml": 8,
    "page_count_pdf": 8,
    "bbox_match": true,
    "bbox_expected": [432, 648],
    "bbox_actual": [432, 648],
    "image_hash_delta": 3,
    "image_hash_verdict": "identical",
    "text_checks": {
      "title_found": true,
      "heading_count_match": true,
      "paragraph_variance_pct": 1.2
    }
  }
}

Markdown Summary

Generated in docs/REBRANDED_VISUAL_AUDIT.md:

Common Issues and Fixes

Issue: Page Count Mismatch

Symptom: XHTML renders as 8 pages, PDF has 9 pages

Possible causes:

1. Extra blank page in PDF (page break issue)
2. Different margin settings between XHTML and PDF export
3. Widow/orphan control differences

How to fix:

1. Open PDF in Acrobat to verify blank page
2. Adjust print-pod.css orphans/widows settings:
   cssCopy

   p { orphans: 2; widows: 2; }
   

3. Re-export PDF from InDesign or print-to-PDF workflow
4. Re-run parity check to verify

Issue: Visual Hash Mismatch

Symptom: Hash delta >30 (significant visual difference)

Possible causes:

1. Font substitution in PDF vs XHTML
2. Image resolution difference
3. Different CSS applied (print vs digital styles)

How to fix:

1. Open both side-by-side:
   • XHTML screenshot: docs/screenshots/{basename}/title_1080x1440.png
   • PDF page 1: Open in Acrobat, export as PNG
2. Identify visual difference (typography, spacing, images)
3. Adjust print-pod.css to match digital rendering
4. Regenerate PDF
5. Re-run parity check

Issue: Missing PDF

Symptom: pdf_status: "missing"

Action:

1. Check if PDF was not generated yet for this chapter
2. Verify PDF naming matches XHTML basename:
   • XHTML: 9-chapter-i-unveiling-your-creative-odyssey.xhtml
   • PDF should be: 9-chapter-i-unveiling-your-creative-odyssey.pdf
3. If PDF exists with different name, rename to match
4. If PDF doesn't exist:
   • Generate from XHTML using print-to-PDF workflow
   • Place in appropriate pdf-pod/ subdirectory

Issue: Text Extraction Failure

Symptom: text_checks.title_found: false

Possible causes:

1. PDF is image-only (scanned, not text-based)
2. PDF has font embedding issues (text not extractable)
3. PDF was generated with text as outlines/curves

How to fix:

1. Verify PDF is text-based (select text in Acrobat)
2. If image-only, regenerate PDF from source XHTML
3. Ensure fonts are embedded as subsets (not outlines)

Example Invocation

User: "Are the print PDFs in sync with the latest EPUB chapters?"

Agent response:

1. Run PDF parity checker
2. Read results from updated JSON
3. Summarize:

PDF Parity Check Complete
=========================

Total chapters: 44
PDFs verified: 42
Missing PDFs: 2

Results:
✅ PASS: 38 chapters (perfect parity)
⚠️ WARN: 4 chapters (minor variance)
❌ FAIL: 0 chapters (significant mismatch)
🔍 MISSING: 2 chapters (PDFs not found)

Warnings (4 chapters):
- Chapter VI: +1 page variance (11 vs 10) - Extra blank page in PDF
- Chapter XI: Visual hash delta 18 - Minor font rendering difference
- Chapter XV: Text variance 12% - Some paragraphs reflowed
- Chapter XXII: Media box 432×660 - Height should be 648

Missing PDFs (2 chapters):
- 30-SelfAssessment.xhtml (no matching PDF found)
- 43-DoodlePage.xhtml (no matching PDF found)

Recommended actions:
1. Fix page break in Chapter VI
2. Review font settings for Chapter XI
3. Generate missing PDFs for Self-Assessment and Doodle pages
4. Verify media box for Chapter XXII

Full report: docs/REBRANDED_VISUAL_AUDIT.md (PDF Parity column)
Detailed JSON: docs/REBRANDED_VISUAL_AUDIT.json (pdf_check objects)

Integration with Other Skills

Run after:

• epub-visual-auditor - Ensure XHTML rendering is correct first

Run before:

• Sending POD files to print vendor
• Uploading to IngramSpark or KDP Print
• Final publication package

Pair with:

• epub-publication-validator - Comprehensive pre-publication check

Notes

• PDF comparison requires pypdf and Pillow Python libraries
• First run may be slower (generates temporary PDFs for missing files)
• Temporary reference PDFs are stored in /tmp/ and not committed to repo
• Visual hash comparison is perceptual (small rendering differences are OK)
• Re-run after any CSS or XHTML changes to verify parity maintained