DocAccessible
Checking account

Published research release / updated August 6, 2026

PDF-to-HTML conversion benchmark corpus

Fifty public PDFs. Fifty reviewed HTML revisions. Every result stays connected to its source, conversion hash, automated evidence, and final human review record.

Explore 50 recordsHuman-review CSV

Human review complete

Project owner / exact revisions

Human-reviewed revisions
50/50
Published HTML versions
50
Text threshold met
23/50
Rendered pages compared
373

Human-reviewed does not mean independently audited, certified, or declared conformant with WCAG or PDF/UA.

What the corpus shows

Automated scores describe risk. Human review closes the loop.

The two automated measurements move together, but they are not interchangeable. The project-owner review resolves the final route and publication decision for each exact conversion.

Automated text fidelity compared with rendered visual coverageFifty points show each benchmark document. Coral points retained an automated visual attention signal. Dark points did not.0025255050757510010090% text threshold01-01: New Requirements on Accessibility of Web Content Mobile Apps and Kiosks. Text 98.98%. Visual 91.68%.01-02: Federal Records Centers Digital Delivery FAQ. Text 97.61%. Visual 75.23%.01-03: Civics and Citizenship Toolkit Quick Start Guide. Text 98.86%. Visual 77.08%.01-04: USCIS Welcomes Refugees and Asylees. Text 96.83%. Visual 72.53%.01-05: 2025 MEP State Competition FAQs. Text 95.58%. Visual 69.2%.02-01: The Nation's Fiscal Health 2025. Text 86.95%. Visual 82.81%.02-02: The NIST Cybersecurity Framework (CSF) 2.0. Text 90.63%. Visual 76.73%.02-03: Federal Rulemaking Potential Effects of PAYGO Legislation. Text 95.27%. Visual 83.15%.02-04: COVID-19 Key Indicators and Federal Response 2025 Update. Text 89.68%. Visual 76.02%.02-05: GAO Performance and Accountability Report FY2024. Text 91.79%. Visual 59.3%.03-01: Legionnaires Disease General Fact Sheet. Text 94.79%. Visual 64.08%.03-02: CDC Global Health Overview Fact Sheet. Text 93.36%. Visual 83.5%.03-03: Measles Is Not Just a Little Rash. Text 95.75%. Visual 68.89%.03-04: The Urgent Threat of TB Drug Resistance. Text 95.26%. Visual 63.46%.03-05: Tuberculosis in Children and Adolescents. Text 90.35%. Visual 55.94%.04-01: SSA Alternative Text Guide. Text 78.12%. Visual 72.88%.04-02: Executive Guide to Federal IT Accessibility. Text 96.64%. Visual 73.75%.04-03: PDF Tagging Guide. Text 62.41%. Visual 80.48%.04-04: How to Identify Section 508 Errors in HTML. Text 79.49%. Visual 75.12%.04-05: Welcome to the United States Orientation Brochure. Text 98.9%. Visual 76.7%.05-01: Water Availability in the United States Key Findings. Text 89.42%. Visual 82.94%.05-02: CASTNET Quality Assurance Report Third Quarter 2025. Text 63.9%. Visual 72.66%.05-03: NASA Aerospace Safety Advisory Panel Annual Report 2025. Text 95.84%. Visual 77.25%.05-04: Births in the United States 2024. Text 84.82%. Visual 80.16%.05-05: Birth Cohort Geographic Mobility in the United States. Text 80.37%. Visual 79.6%.06-01: Unintentional Fall Deaths in Adults Age 65 and Older. Text 73.99%. Visual 78.18%.06-02: Perinatal Mortality in the United States 2022 and 2023. Text 76.74%. Visual 79.59%.06-03: Emergency Department Visits for Tooth Disorders. Text 88.31%. Visual 84.8%.06-04: Prevalence of Cardiovascular Disease Risk Factors in Adults. Text 85.06%. Visual 83.31%.06-05: Changes in Timing of Prenatal Care Initiation. Text 75.6%. Visual 81.69%.07-01: Q Fever Fact Sheet Arabic. Text 60.74%. Visual 45.62%.07-02: Q Fever Fact Sheet French. Text 93.3%. Visual 87.6%.07-03: Q Fever Fact Sheet Vietnamese. Text 92.92%. Visual 86.98%.07-04: Q Fever Fact Sheet Simplified Chinese. Text 93.4%. Visual 87.31%.07-05: Q Fever Fact Sheet Spanish. Text 92.49%. Visual 88.1%.08-01: Income in the United States 2024. Text 54.96%. Visual 52.65%.08-02: Health Insurance Coverage by State 2023 to 2024. Text 70.63%. Visual 73.64%.08-03: America's Families and Living Arrangements 2022. Text 73.7%. Visual 69.26%.08-04: Health United States 2000. Text 62.36%. Visual 54.83%.08-05: EPA Third Quarter 2025 FOIA Log. Text 91.51%. Visual 88.19%.09-01: Form W-9 Request for Taxpayer Identification Number. Text 95.57%. Visual 73.28%.09-02: Form W-4 Employee Withholding Certificate. Text 81.86%. Visual 63.02%.09-03: Form I-9 Employment Eligibility Verification. Text 77.61%. Visual 60.56%.09-04: OF 306 Declaration for Federal Employment. Text 95.4%. Visual 70.2%.09-05: SF 15 Application for Veterans Preference. Text 88.91%. Visual 71.96%.10-01: JFK Record 104-10181-10007. Text 37.52%. Visual 47.29%.10-02: JFK Record docid-32313670-1. Text 78.99%. Visual 57.02%.10-03: JFK Record docid-32313410. Text 62.22%. Visual 49.62%.10-04: JFK Record 157-10011-10024. Text 49.58%. Visual 58.85%.10-05: JFK Record docid-32339281. Text 40.14%. Visual 51.56%.Automated text fidelity (%)Rendered visual coverage (%)
Visual signal retainedNo visual trigger

0.60

Correlation between automated text fidelity and rendered visual coverage. It is directional evidence, not equivalence.

23/50

Conversions met the 90% automated text-fidelity threshold before human review.

41/50

Documents retained at least one visual attention signal for inspection and reproducibility.

50/50

Exact source and conversion revisions have a completed project-owner review record.

Class matters

Simple linear and multi-column documents reached the automated text threshold most often. Scans and simple-table files did not, despite stronger visual results in parts of the set.

Signals stay visible

A completed human review does not erase low automated scores or preflight flags. They remain attached to the record so the evidence can be inspected and reproduced.

Routes are explicit

Every record publishes a final route: 10 HTML-first, 25 automation plus review, and 15 specialist conversions.

Evidence method

Four evidence layers, kept separate

The corpus is a purposive diversity sample, not a popularity ranking or random sample. Each layer answers a different question and retains its own method, date, and limitations.

  1. 01

    Source acquisition

    Official publisher URLs, malware scanning, PDF signature and metadata inspection, and frozen SHA-256 evidence. Source PDFs are linked, not republished.

  2. 02

    Automated text fidelity

    Normalized token and ordered five-token comparison. 23 of 50 met the 90% threshold. This score does not evaluate semantics or conformance.

  3. 03

    Rendered visual coverage

    Poppler and fixed Chromium rendering across 373 pages. Coverage has no pass threshold and can be lowered by valid reflow.

  4. 04

    Human review and route decision

    Project-owner source-grounded review of exact hashes, meaningful blocks, applicable semantics, keyboard and assistive-technology behavior, and final routing.

Read the full methodologyDownload the blank review template

10 classes / 50 records

Inspect every record

Search and combine filters across the full corpus. Filter state and result pages are stored in the URL so a view can be shared.

50 of 50 records

Compare all ten document classes
Class-level automated text threshold and rendered visual coverage
Document classText thresholdMean visual coverageVisual signals
Simple linear5/5 (100%)77.14%4/5
Long reports3/5 (60%)75.6%5/5
Multi-column5/5 (100%)67.17%5/5
Lists and links2/5 (40%)75.79%5/5
Figures and maps1/5 (20%)78.52%4/5
Simple tables0/5 (0%)81.51%3/5
Multilingual4/5 (80%)79.12%1/5
Complex tables1/5 (20%)67.71%4/5
Forms2/5 (40%)67.8%5/5
Scans0/5 (0%)52.87%5/5

Text threshold is the share at or above 90% automated source fidelity. Visual coverage is diagnostic and has no pass threshold.

01-01Simple linearHHSEnglish

New Requirements on Accessibility of Web Content Mobile Apps and Kiosks

fact sheet / civil rights

Open reviewed HTMLOfficial PDF
Text
98.98%
Threshold met
Visual
91.68%
No trigger
Preflight
0
flags retained
Human
Reviewed
html first

99.33% token F1 / 98.74% ordered five-token F1 / 3 rendered source pages

No automated attention signal

01-02Simple linearNational ArchivesEnglish

Federal Records Centers Digital Delivery FAQ

faq / records management

Open reviewed HTMLOfficial PDF
Text
97.61%
Threshold met
Visual
75.23%
Signal retained
Preflight
0
flags retained
Human
Reviewed
html first

99.44% token F1 / 96.39% ordered five-token F1 / 2 rendered source pages

Inspect 1 automated signal
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

01-03Simple linearUSCISEnglish

Civics and Citizenship Toolkit Quick Start Guide

quick start guide / immigration

Open reviewed HTMLOfficial PDF
Text
98.86%
Threshold met
Visual
77.08%
Signal retained
Preflight
0
flags retained
Human
Reviewed
html first

99.14% token F1 / 98.67% ordered five-token F1 / 4 rendered source pages

Inspect 1 automated signal
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

01-04Simple linearUSCISEnglish

USCIS Welcomes Refugees and Asylees

orientation guide / immigration

Open reviewed HTMLOfficial PDF
Text
96.83%
Threshold met
Visual
72.53%
Signal retained
Preflight
0
flags retained
Human
Reviewed
html first

98.68% token F1 / 95.59% ordered five-token F1 / 12 rendered source pages

Inspect 1 automated signal
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

01-05Simple linearNISTEnglish

2025 MEP State Competition FAQs

faq / economic development

Open reviewed HTMLOfficial PDF
Text
95.58%
Threshold met
Visual
69.2%
Signal retained
Preflight
0
flags retained
Human
Reviewed
html first

98.4% token F1 / 93.71% ordered five-token F1 / 6 rendered source pages

Inspect 2 automated signals
  • Fewer than 70% of sampled source-page text sequences aligned to the HTML render.
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

02-01Long reportsGAOEnglish

The Nation's Fiscal Health 2025

annual report / public finance

Open reviewed HTMLOfficial PDFPublisher context
Text
86.95%
Below threshold
Visual
82.81%
Signal retained
Preflight
0
flags retained
Human
Reviewed
automation review

89.89% token F1 / 84.99% ordered five-token F1 / 12 rendered source pages

Inspect 2 automated signals
  • Automated source fidelity is 86.95%, below the 90% text threshold.
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

02-02Long reportsNISTEnglish

The NIST Cybersecurity Framework (CSF) 2.0

framework / cybersecurity

Open reviewed HTMLOfficial PDFPublisher context
Text
90.63%
Threshold met
Visual
76.73%
Signal retained
Preflight
0
flags retained
Human
Reviewed
automation review

95.63% token F1 / 87.29% ordered five-token F1 / 12 rendered source pages

Inspect 1 automated signal
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

02-03Long reportsGAOEnglish

Federal Rulemaking Potential Effects of PAYGO Legislation

audit report / government operations

Open reviewed HTMLOfficial PDFPublisher context
Text
95.27%
Threshold met
Visual
83.15%
Signal retained
Preflight
0
flags retained
Human
Reviewed
automation review

98.11% token F1 / 93.38% ordered five-token F1 / 12 rendered source pages

Inspect 1 automated signal
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

02-04Long reportsGAOEnglish

COVID-19 Key Indicators and Federal Response 2025 Update

audit report / public health

Open reviewed HTMLOfficial PDFPublisher context
Text
89.68%
Below threshold
Visual
76.02%
Signal retained
Preflight
0
flags retained
Human
Reviewed
automation review

94.13% token F1 / 86.72% ordered five-token F1 / 12 rendered source pages

Inspect 2 automated signals
  • Automated source fidelity is 89.68%, below the 90% text threshold.
  • At least one sampled PDF page with image objects has no aligned HTML visual element.

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

02-05Long reportsGAOEnglish

GAO Performance and Accountability Report FY2024

annual report / government operations

Open reviewed HTMLOfficial PDFPublisher context
Text
91.79%
Threshold met
Visual
59.3%
Signal retained
Preflight
1
flags retained
Human
Reviewed
automation review

98.52% token F1 / 87.31% ordered five-token F1 / 12 rendered source pages

Inspect 3 automated signals
  • At least one sampled PDF page with image objects has no aligned HTML visual element.
  • Planned visual representations are absent: semantic tables.
  • Table representation: Tables are planned, but no semantic TABLE element is present. The fragment contains 0 layout-text block(s).

These deterministic signals remain visible for reproducibility. The project-owner review for this exact revision is complete.

The 90% text threshold and visual-coverage score are automated diagnostics. Human review is recorded separately for every exact source and HTML revision and does not assert certification or conformance.

Publication record

A release you can cite and audit

Version 2026-08-06. Reviewed by the project owner on August 6, 2026. Every document record contains its exact source and conversion digest.

Citation

DocAccessible. “PDF-to-HTML conversion benchmark corpus.” Version 2026-08-06. 50 public-source documents and reviewed HTML conversions. Accessed [date].

https://docaccessible.com/benchmarks/pdf-to-html

Review boundary

Review status is project-owner attestation, not independent certification, a legal opinion on third-party source rights, or a WCAG/PDF-UA conformance declaration.

Corpus hashes

Source corpus SHA-256
db2e210a7416bbe0bf5483205de31dbd0bee552cdba631c28fb74b9e4abd3854
Conversion corpus SHA-256
1b4cb1fc8a47bf2f58b0780da71a0289e0bb5721ac28c0ad4f4024fd8744f9ad

Download the evidence

Source PDFs remain documents of record with their publishers and are not redistributed here. DocAccessible publishes benchmark measurements, exact-version review records, and reviewed HTML routes for inspection. Publication review is complete for all 50 records; 6 sources carry public-domain evidence and 44 carry project-owner publication-review evidence.

Start with the record, then inspect the evidence

Each detail page keeps the publisher source, reviewed HTML, automated diagnostics, human-review dimensions, and exact version hashes together.

Choose a record