Research program
Seven studies on the questions nobody benchmarks: whether headings, tables, reading order, links, and equations keep their meaning when PDFs become HTML, and what real-world tagging does to assistive technology along the way.
Every number we publish traces to recorded evidence with a stated method and version. Findings that have not been measured are labeled planned, never implied.
Our own conversion pipeline is scored with the same frozen methods as everything else we measure, and its failures are published with the same detail as its wins.
Corpora are acquired from official public URLs with recorded hashes and malware scans. Methods are versioned. When a method changes, the change is disclosed.
7 studies. Statuses are factual, not aspirational.
01 · Flagship benchmark
Collecting evidenceA public benchmark for what no document parser is measured on today: whether a converted document's headings, tables, reading order, links, and alternatives are faithful and usable with assistive technology.
02 · Field study
Collecting evidenceA prevalence study of the tagging mistakes that make technically tagged PDFs unusable in practice, measured across public government documents.
03 · Recurring index
PlannedA recurring, sector-stratified measurement of how ready state and local government documents are for the April 2027 and April 2028 web accessibility deadlines.
04 · Measurement study
Collecting evidenceThe first measurement of whether hyperlinks keep their text and their destinations when PDFs become HTML, across the three link classes that behave completely differently.
05 · Data study
Collecting evidenceHow much of a government PDF is repeated running headers, footers, and page numbers, and what that repetition costs a screen reader user in listening time.
06 · Replication and extension
PlannedA checkpoint-level measurement of where PDF/UA validators diverge on the same real files, extending published research that found disagreement on half of documents tested.
07 · Measurement study
PlannedEvery major browser now renders MathML natively, yet equations remain the content most likely to vanish in PDF conversion. This study measures the gap.