🔬 Research project

Atlas — Measuring Real Editability, Not Just Looks

A PDF-to-Word converter can produce a page that looks perfect and is still unusable — text trapped in boxes, tables that aren't tables, paragraphs in the wrong order. Atlas is our attempt to score that honestly.

What Atlas actually measures

Most conversion quality checks are visual: does the exported page look like the source PDF? Atlas asks a different question: if you had to actually edit this document, how much work would it take?

The underlying metric — the Editing Readiness Index (ERI) — inspects the actual document structure (a .docx is a zip of XML) and checks things a rendered screenshot can't show you:

Is the text in a real, editable paragraph — or trapped in a fixed text box?
Is a table a real table you can resize — or a flattened image?
Does reading order top-to-bottom actually make sense?
Is any text visibly garbled, reversed, or duplicated?

Where this stands today

The scoring engine exists and is tested against a hand-built corpus of documents with known, injected defects. What it is not yet: independently validated against real human judgment.

We're currently running that validation — recruiting people to rate anonymized converted documents by hand, without knowing the score in advance, and comparing their ratings to Atlas's. Until that's done, we're not publishing converter comparisons or accuracy claims. This page will be updated with real results, not projected ones.

🙋 Help us validate Atlas

We're looking for a few volunteers to rate a small set of anonymized documents by hand — open each one in Word, LibreOffice, or Google Docs, do a short edit, and rate how much rework it needs. About 15-20 minutes total. Click below, tell us your email and what language you're comfortable working in, and we'll send you the sample set and instructions.