PDF OCR quality checker

This tool audits text that already exists in the file; it never performs OCR. Each page is classified by what pdf.js extracts: a page with no letter words is a scan or a blank, and a page whose glyphs are drawn invisible (text render mode 3) carries a hidden OCR layer. Pages rate poor at more than 5 percent letter/digit mix like vi11age or c1erk, more than 5 percent garbage symbols such as replacement and private-use characters, or a dictionary share under 70 percent when at least 20 words are checkable; between those marks and 1 percent, or a share under 85 percent, they rate fair. The dictionary holds 25,322 common English words with plural trimming, and capitalized unknown words count as names rather than damage, so a short English page under 20 words gets no verdict. Two faults stay invisible to every check: real word substitutions such as form for from, and born-digital pages whose fonts ship no Unicode map, which is a font problem rather than an OCR one. Everything runs in your tab and nothing is uploaded.