No page image
Layout elements whose box matches the GT but whose text content differs.
Each item is a ParseBench rule where the ground truth (ParseBench GT) and an independent transcription of the page (Model parse) disagree. Neither side is assumed correct: compare both against the page image.
| ↓ ↑ / j k | next / previous item |
| → ← | next / previous page |
| / | search by id, file name or text |
| v | switch the comparison view |
| + - 0 · Ctrl+scroll | zoom the page · 0 shows the whole page |
| c · o | copy the id · open the row on Hugging Face |
| ? | this help |
Each item shows its ParseBench id (the id column of
llamaindex/ParseBench).
Click it to copy; view on Hugging Face opens that row in the dataset viewer.
The browser URL (…#<id>) always points at the current item, so you can paste it to refer to an issue.
Single offline HTML file; nothing is uploaded. View preferences are kept in your browser.