Sheaf.us
Evals

OCR scoreboards

Evals of reading a page: OCR engines, their settings, and what is built on their output.

The scoreboards

Scoreboard What it measures Created Updated
OCR accuracy Which checked values each OCR engine reads correctly, from the PDF and from the page image, with time and cost. Oct 8 2026 —
Grounding Where each grounding option puts a checked value: its own bounding box, its row or line, or its block. Oct 7 2026 —
Table formats How OCR engines write a table: HTML, Markdown or other, and whether merged table cells survive. Oct 3 2026 —
Form blocks by engine The block types Mistral, Landing AI and Reducto have for the parts of a form. Oct 3 2026 —
Mistral OCR settings What Mistral OCR 4.1's page annotation corrects on form pages, and what it costs. Oct 3 2026 —
Chinese OCR Chinese OCR models: precision, speed, hosted price, bounding boxes and input. Oct 3 2026 —
Form review variations The Level 2 form review on handwritten pages, by model, picture and scope. Oct 3 2026 —
Forms · Level 2 Parse Second passes that correct Mistral OCR's reading of a form page. Oct 3 2026 —
Forms · Level 1 OCR engines on the same form pages, scored against ground truth. Oct 3 2026 —

Recent changes

Glossary

Term Definition
OCR Optical character recognition: reading the text, tables and marks of a page from its image or PDF.
engine An OCR product or model that reads a page and returns its text, such as Mistral OCR, Landing AI or Reducto.
block One part of a page as an engine returns it: a paragraph, a table, a heading or a figure.
bounding box The rectangle that marks where something is printed on the page.
table cell One value of a table, where a row and a column meet.
form field A label on a form and the value written for it.
grounding Linking each value of a reading to its bounding box on the page.
ground truth The correct answers for a page, one fact per value, checkbox, choice or signature. Every reading is scored against them.
test set The pages a scoreboard is measured on, with their ground truth.
grader The program or model that compares a reading with the ground truth and counts what is right.

Test us with your documents. Send the bundle that currently ruins someone's afternoon, exactly as it arrived.