[Source](https://sheaf.us/scoreboard-split-classify-extract.html)

Scoreboard

# Split, Classify and Extract: text models compared

How each text model cuts a file into documents, names each one and reads its values, with what it costs.

Its key points, its tables, how it was scored and its log load from [/api/public/scoreboards/split-classify-extract](/api/public/scoreboards/split-classify-extract).

## Test documents

The test sets are not published: Claude wrote their ground truth from the pages, and no person has checked it yet. The documents are public filings: SpaceX's [prospectus](https://www.sec.gov/Archives/edgar/data/1181412/000162828026042639/spaceexplorationtechnologi.htm) (37 of its pages), [second-quarter earnings release](https://s21.q4cdn.com/184289198/files/doc_financials/2026/q2/SpaceX-Reports-Second-Quarter-2026-Results.pdf) and [bond offering press release](https://www.sec.gov/Archives/edgar/data/1181412/000162828026044489/exhibit991-8xk.htm), and, in the states' case against Paramount, the [complaint](https://storage.courtlistener.com/recap/gov.uscourts.cand.474157/gov.uscourts.cand.474157.1.0_2.pdf) (31 of its pages), the [consent decree](https://storage.courtlistener.com/recap/gov.uscourts.cand.474157/gov.uscourts.cand.474157.267.0_3.pdf) (22 of its pages) and the [order entering it](https://storage.courtlistener.com/recap/gov.uscourts.cand.474157/gov.uscourts.cand.474157.266.0_1.pdf); and, in Musk v. Altman, the orders on a [preliminary injunction](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.121.0_1.pdf) and on [summary judgment](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.390.0.pdf), the [jury instructions](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.574.0.pdf), the [verdict form](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.572.0.pdf), the [post-trial order](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.580.0_1.pdf) and ten trial exhibits, filed as docket entry 601 ([the first](https://storage.courtlistener.com/recap/gov.uscourts.cand.433688/gov.uscourts.cand.433688.601.1.pdf)).

## 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. |