For the complete documentation index, see llms.txt. This page is also available as Markdown.
Extraction schema
An extraction schema is the set of fields you want pulled out of every document — the column definitions of the table you are building.
Shape
json
{ "object": "extraction_schema", "id": "exsc_2hq7bnv6ld", "name": "RCT screening",
"fields": [
{ "name": "sample_size", "type": "integer", "description": "Total participants enrolled" },
{ "name": "intervention", "type": "string", "description": "The treatment being tested" },
{ "name": "is_randomized", "type": "boolean", "description": "Was allocation randomized?" }
] }Field types
string · integer · number · boolean · date · enum · string[]
Nested objects are not supported in v1. A schema may hold up to 50 fields.
Writing good descriptions
The description is the instruction the extractor follows — it matters more than the field name.
Say what counts and what does not, and name the unit.
Every cell is cited
Extraction rows return a Citation per value, so a reviewer can verify any cell against
the source page. A field the extractor could not find returns null with no citation — never a guess.
Related
Last updated