Extract (extract)
Extracts the fields of a JSON Schema from a parsed document or segment, with a citation (page, quote, boxes) and a calibrated confidence per field. null means not found. See Extraction harness.
Group: understand · Inputs: ParsedDoc, Segment · Outputs: Extraction
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
field_routes |
object | Per-field model routing (T42): group_id → {tier, model, fields}. One extraction pass per distinct route; fields in no group use tier/model_policy. |
||
learning_snapshot_id |
string or null | ExampleSnapshot (T42) whose learned examples enter the prompt as data. |
||
lookups |
array of string | |||
model_policy |
string or NodeModelPolicy or null | A model spec (provider:alias, a pin) or a policy fragment (T49) that tightens the organization and project model policy for this node. |
||
prompt_addendum |
string or null | |||
schema_id |
string | yes | "schema" |
Key of the extraction schema inside the version's schemas. |
tier |
"fast" | "standard" | "agentic" |
"standard" |