Types
Choose the type your code expects
Use numbers for amounts, booleans for visible checkbox states, and strings for identifiers. Deterministic normalization is easier when the type is explicit.
Guide / Schema design
Design a schema your workflow can trust.
A dependable extraction schema names exactly the fields your system consumes, in the types it expects. Start narrow, describe ambiguous fields, and require only what must always be present.
Start here
Name the fields your code already uses.
Routeser requests carry user-defined field names, types, descriptions, and required flags. Mirror the vocabulary of the system that will store the result so downstream code needs no translation layer.
Resist a wide schema. A focused set of fields is easier to validate, cheaper to review, and more predictable across documents than a catch-all object.
Descriptions
Explain where a value appears
Field descriptions tell the extractor which visible label or region a value comes from. They are the cheapest way to disambiguate lookalike fields.
Required
Require only the non-negotiable
Mark a field required only when its absence should be treated as a problem. Optional fields let the extractor preserve uncertainty instead of inventing a value.
Before you submit
A short pre-flight for a new schema.
- List the exact fields the downstream system consumes.
- Assign each field a concrete JSON type.
- Add a one-line description for any field with an ambiguous label.
- Mark required flags conservatively and validate the terminal result before acting on it.
Related
See a schema in a real workflow.
Questions, answered plainly
Frequently asked questions
Does a larger schema always extract more?
No. A narrow schema is easier to validate and review. Add fields when a downstream consumer needs them, not speculatively.
When should a field be required?
Require a field only when its absence should fail or flag the workflow. Optional fields let the extractor preserve genuine uncertainty.
Do descriptions change the output type?
No. Descriptions guide where a value is found; the field type defines what your system receives. Keep both explicit.
Next step
Define your first schema
Open the console to create a schema and evaluate it against your own documents in an authenticated workspace.