The visual workbench
Upload a sample document, refine the fields and rules the platform inferred, and see live extraction results in the same dialog. Save when the result matches what your team needs.
Audience: builders
Why you're here
Your AP team gets the same Acme Logistics invoice every week, and the totals never quite add up the way the agent expects. Your shipping desk wants to extract carrier, weight, and consignee out of a bill of lading where the carrier moves the fields around per region. Your quality team needs pass/fail readings out of a scanned certificate. In every case the right path is the same: drop the document the team actually receives, refine the template against what came out, and try it again until the extraction is honest.
The workbench keeps the sample, the Schema, and the extraction result in one dialog so you can iterate in seconds instead of trial-and-error against production.
Open the workbench
Open your Project. Click the Context tab. Pick Schemas. From there you reach the workbench three ways:
- First Schema. Drop a sample on the Pull structured data out of your team's documents panel. The platform drafts the Schema, then the workbench opens on the draft.
- New Schema by hand. Click New schema in the section header. The workbench opens on a blank draft.
- Open an existing Schema. Click any row in the catalog. The workbench opens on that Schema in view mode; click into the fields, prompt, or rules to edit.
Draft a Schema from a sample
Drop a real document
In the empty state or the Drop a sample to draft a schema strip above the catalog, drop a PDF, XLSX, DOCX, or image. The platform accepts one file at a time for inference. Real documents from your team work better than vendor-supplied examples.
Watch the analysis dock
A card appears above the catalog with four steps: Reading your document, Looking for known patterns, Drafting a schema, Finalizing the schema. Each step shows elapsed seconds while it runs and "Done" when it lands. The work runs server-side; you can leave the page and the analysis keeps going, then reappears when you return.
Open the draft
When the card flips to Your schema is ready to review, click Open. The workbench opens with fields, the prompt, and any inferred rules pre-filled from the document.
What the workbench shows
The workbench is a full-screen dialog with three columns:
- Samples (left). The document you dropped, with a thumbnail preview. PDFs render inline; spreadsheets show a placeholder icon; images preview as images. Click Upload to add more samples for the same Schema.
- Editor (center). Fields, the prompt (extraction guidance), and rules. Each block has a Simple view and an Advanced view; Advanced is the raw JSON.
- Sample Results (right). The extraction result when you click Try on sample. Resizable, with a maximize button. Drag the handle on its left edge to widen or narrow.
The top bar carries: a breadcrumb, the scope picker (Project, Personal, Team, or Workspace), Try on sample, and Save Draft.
Refine the fields
Open Fields in the editor. Each row shows the field name, its type, whether it is required, and a description. Click a row to edit inline. Use + Add field to extend the Schema. The Advanced toggle shows the raw extraction_schema JSON, which is a JSON Schema object with a properties map.
The Prompt block is plain-language guidance for how to read this document type. Example: "Treat negative amounts as credits." or "Line items appear after the footer when the vendor splits the invoice."
Try the Schema on the sample
Click Try on sample in the top bar. The platform runs an extraction against the document you have selected in the Samples column, using the current draft of your Schema. While it runs, the button reads Trying with an elapsed counter.
When the result lands, the right column shows:
- A headline: "Pulled X of Y fields."
- The file name, page count, and field count.
- A Rules block (if rules are defined) showing each sentence with a pass/fail status.
- An Extracted fields tree with each value, its type, and a page reference chip (for example,
p2) showing the page the value came from. - A disclosure for the raw response JSON.
The result is live, not cached. Edit a field name or a rule, click Try on sample again, and the result panel updates against the current draft.
Save the draft
Click Save Draft in the top bar. The Schema is written to your Project's volume at .blockparty/schemas/<scope>/<slug>.json. The toast tells you which: Saved as Draft when you have direct write authority for the scope, Submitted for review when your scope only allows candidate proposals.
For how candidates flow through review and how to use a saved Schema in an App contract, see Schemas.