Test a document in a pipeline
Upload a document to a pipeline's Test tab, run it against a draft or published version, and review the extracted data field by field — with confidence and on-page coordinates — before you deploy or deliver.
The Test tab runs a real document through a pipeline and shows you every stage's output, so you can check a change before you publish it or wire it to a destination. A test run never delivers anywhere — it is for looking, not shipping.
Pipeline builders. You can test any version you can open, including an unpublished draft.
1. Open the Test tab
Open a pipeline and switch to Test. The tab has a version selector, an upload control, and a Run button, with a per-stage progress strip across the top.
The Test tab before a run. The stage strip on the right shows the pipeline's stages in order; the two panels fill in once a document is uploaded and run.
2. Choose a version
Pick the version to test from the selector. A draft lets you check unpublished edits; a published version reproduces exactly what production is running. The run is stamped to the version you pick, so you can compare a draft against the live version deliberately.
3. Upload your documents
Upload one document, or a batch. The left panel previews the document you are looking at; when you upload several, a switcher lets you move between them and each is processed independently.
4. Run
Choose Run. The stage strip advances as each stage completes — Initialization, Parse, Split / Classify, Extract, Post processing — and the run ends in one of a few outcomes: it completes cleanly, it needs review, it is awaiting review, or it failed or was blocked. The outcome tells you whether the pipeline would have shipped this document as-is.
5. Review the extracted data
The result is a split view: the document on the left with a bounding-box overlay marking where each value was read, and the extracted fields on the right, each with its confidence. Hovering a field highlights its box on the page. The results are grouped into Fields, Tables, Figures, and Selections, with a Raw JSON view, and per-stage panels let you see what Parse read, what Classify decided, and what Extract produced — so when a value is wrong you can tell which stage to fix.
A completed run. The document and its field overlays sit on the left; the extracted fields, each with a confidence score, on the right. The overlay is the fastest way to confirm a value was read from the right place on the page.
6. Verify or correct
If the run needs review, the document carries the same exceptions it would in production, and the review happens in the Reviews workspace rather than here — see Correct a document in review for the claim-correct- submit loop. Use the Test tab to find the problem; use Reviews to work it.
Test runs never deliver
A test run reads and reports, but it never writes to a connector or fires a destination — the same guarantee the API gives a submission made with run_kind=test (see Submit a job). To send real output somewhere, publish the pipeline and submit a production job, or wire delivery with Deliver to a destination.