Find the first clue in one claim
Import the scenario and distinguish the source document from intake metadata.
Question for this chapter
What is already known before a claim is judged, and what still has to be read from the document?
Why this matters now
If a value's origin is unknown, finding an error later becomes difficult. Establish the boundary between intake information and source-document content before reviewing the structured result.
Try it
docai.zip Download(43 KB)Select Collections in the Portal's left sidebar. On the collections page, choose More (⋯) → Import in the upper-right and upload the ZIP.
Import registers the assets defined in manifest.json. Open Deep dive below to inspect their types
and contents.
When import completes, open claim_documents in the processed collection and find the row whose
claim_id is CLM-2025-006.
Success looks like this
The collection tree contains inbox and processed, and the CLM-2025-006 row contains this intake
information:
| Value to inspect | Current value | What it tells you |
|---|---|---|
document_key | claims/inbox/clm_006.pdf | Location of the source PDF |
claimant_name | Frank Wilson | Claimant |
policy_number | POL-60444 | Insurance policy |
received_at | 2025-03-01 11:15:00 | Intake time |
source_channel | portal | Intake channel |
Interpret the result
This row is metadata known when the document arrived, not the contents of the PDF.
- Source document: the claim amount and diagnosis written in
clm_006.pdf. - Intake metadata: the document location, claimant, policy number, and time known before processing.
- Structured result: values that the next pipeline combines from the document and metadata.
Deep dive — complete imported asset inventory
| Asset type | Count | Contents |
|---|---|---|
| Collections | 2 | inbox, processed |
| Datasets | 3 | claim_documents, extracted_fields, decision_log |
| Code | 3 | Document structuring, claim decision, ontology materialization |
| Pipelines | 3 | Claim extraction, decision flow, ontology materialization |
| Ontology | 3 entities / 2 relations | Document, claimant, policy / filed-by and covered-by relations |
| Knowledge | 1 | claims_handbook |
| Dashboard | 1 | claims_processing |
inbox is an empty collection reserved for future receiving assets. All practice assets live in
processed. Across the ten samples, channels are email 4, portal 4, and fax 2.
Next decision
Intake information does not reveal the amount and diagnosis in the PDF. Next, structure the document and decide whether its output is trustworthy enough for an automated decision.