Decisions

A decision is the outcome of a document verification. Decisions can be made automatically by the validation pipeline or manually by a reviewer through the API or dashboard.

Decision Statuses

Status	Value	API Code	Description
Pending	0	Default	Decision has not been made yet
Approved	1	APPROVED	Document passed all validation checks
Rejected	2	REJECTED	Document failed one or more validation checks
Double Check	3	DOUBLE_CHECK	Document requires manual review
Canceled	4	CANCELED	Document was canceled before a decision

How Decisions Work

Automatic Decisions

When a verification is submitted, the system:

1. Runs the extraction pipeline on all documents
2. Applies validation rules (comparing extracted data to expected data)
3. Evaluates all rule results to determine the decision

The decision logic follows these rules:

• Approved — All rules pass
• Rejected — One or more rules fail with error-level severity
• Double Check — Rules produce warnings but no errors (ambiguous results)

Manual Decisions

You can manually make decisions through the API on individual documents:

• Approve — Accept the document as valid
• Reject — Reject the document with a reason
• Double Check — Flag for additional review
• Cancel — Remove the document from consideration

Manual decisions override automated results.

Decision Cascade

When a decision is made on a document, it may cascade to the verification level:

1. Document decision is created
2. Document status is updated to match
3. If all documents have a final decision (approved or rejected), the verification status is updated:
   • All documents approved → Verification Approved
   • Any document rejected → Verification Rejected
   • Any document in double check → Verification Double Check

Validation Workflow Tracking

Each decision includes a validation_workflow field that documents how the decision was reached:

Workflow	Description
Automatic	Decision made by the automated system
Manual by {email}	Decision made by a reviewer via the API

Best Practices

• Review extracted data before making a manual decision to ensure the comparison is accurate
• Add meaningful comments so the end user understands what needs correction
• Use double check as an intermediate state when you need to escalate to a supervisor
• Monitor automated decision quality to identify patterns in false positives or false negatives
• Correct extracted data before making a decision if extraction pipeline results are inaccurate