> ## Documentation Index
> Fetch the complete documentation index at: https://docs.veridox.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Risk Scoring

> Understanding and managing AI-powered risk scores for document verification

## Overview

Veridox provides quantified risk assessments for every document analysed, enabling automated decision-making in verification workflows. This guide explains how risk scoring works, how to interpret scores, and how to implement risk-based automation.

## Risk Score Levels

Veridox uses a four-tier risk scoring system:

<CardGroup cols={2}>
  <Card title="Informational" icon="circle-info" color="#3B82F6">
    No significant concerns identified. Standard documentation with minor or expected variations.
  </Card>

  <Card title="Low" icon="circle-check" color="#10B981">
    Minor anomalies detected that don't indicate fraud. May require light review but generally acceptable.
  </Card>

  <Card title="Medium" icon="triangle-exclamation" color="#F59E0B">
    Moderate risk factors present. Requires additional verification or manual review before proceeding.
  </Card>

  <Card title="High" icon="shield-exclamation" color="#EF4444">
    Significant fraud indicators found. Document shows clear signs of tampering, fabrication, or inauthenticity.
  </Card>
</CardGroup>

## How Risk Scores Are Generated

Risk scores are derived from comprehensive forensic analysis across multiple engines, including metadata examination, compression inconsistency detection (ELA), noise pattern analysis, and AI-powered context understanding.

The **overall risk score** is determined by evaluating all component assessments, with higher-severity findings taking precedence.

## Confirming or Overriding Risk Scores

As an expert reviewer, you can confirm AI assessments or override them based on additional context using the [Confirm Risk Score](../api-reference/endpoints/files-risk-confirm) endpoint.

<Warning>
  **Audit Trail**: All risk score confirmations and overrides are logged with timestamps and user information for compliance and audit purposes.
</Warning>

### When to Override Risk Scores

Consider overriding when:

* **Additional context available**: External information not visible to AI (customer history, supporting documentation).
* **Domain expertise**: Specialised knowledge about document types or regional variations.
* **False positives**: AI flagged expected variations as anomalies.

**Best practice**: Always provide a detailed confirmation reason when overriding scores to maintain a clear audit trail.

## Best Practices

### Score Interpretation

<Check>**Consider context** - Risk scores are one input; combine them with your own business rules and domain knowledge.</Check>
<Check>**Review findings** - Don't rely solely on the overall score; examine specific findings and evidence in the full analysis report.</Check>
<Check>**Understand limitations** - AI is a powerful tool but requires human expertise for highly sophisticated or unique fraud cases.</Check>

### Compliance and Governance

<Check>**Document decisions** - Always provide confirmation reasons for a robust audit trail.</Check>
<Check>**Security** - Remember that users can only access and confirm risk scores for files they have created.</Check>
<Check>**Escalation paths** - Define clear escalation procedures for documents with "High" or "Medium" risk scores.</Check>

## Related Resources

<CardGroup cols={2}>
  <Card title="Confirm Risk Scores API" icon="shield-check" href="/api-reference/endpoints/files-risk-confirm">
    API reference for confirming or overriding risk assessments
  </Card>

  <Card title="File Analysis API" icon="magnifying-glass" href="/api-reference/endpoints/files-analysis-get">
    Retrieve detailed forensic analysis results
  </Card>

  <Card title="Document Analysis Guide" icon="file-magnifying-glass" href="/guides/document-analysis">
    Complete document analysis workflow
  </Card>

  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Get started with document verification
  </Card>
</CardGroup>