Judge Agents Quality layer
Judge Agents audit the reasoning behind every critical output
Two independent evaluators check that the reasoning behind the platform's most important results is coherent with the inputs each run received, before those results move on. If it isn't, the work goes back.
Ransomware on OT
Expected annual loss$4.2MRun inputs: 60+ facilities · Oil & Gas · FY2026
3.1 Scenario reasoning
| Loss driver | Annual |
|---|---|
| Production downtime | $2.6M |
| Recovery and restoration | $1.1M |
| Third-party liability | $0.5M |
Review0 / 3
The Judge only checks. It never writes, edits or replaces a result.
They check the reasoning, not only the result
The scenario and risk-calculation agents hand over their results together with the reasoning behind them. The Judge Agents read both against what the run started from.
The output, its reasoning and the run's inputs.
Is the reasoning coherent with the inputs?
One of two outcomes.
They can send work back
Incoherent reasoning never moves on to the next step.
They never generate content
A Judge only checks. It doesn't write, edit or replace a result.
Underneath DeRISK™ CRQ and DeRISK™ UWA
Judge Agents run quietly underneath both products. When a critical result is produced, they check the reasoning behind it, and only coherent work moves on.
- CRQ
- UWA
- Pass
- Sent back
- Illustrative
Every critical result, checked before it moves on.
Judge Agents are a platform quality layer under DeRISK™ CRQ and DeRISK™ UWA. See them at work in a walkthrough with our team.