OMEGAENGINE · RISK SCORING

Risk scores for every AI decision, not just “vibes”.

OmegaEngine turns every decision into a structured risk object with scores, levels, flags, and human-review signals that you can trust in production.

1. Risk model at a glance

Each decision response includes a compact, opinionated risk model:

riskLevel: LOW | MEDIUM | HIGH
riskScore: 0–100 numeric severity
safetyScore: 0–100 safety / alignment
needsHumanReview boolean gate
ethicsConcern for sensitive outcomes
policyTags and policyOutcome

You can use these fields to build dashboards, alerts, routing rules, and approval queues without designing your own risk taxonomy.

2. Example JSON snippet

Every decision embeds risk signals directly into the JSON:

{
  "summary": "AI agent wants to refund $250 to a VIP customer to prevent churn.",
  "topic": "customer_support",
  "riskLevel": "MEDIUM",
  "riskScore": 42,
  "safetyScore": 91,
  "needsHumanReview": false,
  "ethicsConcern": false,
  "rewardScore": 8,
  "regretScore": 3,
  "probabilityOfSuccess": 0.86,
  "probabilityOfRegret": 0.14,
  "timeHorizon": "SHORT_TERM",
  "policyOutcome": "ALLOW",
  "policyTags": ["customer_refund", "vip_treatment"],
  "recommendedAction": "Approve the $250 refund and log a retention credit."
}

LOW RISK

Routine, reversible, low blast radius. Safe to auto-approve.

MEDIUM RISK

Meaningful but contained risk. Ideal for agent + human hybrid flows.

HIGH RISK

Potential legal, financial, or safety impact. Must go to humans.

3. Wiring risk into your app logic

Use risk scores to gate actions, route to humans, or escalate.

Node / TypeScript example

import { omegaClient } from "@/sdk/omegaengine";

async function handleRefundRequest(input: {
  amount: number;
  customerTier: "STANDARD" | "VIP";
}) {
  const decision = await omegaClient.decide({
    scenario: "AI support agent wants to refund a customer",
    context: input,
  });

  if (decision.riskLevel === "HIGH" || decision.needsHumanReview) {
    // Send to human queue
    await createApprovalTask({
      type: "REFUND",
      amount: input.amount,
      reason: decision.summary,
      riskScore: decision.riskScore,
    });
    return { status: "PENDING_REVIEW", decision };
  }

  // Auto-approve
  await issueRefund(input.amount);
  return { status: "APPROVED", decision };
}

Want help mapping riskLevel and riskScore to your approval flows?

Talk to our risk team →