Use case: ticket triage

Problem: incoming support tickets need a team, a priority, and refund detection — human sorting is slow and expensive; full automation risks misroutes.

Jev’s role: the “all three types” showcase. The official Quick start plus the Confidence-gated routing pattern: one request asks everything; confidence sets the automation boundary. This is exactly what the site’s runnable examples implement (JavaScript / Python).

Request design

{
  "model": "jev-latest",
  "state": "Hi, I've been trying to connect my Stripe account for 3 days and the integration keeps failing. I'm losing sales. Please refund my subscription and help ASAP.",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this",
      "criteria": {
        "billing": "Payment or subscription issues",
        "technical": "Bugs or integration problems",
        "sales": "Pricing or account questions"
      }
    },
    "frustration": {
      "type": "score",
      "instructions": "How frustrated the customer appears",
      "criteria": ["Calm, just stating facts", "Frustrated but civil", "Very angry, strong language"]
    },
    "asks_refund": {
      "type": "noul",
      "instructions": "The customer explicitly requests a refund"
    }
  }
}

Routing rules (all in code)

// Department: auto-assign when confident, human triage otherwise
const route =
  department.confidence >= 0.7
    ? `auto-assign to ${department.choice}`
    : 'route to human triage';

// Refund: asymmetric costs — conservative thresholds, middle band to humans
const refund =
  asks_refund.noul >= 0.9 ? 'auto ticket'
  : asks_refund.noul >= 0.5 ? 'human review'
  : 'no action';

// Frustration: feeds SLA ordering only, never triggers alone
priority = frustration.score >= 1.5 ? 'high' : 'normal';

Threshold design points (official Noul guidance plus our advice):

• department gates on confidence (0.7 to start) — a misroute only costs time, so higher automation is tolerable
• asks_refund uses the noul probability in three bands — refunds are money; 0.5–0.9 must reach a human
• frustration’s Score only feeds ordering, never triggers actions — single-signal triggers misfire

Why one request instead of three

The three questions share one state; batching saves two-thirds of input tokens and round trips (the direct payoff of parallel questions — see Capabilities). Response answers keys mirror the questions one-for-one, so parsing is unambiguous.

Rollout checklist (our advice)

1. Run a week of shadow mode: log Jev’s routing next to human sorting, measure agreement
2. Calibrate criteria per team — only you know where “technical support” ends and “billing” begins
3. Keep probabilities/confidence logs for every automated action — appeals and audits will need them
4. Watch for PII: state leaves your system for the API; sanitize first per your compliance rules

Back to the overview; errors and rate limits: API reference.