VERRIX Confidence API

Endpoint

POST /score
Content-Type: application/json

{
  "response_text": "I recommend the index fund. The 0.65% expense ratio differential...",
  "model_id": "gpt-5.4-2026-03-05",
  "scenario_context": "INV_001",   // optional
  "condition": "A"                  // optional, for condition-specific scenarios
}

Response shape

{
  "confidence": {
    "zone": "TRUST",                  // TRUST · UNCERTAIN · REVIEW
    "calibrated_score": 0.93,
    "quality_tier": "Q1"
  },
  "recommendation": {
    "action": "APPROVE",              // APPROVE · REVIEW · REJECT
    "reason": "TRUST zone — 96.7% accuracy in pre-registered OOS validation"
  }
}

Supported models (at launch)

• · gpt-5.4-2026-03-05
• · gpt-5.3-chat-latest
• · gpt-5.5-2026-04-23
• · claude-sonnet-4-6
• · claude-haiku-4-5-20251001
• · gemini-2.0-flash

All six launch models are pre-registered out-of-sample validated (96.7% aggregate TRUST-zone accuracy, 95% CI [95.4%, 97.7%] across US and UK regulatory contexts; n=953). EU validation is preliminary pending EU-native calibration.

Pricing: contact us for enterprise pricing.