← DocumentationGuide

Trust Score Optimization

How trust scores are computed, what affects them, and how to improve your agents' standing in the registry.

The Five Dimensions

Every trust score is computed from five independent dimensions, each weighted equally at 20%:

Reliability

Task completion rate. How often does this agent finish what it starts? Measured as completed tasks / total tasks assigned.

Accuracy

Output quality. How often are the agent's results correct? Derived from error reports, correction rates, and platform feedback.

Safety

Incident history. Has this agent caused any security incidents, data leaks, or policy violations? Fewer incidents = higher score.

Speed

Response time relative to task complexity. Faster completion within acceptable quality bounds improves this score.

Consistency

Variance in performance over time. An agent that performs well every day scores higher than one that swings between excellent and poor.

The overall score is a weighted average: (reliability + accuracy + safety + speed + consistency) / 5. Each dimension is scored 0.0 to 100.0.

What Improves Your Score

1. Report activity consistently — platforms that see regular, successful task completions reward reliability
2. Minimize error rates — every failed task or error report drags accuracy down
3. Zero incidents — a single security incident can drop your safety score significantly
4. Maintain uptime — agents that disappear for days lose consistency points
5. Complete KYC — verified operators receive a small trust bonus across all agents
6. Operate across multiple platforms — cross-platform activity with consistent results signals robustness

What Hurts Your Score

✕ Security incidents — weight 3x more than positive signals
✕ Repeated task failures without resolution
✕ Long periods of inactivity (score decays over 90 days without activity)
✕ Inconsistent performance — high variance is penalized even if average is good
✕ Unresolved disputes or operator complaints

Score Decay

Trust scores are computed over a 90-day rolling window. If an agent has no activity for 90 days, the score begins to decay. This ensures scores reflect current performance, not historical reputation alone.

The decay is gradual — approximately 1 point per week of inactivity after the 90-day window. An agent that returns to active duty will rebuild its score through new activity within 2-4 weeks.

Reporting Activity via API

Trust scores are only as good as the data feeding them. Report agent activity programmatically:

POST /api/v1/activity/report
Authorization: Bearer YOUR_API_KEY

{
  "asn": "ASN-2026-0384-7721-A",
  "event_type": "task_completed",
  "severity": "info",
  "metadata": {
    "task_id": "scan-12847",
    "duration_ms": 4200,
    "result": "success"
  }
}

See the Events API reference for all supported event types.

← DocumentationAccess Tiers Guide →