Trust Score Algorithm

Trust Score (0-100) represents overall platform trustworthiness based on hallucination detection, correction rates, and compliance posture. This guide explains the calculation.

Score Components

Trust Score = Hallucination Detection Rate × 0.35 + Correction Rate × 0.35 + Compliance Score × 0.30

Component 1: Hallucination Detection Rate (35%)

Measures how well hallucinations are detected and caught:

Detection Rate = Hallucinations Detected / (Hallucinations Detected + Missed)

Example:
- Detected: 47 hallucinations
- Missed: 3 hallucinations
- Detection Rate: 47/50 = 94%
- Component Score: 94 × 0.35 = 32.9/100

Component 2: Correction Rate (35%)

Measures how many detected hallucinations are corrected:

Correction Rate = Corrected / Detected

Example:
- Detected: 47 hallucinations
- Corrected: 45 hallucinations
- Correction Rate: 45/47 = 96%
- Component Score: 96 × 0.35 = 33.6/100

Component 3: Compliance Score (30%)

Measures regulatory and policy compliance:

Compliance Score = (Policies Enforced / Policies Defined) × 100

Example:
- 12 policies defined
- All 12 enforced successfully
- Compliance Score: 12/12 = 100%
- Component Score: 100 × 0.30 = 30/100

Overall: 32.9 + 33.6 + 30 = 96.5/100

Scoring Bands

Score	Interpretation	Action
95-100	Excellent	Maintain current posture
85-94	Very Good	Monitor for degradation
70-84	Good	Address emerging gaps
50-69	Fair	Significant improvements needed
<50	Poor	Urgent intervention required

Time-Based Averaging

Trust Score is smoothed over time to avoid noise:

Daily Score = Real-time calculation (as above)
Weekly Score = Average of 7 daily scores (30% weight to last 3 days)
Monthly Score = Average of 30 daily scores
Overall = Weighted average (today: 10%, this week: 30%, this month: 60%)

This prevents single bad days from disproportionately affecting the score.

Historical Tracking

View score evolution:

history = client.shield.get_trust_score_history(
    granularity="daily",
    days=90
)

for entry in history:
    print(f"{entry.date}: {entry.score}/100")
    print(f"  Detection: {entry.detection_rate}%")
    print(f"  Correction: {entry.correction_rate}%")
    print(f"  Compliance: {entry.compliance_score}%")

Benchmarking

Compare your score to industry peers:

benchmark = client.shield.get_score_benchmark()

print(f"Your Score: {benchmark.your_score}")
print(f"Industry Average: {benchmark.industry_avg}")
print(f"Your Percentile: {benchmark.percentile}%")  # e.g., top 25%

Next Steps

• Improving Score: Strategies to increase each component
• Monitoring: Set up score tracking
• Benchmarking: Compare against industry