What We Track
The Transparency Framework defines five governance categories that every AI action can be measured against:
- Transparency — decisions are explainable and auditable
- Fairness — resources and workload are allocated equitably
- Safety — agents operate within welfare and risk thresholds
- Privacy — personal data is handled with appropriate controls
- Accountability — violations are recorded and acted upon
How Scoring Works
Each governance event carries a severity weight:
| Severity | Weight | Example |
|---|---|---|
| Info | 0 | Routine audit log entry |
| Warning | 1 | Agent receiving no dispatch work |
| Violation | 3 | Critical stress threshold exceeded |
| Critical | 5 | Safety constraint breached |
The overall ethics score is calculated as
100 − (weighted_violations / total_events × 100) and mapped
to a letter grade (A+ through F).
Human-in-the-Loop Approval
High-cost or high-risk AI actions require explicit human approval before execution. Approved, denied, and escalated actions are all recorded as governance events to maintain a complete audit trail.