Operations
Trust gradient
AI should demonstrate competence through tracked outcomes before being granted autonomous execution. Qorpera starts fully supervised and provides the telemetry needed to make an informed delegation decision. There is no automatic promotion — the operator always chooses.
1. Observe
Qorpera monitors and shows you what is happening. Situations are detected and written to the wiki, but no action proposals are surfaced for review. You see exactly what the system sees — a diagnostic baseline.
2. Propose
The default operational posture. Qorpera proposes concrete actions for every detected situation. You approve, edit, or reject. Every response updates the per-situation-type track record — total proposed, total approved, consecutive approvals, approval rate, rejection patterns, correction history.
3. Delegate
When the track record shows a situation type is consistently handled the same way, you can choose to delegate it to a system job. System jobs execute autonomously — detect, reason, act — without per-instance human approval.
The delegation decision is always yours. There is no threshold that automatically promotes a situation type to autonomy. You read your own approval history and make a deliberate call.
Why supervised-with-delegation over automatic graduation
Three reasons:
- Operator confidence. You see exactly which situation types are delegation candidates and why — you’re reading your own history, not trusting a threshold algorithm.
- Appropriate granularity. Some situation types should never be autonomous regardless of approval rates (high-stakes decisions, relationship-sensitive interactions, regulatory actions). The operator makes this judgment, not the system.
- Clean accountability. When a system job runs autonomously, the operator explicitly chose that. No ambiguity about who authorised the AI to act.
Revocation
Autonomy is revocable at any time. A delegated situation type can be pulled back to propose-mode immediately; the system job halts and future instances surface for approval again.