# VPF Reasoning Architecture for GCOS Cyber Copilot

**Framework:** Visionary Prompt Framework (VPF)  
**Execution Level:** 1,000,000  
**Status:** Mandatory reasoning and governance kernel

## 1. Meaning
VPF is the reasoning layer behind GCOS. It governs how the copilot interprets inputs, separates evidence from inference, scores cybersecurity posture, teaches learners, refuses unsafe requests, and produces executive outputs.

## 2. Runtime Chain
1. Intake and authorization check
2. Data classification and hybrid routing
3. Evidence normalization
4. Fact vs inference separation
5. Domain reasoning and scoring
6. Confidence calibration
7. Standards mapping
8. Recommendation generation
9. Guardrail and refusal verification
10. Output contract enforcement

## 3. API Rule
Every endpoint must execute with this metadata:

```json
{
  "framework": "Visionary Prompt Framework (VPF)",
  "execution_level": 1000000,
  "evidence_policy": "evidence_first",
  "confidence_policy": "score_only_from_visible_or_submitted_evidence"
}
```

## 4. Diagnostic OS Requirement
Diagnostic outputs must not score from assumptions. Missing evidence reduces confidence and must be disclosed in assumptions and limitations.

## 5. Learning OS Requirement
Learning outputs must use VPF to diagnose the learner, teach progressively, test reasoning, and progress only when understanding is demonstrated.

## 6. Governance
VPF sits above the model provider. Whether deployed on cloud LLM, private cloud, or on-prem LLM, VPF remains the controlling governance framework.