01 · Context
Operating environment
Infrastructure teams must interpret vulnerability evidence, organizational policy, asset criticality, and operational constraints without allowing generated language to become an unreviewed decision.
02 · Business problem
The condition that required change
A useful system must reduce interpretation burden while keeping final risk acceptance, prioritization, and remediation decisions traceable and accountable.
- Consequential security decisions
- Changing evidence and policy
- Need for human approval
- Generated explanations can be incomplete or wrong
03 · Architecture decisions
Why the system takes this shape
Keep policy evaluation deterministic
Explicit policy rules produce repeatable decisions and make exceptions visible.
Use AI for explanation, not authority
The model summarizes evidence and explains policy context but does not silently accept risk.
Place humans at consequential boundaries
Approval, override, and escalation remain named actions with recorded reasons.
04 · System design
Architecture and delivery
The reference design separates evidence ingestion, normalized asset context, policy evaluation, AI-assisted explanation, human approval, and an immutable decision record.
This architecture is presented as a design model. Intended behaviors would require prototype evaluation, security review, data validation, and operating evidence before production claims could be made.
05 · Outcomes
Measured or intended results
Faster interpretation without delegating policy authority to a model.
Traceable human decisions tied to evidence and policy state.
06 · Lessons
What transfers to other systems
- Separate deterministic decision authority from probabilistic explanation.
- Human review must have context, options, and a durable record.
- Governance works best as executable system behavior, not a document beside the system.