AI → Signals → Governance → Authorization Decision → Execution Runtime → Attestation
1. Signal ingestion
AI systems generate signals representing proposed actions, observations, or intent.
These signals are structured inputs to the system.
2. Governance evaluation
Signals are passed into the Governance layer (@parmanasystems/governance).
Governance applies:
signed policies
deterministic rules
authorization constraints
No probabilistic reasoning is involved.
3. Authorization Decision generation
Governance produces an Authorization Decision.
This decision is:
deterministic
reproducible
bound to a specific policy version
It is the only valid outcome of governance evaluation.
4. Execution Runtime enforcement
The Execution Runtime enforces the Authorization Decision.
It ensures:
replay protection via executionId
deterministic enforcement behavior
rejection of unauthorized execution attempts
If validation fails, execution is blocked.
5. Attestation generation
Each Authorization Decision produces a cryptographic attestation.
The attestation includes:
policy version
signal hash
decision outcome
runtime identity
It can be independently verified using public keys.
Properties of Authorization Flow
Deterministic
Identical inputs always produce identical outputs.
Secure
Unauthorized or invalid executions are blocked.
Replay-safe
Each execution is uniquely bound and cannot be replayed.
Verifiable
All decisions are independently verifiable without system access.
Summary
Parmana enforces strict separation of concerns:
AI generates signals
Governance evaluates signals
Authorization Decision is computed deterministically
Execution Runtime enforces decisions
Attestation ensures verifiability
