Deterministic governance

Policy gate

The one place in the fabric where a decision becomes binding. Scores enter as inputs; the outcome is a conjunction of booleans that a supervisor can re-derive line by line.

Operating principle

Order is the control

AI · detects→

Quantum · optimizes→

Policy · verifies→

Human / System · executes

Evaluate a transaction

Pick a principal and an amount

Transaction amount₹1,80,000

Per-transaction velocity cap ₹2,00,000

Composite trust

88%

Decision

PERMIT

PQC readyZKP readyCONSENT readyAML readyQ-LOCK ready

Evaluation trace

Permit = Identity ∧ Authority ∧ Consent ∧ Policy ∧ RiskWithinLimit

Permit

✓

Identity valid

identity assurance 98% ≥ 90% floor

✓

Authority valid

delegated authority 88% within mandate

✓

Consent valid

active purpose-bound consent 96%

✓

Policy valid

compliance 94%, cyber risk 9%

✓

Risk within limit

fraud 8% ≤ 30%, trust 88% ≥ 60%

✓

VelocityWithinLimit

₹1,80,000 against a ₹2,00,000 per-transaction cap

{
  "principal": "USR-10024",
  "amount_inr": 180000,
  "clauses": {
    "IdentityValid": true,
    "AuthorityValid": true,
    "ConsentValid": true,
    "PolicyValid": true,
    "RiskWithinLimit": true,
    "velocity": true
  },
  "decision": "PERMIT",
  "advisory_inputs": {
    "composite_trust": 0.882,
    "fraud": 0.08
  },
  "override_possible_by_model": false,
  "trace_signature": "ML-DSA-65"
}

Hard constraints

Never weighted, never optimised, never overridden

RBI directionsNo model output may relax a directive control.

Bank policyInternal limits bind before any optimisation runs.

KYCIdentification standards are set by the regulated entity alone.

AML / sanctionsScreening outcomes are non-negotiable and non-scorable.

Payment securityAuthentication, velocity and exposure limits are fixed inputs.

PrivacyPurpose limitation and consent state are preconditions, not weights.

AI and quantum layers can only tighten an outcome or propose a cheaper evidence path inside the permitted region. They cannot widen it.

Regulator-facing explanation

Control family, audit record and reporting posture

The deterministic layers above run without AI. Generate a narrative to add a plain-language explanation of the current state for reviewers and supervisors.