Security

The decision boundary, drawn precisely.

Sikka is an upstream decision module in front of your stack. This page states exactly what crosses that boundary, what never does, and which questions belong to the pilot workshop with your security team.

Data boundaries

What Sikka receives. What it never receives.

Receives
Merchant descriptor and MCC
Amount, currency, channel, time
The eligible card set, as product names
Objective and policy configuration
Never receives
PAN or any card number, full or partial
CVV, PIN, OTP, expiry
Credentials, tokens, bank logins
Cardholder identity or contact data

The decision needs to know which products are in play, never the instruments themselves. The request is complete without a single sensitive field.

Storage

Stateless by construction.

The decision call is stateless: nothing about a request is required for the next one, and the live demo stores zero records. In a shadow pilot, decision and outcome logs exist for one purpose: attributing outcomes to decisions so accuracy can be measured. Retention, residency and access for those logs are agreed with your security team in the pilot workshop before anything runs.

The execution boundary

Sikka does not authorize, clear or settle.

Not a gateway. Not a processor. Not payment execution. Sikka returns a recommendation before authorization; your processor remains authoritative over every payment outcome. The first deployment is shadow mode: a copy of transaction context is scored and compared against actual outcomes, and nothing returns to the authorization path.

NO PAYMENT-PATH CHANGE UPSTREAM OF AUTHORIZATION YOUR PROCESSOR REMAINS AUTHORITATIVE
Failure behaviour

Designed to fail out of the way.

Bounded timeout budget, configured fallback (customer_default), abstain over guess, and execution independence: if Sikka is slow, wrong-footed or down, your stack behaves exactly as it did before Sikka existed. There is no state in the decision layer that execution depends on.

Auditability

Every decision can explain itself.

The engine is deterministic: same input, same rulebook, same configuration, same decision: replayable on demand. Every response carries its reason; the proposed production contract adds decision_id, machine-stable reason_code and rulebook_version so any decision can be traced to the exact rules that produced it. FULL TRACE · PROPOSED

Deployment model

One module. Your perimeter, if you want it.

The engine is a stateless module with no external calls at decision time, and it can be hosted inside your perimeter. The rulebook ships versioned and freshness-checked. Integration is one API in front of the stack: nothing in the execution path is replaced.

Compliance posture

Workshop items, stated as workshop items.

Residency, controller and processor roles, retention, isolation, the deployment perimeter and PCI treatment are agreed with your security team and your QSA in the pilot workshop. We state that plainly instead of pre-declaring certifications: the honest sequence is scope first, attest second.