Use only the decision-relevant signal.
A credit qualification is a threshold predicate; device posture and velocity are synthetic fixtures. Raw credit scores and customer profiles are not put into the displayed decision receipt.
JointRisk / GFTN reference · controlled walkthrough
A governed decision before an action.
JointRisk is a reference control room for a shared feature layer with separate fraud and early-credit-loss assessments. It records the evidence that led to an approve, step-up, or decline before a simulated action is allowed to proceed.
the decision path
Keep the evidence, the decision, and the action separate.
The reference sequence gives a judge or partner something concrete to inspect: the inputs and their boundary, the policy decision, and the governance record are distinct rather than hidden inside a single “risk score.”
Do not collapse fraud and early loss.
A shared feature layer feeds distinct fraud and early-credit-loss assessments. The system preserves the task boundary instead of turning either one into a vague, unreviewable “badness” score.
Bind a signed decision to one action.
The outcome is approve, step-up, or decline. The signed decision is bound to the simulated transaction, freshness, policy, and evidence so replay or tampering is rejected by the reference flow.
Authority comes before the risk gate.
The core reference sequence verifies bundled GLEIF testnet organization artifacts offline, then checks a key-bound officer delegation and a capped, revocable agent authority. This is technical test-ecosystem evidence, not a fresh production vLEI lifecycle, GLEIF accreditation, or GXS integration.
A label is not a promotion.
Confirmed labels first form a candidate model. A reviewer may promote or roll back only after signed synthetic shadow evidence matches the exact candidate and passes explicit safety and friction gates. Sandbox events never become fraud or credit-loss learning labels by implication.
what this does not establish
The strongest useful claim has a boundary.
- It demonstrates: a reproducible, synthetic control sequence with separate risk heads, signed transaction-bound decisions, step-up/decline controls, and inspectable candidate-promotion governance.
- It does not demonstrate: live GFTN deployment, GXS data access, production bank connectivity, settlement, model validation, operational fraud/loss outcomes, or customer-friction outcomes.
- Optional Open Banking sandbox work: if enabled by an operator-owned test account, it adds sandbox checkout lineage only after an approved local gate. It does not change the deterministic core or create a bank-outcome claim.
next step
Inspect the reference on the terms it can honestly support.
The controlled control room is the right place to discuss its technical sequence. The evidence ledger is the right place to inspect broader protocol and interoperability claims. Neither substitutes for a bank, partner, regulator, or model-validation programme.