Pillar 01 · credit intelligence
Score the cash flow, not the file
Feature extraction turns normalised economic events into a behavioural vector — inflow strength and regularity, net margin, income trajectory, balance resilience, counterparty diversity, digital-rail participation, graph connectivity and repayment outcomes where they exist.
Two governed scorecards run on every request: the registered champion produces the score, the challenger shadow-scores it, and divergence is monitored. The response carries the reason codes that produced the number.
reason codes12-month PDindicative limitfraud probability
Pillar 02 · identity & verification
Verify once, trust everywhere
Identifiers are salted into blind indexes and resolved to a subject on the graph — the spine never needs the raw identifier to know who it is talking about. Verification runs liveness first, then a 1:1 DHA/ABIS biometric match.
A verified subject can be issued an Ed25519 credential that any device holding the issuer public key verifies with no network call at all — which is exactly where connectivity is weakest and per-transaction biometric cost is highest.
entity resolutionliveness1:1 ABISoffline credentials
Pillar 03 · FICA-as-a-service
Five checks, one call, one retention reference
Customer due diligence under the Financial Intelligence Centre Act runs as a single orchestrated operation: identity resolution, DHA verification, liveness, PEP and sanctions screening, and CIPC look-through for juristic persons.
The outcome is risk-rated rather than binary — clear, review or fail — so the review queue holds the cases that need a human and nothing else. Every check writes a retention reference against the five-year obligation.
DHAPEP & sanctionsCIPC look-through5-year retention
Pillar 04 · health intelligence
Price the behaviour, not the declaration
Engagement is measured — steps, gym attendance, screenings, prescription adherence, sleep — and converted through a discrete proportional-hazards form into an explainable morbidity index and a hazard band from H1 to H5.
The actuarial layer prices expected annual claims, a recommended risk reserve and an indicative loss ratio per member. Health data is special personal information under POPIA and is sealed under its own key domain, separate from everything else.
engagement scoringmorbidity indexhazard bandssegregated keys
Pillar 05 · loyalty & engagement
A points economy that cannot quietly lose a point
Points sit on a per-account, append-only, hash-chained ledger. The chain is verified at every boot: a break is treated as an operational incident and the platform refuses to serve rather than serve a balance it cannot prove. Redemption is idempotency-first with an inventory compare-and-set, so a retried request cannot double-spend.
A three-layer guard — database constraint, API validation and runtime check — keeps rewards on the correct side of the National Credit Act's prohibition on inducements to borrow. Partner analytics are exposed only as k-anonymous segments over per-partner pseudonyms.