Architecture & extensions
Cadence is built around a few clean seams: a third-party/account-servicer boundary, a product-parameterised decision core, swappable provider adapters, and per-session persistence. Each seam is where the system grows next.
Third-party provider and account servicer
The open-banking split is modelled explicitly so the wire format never leaks into the decisioning code.
Cadence — the TPP / console
Demo Bank — the mock ASPSP
A product-parameterised engine
One rule core, different parameter sets per product. The consumer instalment loan is active today; the mortgage parameter set is already defined to prove the seam.
Consumer loan — active
The live product runs the full Haushaltsrechnung and rule set: affordability buffer, debt-to-income ceiling, stability and tenure minimums, and a rate-stress test, all expressed as named parameters on a single product config.
- Demo APR
- 7.9%
- Max DTI
- 40%
Mortgage — the next module
The mortgage seam reuses the same core with a heavier parameter set and a few product-specific inputs. It is defined but inactive in the demo — the seam, not the screen.
- Property value
- Purchase price input
- Maximum LTV
- 90%
- Minimum equity
- 10%
- Term
- 120–360 months
- Stress-rate buffer
- +3.0 pp
Provider abstractions & persistence
Swappable providers
Per-session persistence
Open finance as the natural extension
Beyond payment accounts
The same architecture extends naturally from payment-account data toward broader open-finance access — savings, investments, pensions and insurance brought under comparable consent-based sharing. A wider evidence base would sharpen both categorisation and affordability while the boundary, the decision core and the audit trail stay exactly as they are.
This is framed as where the system would grow, not what the demo runs on. The demo deliberately stays within account-information access; the forward path is additive, plugging new providers into seams that already exist.