What's real, and what's illustrative
The fair question about any "art of the possible" demo is whether it's real or just screens over a seed script. Here is the honest answer — capability by capability — and six proofs computed on this page load, right now.
Proofs computed on this request
Just now this page called a public API with a real German IBAN and got back Commerzbank (COBADEFFXXX) in 140 ms.
Try it yourself0 unbalanced journals across 137 journals / 274 entries — the query a regulator would ask for, run live.
Open the ledgerNo executed proposal yet — run the copilot to see it post a real journal.
Open the copilotThe database itself refuses to record an approval by the same person who created the payment — enforced in the schema, not in application code that could be bypassed.
chk_txn_four_eyes: CHECK (checker_user_id IS NULL OR checker_user_id <> maker_user_id)
This page just queried 432 multinationals / 1,296 filing-years and summed our 15 clients' latest disclosed cash to $582,868,225,500 — the same aggregate the console returns, computed live. The engine is real; the filing figures are simulated.
Open the research consoleCall-for-funds CFF-2026-000009 brought €2,200,000 into Helios Renouvelables SA (FR613000600001123456789), posting balanced journal eebf4507-56f6-42ac… through the same locked settlement service a payment uses.
Open call-for-fundsExecutes for real (LIVE)
| SWIFT / BIC resolution | LIVE | Makes a real outbound call to the public openiban.com directory and decodes BICs to ISO 9362; IBANs are checked with the ISO 13616 mod-97 algorithm. |
| Double-entry ledger | LIVE | Every movement posts a balanced journal; the ledger is append-only, enforced by database triggers, and reconciles cached balances to the journal on demand. |
| Payment maker-checker | LIVE | Four-eyes is a database CHECK constraint (chk_txn_four_eyes), not an application if-statement — it cannot be bypassed by a code path that forgets to check. |
| Payment settlement | LIVE | Approving then settling a payment debits the account under a row lock and posts a balanced journal to the ledger you can see in the GCMS console. |
| Liquidity sweeps | LIVE | The ZBA sweeper concentrates real balances with advisory locks and a unique run-date claim to prevent double-sweeping, posting a journal per movement. |
| AI Copilot sweeps | LIVE | When it acts, the agent moves money only through a locked, double-entry settlement service — idempotent per proposal — never by writing a balance directly. |
| Statement export (camt.053 / MT940) | LIVE | Downloads a genuinely well-formed ISO 20022 camt.053 XML (and an MT940 text) built from the ledger — open it in your own tooling. |
| Role-based access control | LIVE | Every route is permission-gated; a persona without the right returns a real 403, and corporate personas are hard-scoped to their own client. |
| Call-for-funds settlement | LIVE | Funding an accepted call-for-funds posts a balanced, double-entry journal to the ledger under a row lock and moves the deposit — idempotent per request, never a direct balance write. The outbound "dispatch" API handshake is honestly simulated and tagged as such. |
Deterministic engines, not a language model (COMPUTED)
| AI assistant | COMPUTED | A rule engine that runs real scoped queries and states the figures — it never invents a number, and it says so. |
| Copilot reasoning | COMPUTED | The rules decide what moves; the reasoning is produced by a deterministic fallback locally (or a live model if an API key is configured), and every record is tagged which. |
| Board briefing | COMPUTED | A deterministic rule engine — every generated sentence carries the exact figure that triggered it, traceable to a query. |
| Executive analytics | COMPUTED | RoRWA, revenue concentration (HHI), plan variance and product economics are live SQL aggregates over the modelled book. |
| Cash-flow forecast | COMPUTED | A transparent velocity model from each account's own trailing inflow/outflow run-rates and payroll/month-end seasonality — no black box. |
| Liquidity & deposit alerts | COMPUTED | A deterministic rules engine over live balances, forecasts and the MNC dataset — forecast shortfalls, buffer breaches, deposit outflows and idle-cash whitespace. Every alert stores the figure that triggered it; no language model decides. |
| AI client research & business case | COMPUTED | A natural-language question becomes a structured plan, executed as real SQL over the client-intelligence dataset; every number in the answer is a reproducible aggregate. A language model (when a key is configured) only writes the prose around those figures — and each record is tagged model-vs-deterministic. |
Illustrative sample data (DEMO)
| Client, revenue & P&L figures | DEMO | Generated and internally consistent so the analytics reconcile, but they are NOT a real bank's numbers. |
| Balances & payment history | DEMO | A seeded showcase dataset for 16 corporate clients across regions — realistic, not real. |
| ESG, green & netting data | DEMO | Modelled to be plausible and self-consistent; the live pieces (a sweep, a lookup) execute on top of it. |
| Sanctions & screening hits | DEMO | Realistic watchlist names and match scores for the workflow — no screening vendor is actually called. |
| MNC filings & share-of-wallet | DEMO | The 400+ multinationals carry real, recognisable names, but their disclosed cash, three-year filings and Coalition-style wallet are simulated and flagged is_simulated in the data — every screen that shows them carries a SIMULATED tag. The engine that reasons over them is real; the numbers are illustrative. |
Proof-of-concept platform. The live capabilities are genuinely live; the figures are illustrative demo data and are not a real institution's numbers.