State model
A decision snapshot binds period, entity hierarchy, measures, formula versions, assumptions and source cut-offs. Scenarios never overwrite actuals.
Cash conversion depends on receivables, payables, inventory, seasonality and operational constraints that simple product propensity cannot explain.
A simulation twin tests transparent working-capital levers, separates customer-provided assumptions from observed facts and shows trade-offs rather than a single recommendation.
Organisations, systems and operating conditions are intentionally anonymised and recomposed. The design demonstrates engineering and banking-domain reasoning; it does not represent a named client estate, vendor product or measured production result.
Typed commands, state changes and receipts cross an event spine without surrendering domain ownership.
Permitted workThe system coordinates documents, parties and obligations. Relationship judgement, legal interpretation and external commitments remain human-authorised.
Consistency ruleModel parties, legal capacities, instruments, versions and deadlines explicitly; an entity or clause conflict keeps the case open.
Hard boundaryThe model is not a system of record, identity provider, policy authority or proof that an external effect occurred.
A business operator needs a reconciled view, explanation or scenario from several data products while retaining ownership of the business decision.
The simulator cannot quote a binding price, assess suitability or execute a facility change.
A deterministic outer workflow contains model-led work inside typed, observable calls. Dashed messages remain proposals until policy or a human grants authority.
A decision snapshot binds period, entity hierarchy, measures, formula versions, assumptions and source cut-offs. Scenarios never overwrite actuals.
Read from curated analytical products rather than operational tables. Use drill-through links for source detail and typed write-back only for approved decisions.
Definitions and units are resolved before aggregation. Missing data stays missing; mixed grains and double-counted entities fail reconciliation.
These roles are deliberately vendor-neutral. Each can be independently owned, versioned and replaced.
Chooses an approved model route by task, risk, evidence quality, latency budget and cost ceiling; enforces structured outputs.
Runs component, route, trajectory, failure, harm and outcome tests against the versioned system manifest.
Builds time-qualified projections from source events and reconciliations without becoming the legal system of record.
Presents claims beside evidence, alternatives, uncertainty, missing information, permitted actions and current custody.
Appends request, versions, policy result, model proposal, approval, action receipt, readback, correction and custody events under one correlation key.
Evaluates identity, purpose, capability, amount, risk tier and policy version; returns allow, deny, step-up or human-review with reasons.
Durable records carry provenance, authority, effect and custody without turning a transcript into an uncontrolled memory store.
Model parties, legal capacities, instruments, versions and deadlines explicitly; an entity or clause conflict keeps the case open.
The selected design is not universally superior. It is the safer fit for this boundary and failure cost.
Reserve larger models for residual reasoning after deterministic and smaller-model gates.
Send every request to the most capable available model.
Cost acceptedRouting adds evaluation work and operational complexity, but controls cost, latency and unnecessary data exposure.
Test prompts, models, tools, knowledge, policies, state transitions and human paths together.
Use a static answer-quality benchmark as the release gate.
Cost acceptedSystem evaluation takes longer and needs synthetic environments, but detects authority and recovery failures that answer scoring misses.
Use event-fed projections for scale and direct readback for consequential effects.
Fan out to all systems of record for every interaction.
Cost acceptedRead models introduce lag and reconciliation work, but reduce source load and make cross-system views feasible.
Keep people at irreversible, ambiguous and policy-exception points; sample lower-risk automated outcomes independently.
Require the same manual approval at every step.
Cost acceptedRisk-tiering reduces review load but needs calibrated thresholds, sampling and immediate withdrawal of authority when drift appears.
Use append-only events plus a rebuildable current-state projection.
Overwrite the case row with its latest status.
Cost acceptedReplay and storage are more complex, but point-in-time reconstruction and correction lineage remain possible.
Compile stable decision logic and retain retrieval for explanation and residual ambiguity.
Ask a model to interpret the source document for every request.
Cost acceptedRule compilation needs controlled change, but creates repeatable decisions, regression tests and clear exceptions.
Retries are bounded by knowledge of business effect; unknown outcome remains visible, owned and independently reconciled.
Actual thresholds belong to accountable service owners. The design exposes the equations and observables that those owners must baseline.
query_load = active_users x decisions_per_hour x source_scansscenario_work = entities x periods x assumptions x alternativesreview_value = avoided_rework + decision_gain - compute - human_effortSegment customer, counterparty and commercially sensitive data by case purpose and delegated authority.
A design is production-ready only when teams can prove what happened, recover it and change it safely.
party and authority resolution
document version and clause lineage
deadline and custody recovery
external communication approval
Input sources, assumption ownership, scenario lineage, sensitivity ranges, exclusions and client-confirmed priorities.
A relationship manager and product specialist select any proposition taken forward.
Begin with explanation and transparent calculations. Add recommendation only where decision rights, counterfactual evaluation and outcome measurement are established.
Relationship, trade operations, legal, product and financial-crime owners govern the journey.