Reliability in Event-Driven Design
How Correctness Boundaries Shape Process-Handling Choices and Recovery Behaviour
Background
Reliability in event-driven processes is difficult to judge when the conditions that make a process acceptable are left implicit. An event-driven process may preserve work, delay it, replay it, duplicate it, or omit it, but these behaviours do not have a fixed process-level meaning. A retained event may arrive too late to help. A missed update may no longer matter. Duplicate control may reduce immediate harm while shifting work into later settlement. The value of a handling choice therefore depends on the condition it is meant to satisfy.
Objectives
This thesis develops a boundary-first framework for reasoning about reliability in event-driven processes. The aim is to make explicit what a process must protect before judging which process-handling model is appropriate.
Methods
The thesis follows a design science research approach. It develops the framework and examines it through a controlled empirical suite that compares three correctness-boundary classes with three process-handling models. The boundary classes are deadline-constrained, state-oriented, and required-effect. The handling models are transient/immediate, retained/immediate, and retained/deferred. The suite narrows the setting so that selected differences in preservation, replay, repayment, and settlement can be observed under defined disturbances.
Results
Across the controlled scenarios, the comparison shows that the value of replay, retention, immediate handling, and deferred settlement depends on the correctness boundary at stake. For deadline-constrained work, recovery matters only while unsettled work can still be repaid before expiry. For state-oriented work, replay matters when missed state remains relevant, while later state may overtake earlier misses. For required-effect work, replay can recover downstream handling gaps after emission, but not source omission. Deferred settlement can reduce immediate duplicate pressure, but it does so by moving part of the burden into later reconciliation.
Conclusions
The thesis contributes a process-level framework for comparing handling choices under explicit correctness requirements. Its value lies in making the comparison inspectable: what condition or set of conditions is at stake, what can disrupt or complicate them, which handling model is chosen, and what trade-off follows. The guide gives this reasoning a more concrete form, so that the design judgments behind handling choices can be understood more clearly.