| Abstract | Complex stochastic simulators are often evaluated mainly through aggregate performance indicators, yet those summaries can obscure execution structure, variant behaviour, and disruption pathways. This makes simulator behaviour harder to inspect and validate. This paper introduces a process-mining-based observability workflow for stochastic simulation, transforming simulator executions into auditable event logs for post hoc conformance, variant, and dependency analysis without modifying core model logic. Using Mercury, a stochastic, event-driven air traffic management (ATM) simulator, as a case study, we implement a passive logging layer and a post-processing pipeline that constructs (i) flight-centric gate-to-gate traces for conformance and performance analysis and (ii) an object-centric event log (OCEL) linking flights, passenger connections, and aircraft. To the best of our knowledge, this is the first published application of object-centric process mining to ATM simulation outputs. In a large-scale European day-of-operations scenario (2,793 flights per run), logging increases wall-clock runtime by 10.63%, enabling detailed trace-level analysis. Across 10 stochastic replications, 97.6% of flights follow the complete ten-step gate-to-gate lifecycle with zero ordering violations, while the OCEL supports traceable analysis of passenger connection outcomes and cascading disruption effects beyond what aggregate key performance indicators (KPIs) can reveal. The results suggest a general pattern for adding process-level observability to complex simulators beyond ATM. |
|---|