Which operational tasks for a self-hosted durable execution engine are most frequently underestimated during evaluation?

Observability pipeline setup, disaster recovery exercises, and handling of corrupted journals often surprise teams. Upgrades that require state schema migrations and rolling updates without data loss add complexity. Include these in your evaluation with a time-allocation estimate, not just infrastructure. A caveat: underestimation is common because initial PoCs rarely exercise failure modes or long-term maintenance routines.