Operational trust is the compound result of resilience, observability, and accountability. It is earned slowly and lost quickly. Intelligent infrastructure makes all three properties harder to achieve, because failure modes become subtler and the surface area of automated action expands faster than human review can keep up.
What we measure
Four signals tell us whether operational trust is degrading: decision provenance, model drift, recovery time under partial failure, and the fraction of automated actions a human would reverse on review. When the fourth metric drifts upward, the system is losing trust faster than it is earning it — even if uptime looks healthy.
Assurance as a continuous practice
Annual audits are a poor proxy for trust in systems that change daily. Assurance must be continuous, machine-readable, and connected to the same evidence layer that engineering uses to debug production. The end state is a single source of truth that satisfies both the on-call engineer and the regulator.