Understanding a system's behaviour from its outputs (metrics, logs, traces) well enough to answer questions you didn't anticipate.