The idea
Metrics summarize numeric behavior, logs record events and traces follow work through components. Combine signals according to the question. A running machine is not proof that users can complete a task. Measure errors, usable response times and workload in addition to infrastructure resources.
Worked example
A fictional checkout has low CPU usage but every payment request fails. A machine-only dashboard looks healthy, while a completed-checkout rate reveals the user impact. A trace or correlated log can then help locate the failed dependency.
Try it
Choose five signals for a fictional checkout and explain the purpose of each. Include one user outcome and one dependency observation. Match a high-level alert to the logs or traces that would help investigate it without recording payment secrets.