A production system is observable when its emitted information lets operators investigate important questions about its internal behavior without first adding a special-purpose diagnostic for every failure.

Three common telemetry signals provide different views:

• metrics summarize rates, levels and distributions efficiently;
• logs record detailed discrete events;
• traces connect work across the path of one operation.

A metric may reveal that latency increased, a trace may identify the slow downstream span, and logs from that component may explain the concrete failure.

Collecting all three signals does not automatically create observability. Their names, identifiers and dimensions must correspond to meaningful system concepts so evidence can be connected.

Dashboards answer known recurring questions; exploratory queries are needed for failures that were not predicted in advance.

Observability is therefore a property of the diagnostic information a system exposes, not the name of one monitoring product.