The ability to ask new questions about a running system from the outside, without shipping new code, by leaning on logs, metrics, and traces.
Observability is the property of a system that lets you understand its internal state from its external outputs, logs, metrics, and traces, without having to ship new code to add a probe. The term comes from control theory and was applied to software by Charity Majors and the Honeycomb team. The practical test for observability is: can you investigate an incident you have never seen before, using the data you already collect, without adding instrumentation. If the answer is yes, you have observability; if it's no, you have monitoring.
Monitoring tells you whether the things you predicted would break are broken. Observability tells you about the things you didn't predict. Modern services produce too many emergent failure modes for predicted-question monitoring to be enough; observability is what catches the long tail.
See the part of the platform that handles observability in production.