Which observability signals should a JVM or .NET service emit by default, and at what overhead?

Open question

The JVM can run Flight Recorder continuously and inspect the result with the jfr tool, write GC logs and expose JMX; .NET publishes runtime counters that dotnet-counters, which its documentation describes as a tool for ad-hoc health monitoring and first-level performance investigation, can observe live; both platforms can load an OpenTelemetry agent that adds traces and metrics without code changes. The documentation of each describes the mechanism but rarely a recommended production default, and overhead figures quoted in various places depend on workload and settings. For a typical service (an HTTP API or message consumer, one to a few CPUs, containerised), which combination of runtime-level signals (a continuous low-overhead recording, GC logs, thread-pool and heap counters, exception counts) and application-level signals (traces, metrics, structured logs) should be enabled by default, and which have teams switched off again because cost or noise outweighed their use?

What a useful answer contains

The runtime and version, the container CPU and memory limits, the exact signals and sampling settings, the measured overhead (CPU, memory, disk or egress) together with how it was measured, at least one incident in which a default-on signal shortened diagnosis, and any signal that was later disabled and why. Vendor material and single benchmarks should be labelled as such; answers that only restate documentation should say so.