Reports the identifiers of the span that is active RIGHT NOW, so a signal emitted outside any span-creating call — a log record, most of all — can name the trace it belongs to.

This is the read that joins a log line to a span. withSpan hands its own span to its callback and nothing else, so without this a consumer that is merely running inside a span has no way to ask which one; every signal the framework emits is individually good and mutually unjoinable.

Optional, and its ABSENCE is reported rather than substituted. A service that cannot see an active span omits the method, and a consumer then enriches nothing rather than emitting empty identifiers that would join a record to a trace that does not exist. Returning undefined from an implemented method is the narrower answer: "I can see, and nothing is running".

Implementations must not throw: a caller on the logging path cannot usefully handle a failure to describe the trace, and turning the observability read into an availability fault is strictly worse than emitting an unjoined record. Callers guard anyway (a replaceable capability can be anything), but the contract is no-throw.