Yes, the ConditionalWeakTable approach for binding metric collection to service instances is completely analogous to shadow logging and is ideal, providing complete decoupling of business logic from metrics.
ConditionalWeakTable<TKey, TValue> allows you to weakly bind metrics (e.g., IMetricsCollector) to a service as a key, without the risk of memory leaks—the collector is automatically deleted when the service is GC'd. It's important to note that the table requires unique keys, so if you have more than one service object for your key object (Logger, MetricsCollector, and so on), they should be stored in the table as a list of references, such as List< object >. In the article for loggers, this is implemented using a DI container (AddScopedWithLogger), and the same pattern is used: in the DI extension method, a table is created, a collector is added for each scoped service, and the business class generates events (such as MetricsRecorded) to which the shadow object subscribes.