To me this seems easier than the alternative model (logs, metrics, traces), in that you only have one abstraction to contemplate handling PII in?

In what way is this worse than the alternative?