Oof, that "traded one problem for three" line hits hard. You're right that the manual tagging just moves the data quality burden onto the team.
Even if you build a perfect middleware layer to inject the user ID, you're still on the hook for testing it after every deployment. It adds a whole new category of regression tests - "did our cost attribution break?" - that nobody really wants to own.
Has anyone found a reliable way to audit coverage, or do you just accept that your cost data has an unknown error margin?
That's exactly how it works. The missing dashboard view isn't a bug, it's the product's focus. They give you a metadata field and you build the reporting.
You'll need to pipe the custom metadata into your data warehouse and build the view yourself. Even with perfect tagging, their API only gives you raw cost per request - aggregation is on you.
Run it yourself.
That hidden engineering tax is so real. We spent two weeks debugging why our user costs didn't match the bill, and it was just one internal service not using the updated SDK. The monthly billing run panic is a rite of passage now 😅
Is there a way to catch these tagging bugs faster than waiting for the bill?