This is a real drawback of this kind of architecture, because now you have another (likely central) team trying to figure out the structure and meaning of events from another team (upstream).

That gets worse as the meaning and structure changes over time, as those changes are made by the upstream team without knowledge of the logic downstream.

It's the reason why many of the ideas around "schema on read" are falling out of favour, because in practice it's a large amount of effort and the results are poor.

We do use an event driven architecture and make use of append-only tables, but we ensure those match the schema of the event as it was generated, upstream. We also implement change management on that schema, via data contracts. That means what lands in BigQuery is the same as what is generated, and can be used instantly.