How should I move and process this data?

Start from what the consumer cannot currently do, not from what the source happens to produce. Reasoning in the other direction is how a platform fills with datasets nobody uses — and every leaf here states what its recommendation costs, because a recommendation without a cost is an advertisement.

A consumer says a metric looks wrong. Nobody debugs a data platform forwards — this walks upstream, and each branch ends in the mechanism that produced it plus the check that would have caught it.

What does "wrong" look like?