I ran into this exact trade-off when migrating a legacy inventory system to event sourcing. The refinement mapping concept is elegant, but implementing it as a live SQL view over the event table killed our read performance once the volume picked up.

We eventually had to materialize the view to keep latencies acceptable, which introduces its own complexity with lag and invalidation. It seems like you always pay the tax somewhere—either in the query cost or in managing the state of the read model.