Let's say you write

df.filter { someValue > myColumn.max() }

This is way faster:
val myColumnMax = df.myColumn.max()

df.filter { someValue > myColumnMax }

Maybe we could solve this by having lazily calculated stats stored inside ValueColumns. Columns are immutable after all, so it would be safe to do so and the performance gain should be significant!

Of course, this wouldn't work when you write:

df.filter { someValue > (myColumn + 1).max() } // or df.filter { someValue > myColumn.maxOf { it + 1 } }

but that's okay I think. We can't have it all :)