I believe you showed in one of your demonstrations (or someone else showing ramda), a very simple elegant example:

    > sum = reduce(add, 0)
    > sum([1,10,20])

With something like trine, composition can only happen really happen at usage-time, while with Ramda it allows you to reason about the operations separate from the data. I suppose trine could achieve the same by further complicating with some sort of dummy data:

    sum = __placeholder__.reduce(add, 0);
    [1,10,20].::sum()

But this gets difficult because __placeholder__.reduce itself needs to return something that can be chained (certainly doable to return a function that is also a dummy object I suppose?). But anyways, to me this starts looking more and more complex, and making it harder to reason about constructing (composing) NEW functions from existing ones.