Grassmann.jl A\b 3x faster than Julia's StaticArrays.jl

Could somebody please help me figure out what’s going on with this performance variance?

So what’s exactly going on here? Why is it faster when I paste it into the REPL?

This doesn’t happen with the Base.in or method, which is fast without pasting into the REPL

julia> @btime $(barycenter(A)) ∈ $A
  61.499 ns (0 allocations: 0 bytes)

So, for the linear solve method there is a performance gain from pasting into REPL but not for .

Can anyone help me figure out why the linear solve method changes in performance like this?