BinaryOpNode supports broadcasting of not just scalars, but any length-1 array. However this leads to some errors when using high demensional length one arrays. E.g.,

model = Model()
lhs = model.set(10)
rhs = model.constant([[[0]]])
expr = lhs <= rhs

print(expr.shape())

model.states.resize(1)
with model.lock():
    lhs.set_state(0, [0, 1, 2])
    print(expr.state().shape)

(1, 1, -1)
(3, 3, 129)

neither of which are sensible.

Further, because we don't store the shape on the data class we cannot even express the shape correctly in the above example.