URDNA2015 Implementation Question

Hello, I'm working on implementing the RDF Dataset Canonicalization
algorithm[0], using the https://github.com/digitalbazaar/rdf-canonize
implementation as a reference.

I've got an implementation with all the tests passing except evil(1),
evil(2), and evil(3), and I was wondering if you could clarify a point for
me - step 5.4 of the Hash N-Degree Quads algorithm which states: "For each
permutation of blank node list:".

I've instrumented the rdf-canonicalize library so I can inspect the order
of execution, and it appears that what differs between my implementation
and the Javascript one is the order of the permutations. The spec doesn't
say how the permutations should be ordered, and my intuition is that the
order does indeed matter - though I'm happy to be corrected if I'm wrong.

So, here is my question(s):
- Does the order of the permutations matter?
- If so, what order should they be in?

The ref-canonicalize library implements the Steinhaus-Johnson-Trotter
algorithm, which I'm happy to do as well if necessary. But if the specific
order matters I think it should be a part of the spec.

[0]
https://json-ld.github.io/rdf-dataset-canonicalization/spec/#normative-references