Dear author, you can republish this one now considered the fact that there's random tests and more active beta stage nowadays!

You can check out several of the higher quality kata that you have already completed in Python and study how they handle producing random tests.

How do I generate truely random tests? If the seed can't change every run through it will always be possible to just make an array of the answers and bypass the problem. An idea I had was to tie the random seed to some attribute of the user's code, like number of characters or something. Alternately a pointer to the top of the stack might work but that's not possible in Python I don't think.

No random tests. I suggest you unpublish immediately and add them. There is documentation to help.

Should I change the name to something more descriptive?

           test.assert_equals(switches(toggles), switches(toggles))

You're comparing the user's solution return value with itself.

The only thing I can think would be to test the user solution against a known functioning one. Is there a way to do that?

Yes, using a reference solution:

@test.describe('Random tests')
def random_tests():
    def reference_solution(toggles):
        ...
    def generate_random_input():
        ...
    @test.it('a test')
    def _():
        toggles = generate_random_input()
        expected = reference_solution(toggles)
        actual = switches(toggles) # user solution
        test.assert_equals(actual, expected)
    ... 

https://docs.codewars.com/authoring/guidelines/submission-tests/#reference-solution
https://docs.codewars.com/languages/python/authoring#example-test-suite

I wrote a random input generator but am unsure how to test. In other examples I saw, the result has some property that can be tested against. Like testing an absolute value function would have a random input and the test would be whether the result is greater than zero. But with this problem I don't know of any property this can be tested against without linear algebra which isn't supported with codewars libraries. The only thing I can think would be to test the user solution against a known functioning one. Is there a way to do that?

https://docs.codewars.com/authoring/guidelines/submission-tests/#random-tests

What is a random test?

The number of lights is already encoded in the list length, so there's no point in passing it in another argument.
Or maybe the number of switches can be different from the number of lights, but then this case should be tested. It doesn't looks like that ("only describes what OTHER lights").

Random tests should be added.