Apparently, graph walking is not necessary for this problem, there are very few path strategies for a symmetric web.

Problem description is too vague and uncertain. Not a single mention of the desired result. It was just broken code, comments and some clues from tests. Please improve the description, at least to the level when it's possible to think of solution without starting the kata and looking at the unfinished code.

It's not algorithm problem, it's pure math. Suggest to remove tag Algorithms, add tag Mathematics and reduce difficulty level to 7 kyu.

Horrible solution. Perfect example of how to kill performance by making code shorter. Solutions here are expected to have O(n) and compete only in number of passes through the list, but this one manages to get to O(n^2).

There is an issue for original kata and you better not inherit it. There really should be explicit tests with repeating minimum in the array. For example [3, 4, 3, 5] should produce 6, not 7. I guess, single fixed test is ok.

The solution is incorrect. It fails if the minimum element repeats anywhere in the array.

Ants NEVER collide more than 2 at once in one place. In this case particularly the middle one first collides with the right one and then (after changing direction) with the left one:
(time goes from top to bottom)

.R..R..L.
..R..RL.. <- first collision
...R.LR..
....!..R. <- second collision
...L.R..R

There is a harder version of this kata. https://www.codewars.com/kata/65e4b8877e8acd0dbae9a14f

There is wider kata about computing regular expressions for divisibility tests. https://www.codewars.com/kata/5993c1d917bc97d05d000068

Instead of having single @Test with loop inside (it is actually many tests!) use @RepeatedTest instead.

Added to description and tweaked slightly some tests

Good point! Done