At least in java the kata doesn't appear to be broken.
The second test block (should_test_predefined(ReverseTest)) does not seem to test for a correct string result; it does, however, test for behavior (it tests that the function has been called the correct number of times). So it's a little misleading because you can technically pass this test block with the wrong result but the right behavior.
However, the first test set (should_tetst_30_randomized(ReverseTest)) tests for both.
If you follow the instructions and you return the right result, you can pass both blocks and complete the kata.

Heavy reliance on array methods (especially reduce) slows this function down, because each "reduce" executes a loop over an array. Once you've calculated the sum to the right and left of the index, you don't actually need to re-calculate ("reduce") the sums on both sides from scratch when you move the index, because the array values are fixed.

Is there a reason for choosing parseInt over Math.floor for rounding down?