I was trying a solution that would calculate the ratio the cut-off point as following: sqrt(shortestDot * longestDash), where dashes have to be strictly higher. I considered it a good idea as it allows both dot and dash the same ratio of error.

But my simple heuristic with dividing the longest dash by two worked a lot better for the most part, and combined with a few more heuristics, I covered it all but the weird case where "1001" is supposed to mean ". .", quite inconsistently with the 4 kyu Kata. So I kinda gave up on trying a more interesting solution and I agree that this Kata could have been more interesting if the test cases weren't so strict in some rather ambiguous or straight out controversial cases.

It's related to caching, it's not a kata issue.

Your points should be updated now.