What you say seems a bit unclear for now.

• Afaik, the soluiton is matching perfectly the description. Can you provide a concrete example of wrong delimiter?
• About the "twice" problem, that's sounds rather unlikely to just run into the situation because it would require a randomly generated word to be the exact double from another randomly generated one. Still possible, yes. But considering the number of random tests, that shouldn't even be noticeable.
• ...the randomly generated test cases measures all occurences of the returned strings as substrings in the input text.

-> nope:

TopWords.top3("mhvhdmhvhd mhvhd mhvhd mhvhd mhvhdmhvhd")
    
# here is what my solution is counting in there (I created the random tests):
{mhvhd=3, mhvhdmhvhd=2}

So, doesn't match at all what your saying.

Really sounds like an invalid issue, so closing. If I'm wrong, reopen a fresh one, but this time, do it "right":

Seems you're "rather new" to cw, so here are some general guidelines about the comments:

• Issue: problem in the kata itself (description, wrong tests, wrong internal solution...)
• Suggestions: well, I guess that part is clear
• Question anything else that is related to you having a problem solving a kata -> that's you, currently.

When you post issues:

• provide ALL the useful information:
  • language
  • input
  • outputs (actual expected) when relevant
  • error message when relevant
• check, DOUBLE check that this IS an issue, meaning that the problem is in the kata itself and not in your code. If it's in your code, post rather a question
• if you pass this step, you still have to prove/explain what the issue is (and if you can provide fixes, it's even better)

When/if you post a question: well, most of the above apply too x)

When/if you post code, use proper github markdown, so that it's readable.

cheers