I changed (again) the way the assertions are made, hopefully providing a better feedback, now (reset the trainer if you don't see any difference)

yeah, I feared a bit about that type of cases, where the message becomes a bit misleading... The problem is that the real count is:

("'vxn", 10)
('ansace', 12)
("das'", 20)      <<<< you're returning "das" instead, which isn't present in the count so has a 0 frequency.
("dtqo'a", 15)
('fhuubd', 3)
('lunceifkq', 2)
('yyaqfy', 24)

yyAQfy;DAS'!ANSAce.DTQO'A.; -DTQO'A?yyAQfy/?ANSAce-/,-?DAS'.?-.yyAQfy,?-/:'vxN.DAS',!!luNCEIFKq, -ANSAce:yyAQfy; luNCEIFKq_-?DAS'-.:!yyAQfy:DTQO'A!yyAQfy-,!'vxN;DAS'.DAS',,ANSAce:.DAS'/;DAS':? -:yyAQfy/- ?/DAS' ;?/DAS';:'vxN,DTQO'A-?ANSAce/,ANSAce.? DAS'?DTQO'A.DAS':;,! ANSAce;yyAQfy- -:,DAS'?_yyAQfy;'vxN//!yyAQfy ;??!DTQO'A-!.fhUUbD/;DAS':DTQO'A;:DTQO'A/-DTQO'A:yyAQfy?'vxN:.;/:'vxN,'vxN-,yyAQfy!-yyAQfy-DAS';?DAS' :'vxN, ! yyAQfy:.-/'vxN.,?-/fhUUbD,..DAS': ;DTQO'A;;yyAQfy?!? .DTQO'A:!!?ANSAce;?:yyAQfy..!yyAQfy:?.-DAS' :.?ANSAce.fhUUbD,/!/ANSAce;.:.yyAQfy ./ANSAce - !;ANSAce.!.:yyAQfy!yyAQfy;.DAS'. :DTQO'A;.;DTQO'A.DTQO'A?!.!.yyAQfy.yyAQfy?,DAS':_DTQO'A _.,yyAQfy_yyAQfy?'vxN?!:
{'yyaqfy': 24, 'das': 20, 'ansace': 12, "dtqo'a": 15, 'vxn': 10, 'lunceifkq': 2, 'fhuubd': 3}
[(24, 'yyaqfy'), (20, 'das'), (15, "dtqo'a")]
['yyaqfy', 'das', "dtqo'a"]
Frequencies of the most common words: [24, 0, 15] should equal [24, 20, 15]

I have:
[(25, "yowr'fr"), (17, 'zxytghkac'), (16, 'unleqfxsu')]

My final return will give the strings like this:
["yowr'fr", 'zxytghkac', 'unleqfxsu']

When it is tested for the numbers, it looks up "yow'fr" which gives the value 0. This makes the test attempt give a [0,17,16] expected [25,17,16]
I cannot change yowr'fr in any way because it has an apostrophe in the middle. Is this an issue?

The sample gives ("//wont won't won't") and I got ["won't", 'wont'] so when the string "won't" is given, wouldn't that give a 0 as well?

IIRC the strings should be split like "sie@##@1/1/1fsefsef" == ["sie", "fsefsef"], so no, you're wrong.

yes, I handled that case as you described it

you don't have to "filter out", you have to "extract" the words:

" //wont won't won't " => ["wont", "won't", "won't"] extracted, so ["won't", "wont"] as an output

then based on this sample test:

Test.assert_equals(top_3_words(" //wont won't won't "), ["won't", "wont"])

are we not required to filter out characters that are from A-Z or (')?

invalid issue:

• without the actual input and your output, it's impossible to tell if that's an issue in the tests or if your code is just wrong.
• language?

for instance, those values are the frequencies of occurences of different words. So [25,25,23] means that there are 2 different words that occur 25 times and another that occur 23 times. While your ouput matches words that do not occur the right amont of times. I'd bet on your code not cutting the sentence correctly.

Frequencies of the returned words are compared to the frequencies of the expected words - I don't see any issue.

oh, right.

Tho, the version shouldn't have been available. I guess it was a "reset the trainer" thing again.

Because from random import choices. For some reason 3.4 was available for me earlier...

and why so?

Python 3.6 should be made the only language version.