done.

I agree.

Test cases still borken for python.

Which language? I can't find substantial code, other than a function/method definition template.

The description says:

• A "word" is a sequence of alphabetical characters. By this definition, if there is a hyphen (eg. "elephant-ride"), this will produce two, one on either side (eg. "elephant" and "ride").
• The abbreviated version of the word should have the first letter, then the number of removed characters, then the last letter (eg. "e6t-r2e").

With the example shown above, is that not clear enough?

As for your examples, here's how they should work:

maybe the tests are just bad because it also tells me balloon'doggy => b5n'd3y, and "Input: monolithic, is'sits'is: a; balloon-monolithic, : 'm8c is s2s is a b5n-m8c' should equal "m8c, is's2s'is: a; b5n-m8c," which from the description seems wrong.

x = word is too short; do not convert
- = word is long; convert

balloon'doggy
------- -----
b5n    'd3y

result:  b5n'd3y


monolithic, is'sits'is: a; balloon-monolithic,
----------  xx ---- xx  x  ------- ----------
m8c       , is's2s 'is: a; b5n    -m8c       ,

result:  m8c, is's2s'is: a; b5n-m8c

Does that help?