My wrong solution successfully get through all 60+ test cases.
Initially, there were 2 errors in ANIMALITY AND ANIMA (reason - incorrect way of blocking used cells).
I have not yet solved this problem, just tried with reverse order and it accepted (tested many times).

If somebody can review my code and tell me why my trail of used cells not resetted to savepoint with recursion collapse, and acts like it global.

Well, anyway, more complex tests are needed.

UPD: Added "True" true solution.

with outside words_dict had 25 passed and [TypeError: 'function' object is not subscriptable] on random tests
after moving this dict inside function - 125 passed and no error

Can't submit this kata when using split string.
'RUN TESTS' expanded to all variations of tests is working fine (21 Passed; 0 Failed; 0 Errors; Process took 101ms to complete)