done

OK, let's do it.

Thanks! I think I fixed all mentioned.

I've added complete test set.

Please share your opinion how kata should work (and why :-)).
From what I think you mean I tend to disagree.
There seem to be many variants of kata on this site.
The one with only initial tests seem quite common and I liked them.
But I may be biased, I didn't solve that much of them.
I think it involes some good practices if you have to write your own tests (a little like TDD approach).

I've added a note that input is correct in context of the cipher algorithm. Thanks for checking this kata out.

Such case will never happen. :-)

Increased the tests to 110

I'm not sure about this.
The initial tests are very simple on purpose.
All the test cases are expressed in the description with precise examples.
Writing your own test cases is part of the work on the solution (tests are implementation).
Please mark the issue as resolved if you confirm.

As for the second comment, please explain.
It is not possible, because occurence is based on the letter position in a word.
There is only one letter at each position.
There can be words with specific letter repeated, like BELLY (double l).
Do you mean such case?

Edit: I updated description and republished (was not sure if it will save my changes)

Hey, thanks for checking this kata out. Please review again, should be fixed.

The repeated access to string is cache friendly, so has a good chance to be stupid fast.
On the other hand, memory allocations are stripped to (what seems to be) a minimum.
The clever avoidance of count and no intermediate collection to store the counting is appealing to the audiance. :-)
Also, codewars community tend to favor 'less code' approach. :-))
That's my understanding, but would like to hear someone more experienced in Rust.

Run it locally, looks good. Can't approve though. Is anybody here who can? It's quite nice kata for training Rust.

Because of random tests, this solution sometimes does not finish before server timeout.