thrice in my benchmark. it operates inplace

I hit this issue as well. After a couple dozen attempts I managed to pass all the tests anyway.

Afterwards I tried out the reference solution against my own, and the reference certainly has some bug in it. For example, reference::elder_age(3,3,1,100) returns 5 instead of the correct 6 (easily verified by hand).

As far as rust is concerned, mut a is not a change of signature.

Is this solution really twice faster than approach with iterators?

I'm not sure how the term "signature" is strictly defined, but for all practical purposes I wouldn't consider mut a part of a signature. The vector is borrowed the same way and mut refers to a local variable, so the difference can't be observed outside.

Is it ok to change function signature?