You cannot read octets from a file and then hope that Perl will know that you meant UTF-8. You always have to decode your input and encode your output.

Maybe we talk past each other?

My utf8 to iso-8859-1 should do the following:
- read file into scalar
- eval decode utf8
-- if we get an error, then i simply say it's iso-8859-1 and I don't do anything
-- it we get no error, then i use the decoded string

On the other script iso-8859-1 to utf8, I wanna go the same way:
- read file into scalar
- eval decode iso-8859-1
-- if we get an error, then i simply say it's utf8 and I don't do anything
-- it we get no error, then i use the decoded string

Now I expect we get an eval decode iso-8859-1 error if the input file is utf8, but I don't.

-- if we get an error, then i simply say it's utf8 and I don't do anything

That's wrong.

If your input is bytes that are a valid UTF-8 sequence, you still need to call decode on it to have usable strings in Perl.