According to your link, everything should be hunky-dory, because FR pages use UTF-8 encoding.
But what is happening is that the FR server is now converting incoming UTF-8 to garbage by replacing the individual bytes of each multi-byte UTF-8 group with the corresponding HTML entity. Thus, for example, the UTF-8 for the left double curly quote, e2 80 9c, gets replaced by three entities, circumflex-a, the Euro sign, and the oe ligature. As a result, the browser, even though it is correctly in UTF-8 mode, doesn't see the pristine UTF-8 character, but instead the three bogus characters.
If what you say is correct, then it sounds like something has corrupted the translation processing possibly? Strange though since you can remove the character, retype it and it comes out correctly. Definitely a strange problem. Is there anyone at FR that can step through the conversion process, if that is even possible, to determine what exactly is happening? Being a mainframe person, not really knowledgeable if stepping through a process is even possible in this realm.