In UTF8 files with 3-byte characters (e.g. japanese text), MadEdit displays phantom characters as first characters in the first visible line whenever the first visible bytes are continuation bytes of the last multi-byte character in the previous (not displayed) line.
In other words: whenever a multi-byte character is line-wrapped into the next line, and the user scrolls down so that he sees only the second half of this character, MadEdit interprets these incomplete bytes as characters of their own, displays 'phantom characters' which are not really contained in the text, and which disappear when the user scrolls up.
In UTF-8 documents, MadEdit should read back the previous bytes to be able to display the correct character, or alternatively highlight these bytes (e.g. by background color) as intentionally not displayed.