A BOM should only occur in files, but a lot of CM/MD parsing is not file-based (think <textarea>). If a parser does not enforce a certain output encoding it should probably keep the original BOM. For parsing purposes, it must always be treated as if it weren’t there.