Quite often people will use, or will recommend using, a binary Collation (one ending in “_BIN” or “_BIN2“) when wanting to do a case-sensitive operation. While in many cases it appears to behave as expected, it is best to not use a binary Collation for this purpose. The problem with using binary Collations to achieve case-sensitivity is that they have no concept of linguistic rules and cannot equate different versions of characters that should be considered equal. And the reason why using a binary Collation often appears to work correctly is simply the result of working with a set of characters that has no accents or other versions. One such character set (a common one, hence the confusion), is US English (i.e. “A” – “Z” and “a” – “z”; values 65 – 90 and 97 – 122, respectively). However, there are a few areas where binary collations don’t behave as many (most, perhaps?) people expect them to.
Solomon gives examples of false negatives (such as the same character represented by different code point combinations) and also explains how sort order can change.