How do you choose a StringComparer for a Dictionary, HashSet, sorting, or user-facing search?

In production I define one normalization and comparison policy at the boundary and reuse it across validation, storage, caches, and collections. Lowercasing strings manually can allocate and can produce different results across cultures; a comparer usually expresses the intent better. Unicode normalization may still be required when visually equivalent text can arrive in different forms, and database collation must agree with application uniqueness rules. I test difficult cases rather than only ASCII, especially Turkish casing, composed characters, and identifiers that cross services. Display ordering, search matching, and identity equality are distinct product decisions and should not share a comparer accidentally.