The AST string table is WTF-8, not UTF-8: an unpaired UTF-16 surrogate is encoded as the three bytes `ED A0-BF 80-BF`, which strict UTF-8 rejects. `Encoding.UTF8.GetString` replaces those with U+FFFD, so identifiers and string literals carrying a lone surrogate come back silently corrupted. TypeScript permits them, so we decode them ourselves.