#Emacs uses the Unicode Character Database (UCD) compiled into its binary. Every codepoint has properties β category, bidirectional class, case folding, combining class, etc. Emacs loads this at startup and uses it for rendering, input methods, and font selection. Agda goes further β it uses the UCD to determine which codepoints are valid identifiers, operators, or whitespace, since
#Agda allows Unicode operators like β β β’ Ξ» directly in source.
Fontconfig then maps codepoints to glyphs via font lookup. If no font covers the codepoint you get a replacement box.
#GraphQL OTOH has no native binary or raw-text type β everything is a JSON string, so any unescaped byte that's invalid in JSON (control chars, malformed UTF-8 sequences like your 0xD6 case) breaks the entire response. Emacs and Agda never have this problem because they read bytes directly and interpret encoding explicitly β they don't depend on an intermediate serialization format like JSON that has strict validity rule