-T examines a file, and tries to guess if it's an ASCII text file. But what heuristic, exactly, does it use?

To see the implementation of it, search for "fttext" within pp_sys.c of the Perl source code.

Basically:

• the first 512 bytes of the file are scanend
• it evaluates each character in turn
  • if a null character is found anywhere, the file is considered binary
  • if less than 1/3 of the bytes are classified as "odd", then the file is considered text
  • "odd" is defined as:
    • UTF8 characters don't count as odd
    • characters that are <32 are odd, with several exceptions:
      • \n, \r, \b, \t, \f, \0x1B