DEFINITIVE FINDING: XIE's RENDER PATH NEEDS THE IMAGE ACCELERATOR, WHICH THIS
MACHINE DOES NOT MODEL.  ITS REGISTER WINDOW IS SWALLOWED BY THE VRAM MAPPING.

A) THE CONVERSION ROUTINE THE EXPORT ENDS IN
   The post-gate trace showed the failing callee:
       r6CF4 a8=00000002        ; call at 0x36CEE -> file 0x122D94 returns 2
   file 0x122D94 (ddx/smi conversion) begins by driving hardware registers:
       0x122DA2  MOVE A11,@0x5600000,0x0
       0x122DA8  MOVE A11,@0x5600100,0x0
       0x122DBC  *A11(A9) reads @0x5600100, ANDI 0xfffd, writes back
       0x122DD8  MOVE 0x258,@0x5600020,0x0
       0x122DE2  MOVE 0x32, @0x5600030,0x0
       0x122E24  MOVE A11,@0x5600080,0x0
       0x122E44  *A11 reads @0x5600030, OR 1, writes back
   i.e. the VT1200's image accelerator (the "scalution" device the firmware also
   names: "error initializing scalution") at bit window 0x5600000.

B) THAT WINDOW IS NOT A DEVICE IN THE DRIVER
   harness/mame/vt1200.cpp:
       1127:  map(VRAM_BASE_BITS, 0x1fffffff).ram().share("vram");
       VRAM_BASE_BITS = 0x04000000
   so 0x04000000..0x1fffffff is plain shared VRAM and **0x5600000 lies inside it**.
   Reads therefore return pixel data, writes land in video memory, and the
   accelerator never behaves like hardware.

   Confirmed with the driver's own unmapped-access log (VT1200_LOG_UNMAPPED):
   during a full export the log contained ONLY its 2 header lines -- no entry for
   0x5600000 -- precisely because that address is "mapped" (as RAM), not unmapped.

C) WHY THIS EXPLAINS EVERY NEGATIVE
   * the conversion polls accelerator registers expecting status bits; with VRAM
     contents it never sees them and returns BadValue(2);
   * every client-side attribute is correct and unchanged across ~45 turns
     (transport mode 3, scheme PCM, photomap levels 2 / comp_COUNT 1, drawable,
     GC, depth 1, geometry, imported data 0xFF white / 0x00 black);
   * both the hand-rolled client and DEC's own 2.3 client fail identically;
   * the two earlier DDX gates (compression, src_base) are upstream of this and
     can be made to pass, after which this hardware access is what remains.

STATUS.  NO client-chosen XIE image has been displayed, and in this emulator it
cannot be: the display path terminates in a hardware device that is not modelled.

WHAT WOULD BE REQUIRED
   A model of the VT1200 image accelerator (register block 0x5600000.., and the
   pixel-engine semantics behind "scalution"), mapped over that address range and
   with the VRAM window narrowed around it.  That is a substantial new device to
   implement -- not a client or protocol issue.
