THE STORE AT 0x125528 **IS** FILL-DEPENDENT -- THE ENGINE PORT IS THE PIXEL STREAM.

Read back the destination immediately after each store (breakpoint at 0x12552A,
before B1 is advanced), one export per fill:

  fill=ff : 384 stores -> FFFF x283, 0000 x26, 1F00 x4, FF00 x4
  fill=00 : (cumulative 768) -> FFFF x294, 0000 x83, 3200 x17, 3FBD x11
            subtracting the white run: black alone is ~11 FFFF and ~57 0000

So the engine output port 0x5600200 IS the pixel stream: a white tile yields 283
words of 0xFFFF stored at B1, a black tile yields 0x0000.  The value read is the
value stored -- the store is not losing anything.

That resolves the question this turn was asked to settle: 0x5600200 is not a
mis-modelled status/echo register; it carries the pixels.

WHERE IT IS THEN LOST
  The staging buffer's own write-watch (0x4FE780..0x497F, 512 bytes) shows 0xFFFF
  x260 for BOTH fills -- i.e. after our 283 fill-dependent stores, the buffer is
  overwritten by constant allocator/manager traffic (0x5555, 0x0000, 0x4100 from
  the always-present writer PCs).  And the byte-level VRAM diff between white and
  black is ZERO.

  So the tile is written correctly into the staging buffer and then discarded; no
  step copies it to video memory.

STATE
  * NO client-chosen XIE image has ever been displayed;
  * fill-dependent proof now extends one step further: client -> photomap ->
    accelerator FIFO -> engine output -> CPU store into the staging buffer;
  * from there it is overwritten, and VRAM never differs (0 bytes / 128 KiB);
  * model opt-in (VT1200_ACCEL=1); default configuration unchanged;
  * xshow (core X) is the only viewer that has shown an image.
