PUT_TILE state gate, single MAME instance, live resource found by scanning the
XIE table entry (ID + handler 0x3F566ED0, resource pointer at +12).

Write watchpoints during PUT_TILE (res, band, resource+44, band+0x30):
  WPHIT res  addr=003FDC80 data=00000000   ; resource byte 48 = 0
  WPHIT band addr=008F6750 data=00000001   ; band byte 12 = 1
No other writes in those ranges.

Slot 20 compares band+0x60 (byte 12) with request byte 16 (v16) and requires
the AND to be zero.  Therefore after PUT_TILE the second execute must use
v16 = 0:

  arm1 v16=1 -> OK
  put          -> OK
  arm2 v16=0 -> OK
  arm2 v16=2/3/4/0x10 -> BadMatch

So the post-put gate is solved: band byte 12 = 1, second arm v16 = 0.

BUT export is still independent of the pixels:
  solid-white nz=512 first=88888888888888882222222222222222
  solid-black nz=512 first=88888888888888882222222222222222

Binding band+0x30 (which is zero) to the resource+44 pointer by rdebug poke
also produced the same pattern.  The remaining gap is the destination/sink for
the tile data and the export draw path, not the arm-2 gate.
