THE BRANCH THAT SKIPS THE IMPORT IS FOUND: request payload WORD 5.

Slot 19, after the geometry check passes (0x6A87C main path):

  6A88C MOVE *A9(0xc0),*A13(0x40),0x1   ; A9(0xc0) = byte 24 = payload WORD 5
  6A892 MOVE *A13(0x40),A7,0x1
  6A896 JREQ 0x0006AA28                 ; word5 == 0  -> skip to exit
  6A89A MOVE *A13(0x40),A11,0x1
  6A89E BTST 0x0,A11
  6A8A0 JREQ 0x0006AA10                 ; bit0 clear  -> another path
  6A8A4 ...                             ; then the import loop (reaches 0x6A9A2)

My tile header has always been [id, x, y, w, h, band, flags] with band = 0, i.e.
word5 = 0 -- which takes the JREQ and exits without importing.  That is exactly
why 0x7EAB640 never ran and src_base stayed 0.

MEASURED with band (word5) varied, gate satisfied, byte8=1:

  band = 0  -> PUT_TILE SILENT(ok)   handoff trace: (none)
  band = 1  -> PUT_TILE ERROR 8      handoff trace: (none)
  band = 3  -> PUT_TILE ERROR 8      handoff trace: (none)

So word5 != 0 does move past the skip (the request now fails with BadMatch
instead of silently succeeding), but the handoff still does not execute -- a
further check on that path rejects it (candidate: the band/subobject lookup that
0x6A8A4.. performs).

ALSO CONFIRMED THIS TURN: the trace now shows the main path being taken
  S19_C0 -> S19_852 -> S19_87C_mainpath -> S19_EXIT
where earlier runs went S19_850 -> S19_EXIT directly, so the geometry condition
(A13(0xc0) > 0) is satisfied with a 64x64 tile at 0,0.

STATUS: no client-chosen XIE image.  Progress: the "success is a no-op" is now
explained (word5 == 0 skips the import), and the next obstacle is the BadMatch
raised when word5 != 0.
