TURN RECORD: THE src_base CHAIN IS FULLY IDENTIFIED, AND THE IMPORT DOES PRODUCE A
SOURCE -- BUT THE TRANSPORT ORGANIZATION DECIDES WHETHER IT SURVIVES.

A) WHERE src_base COMES FROM (static)
   0x1B238  MOVE *A5(0x28),*A11(0x610)
   i.e. rp+0x610 = *(element + 0x28 bits).  Measured at that store during a
   working export:
     SRCBASE_ST a5=003FD590  elem28=00000000  rp=009261D0
     RP         a11=009261D0  src_base610=00000000
   So the renderer reads the source pointer straight out of the band element.

B) THE IMPORT **DOES** WRITE IT
   Instrumented slot 19 during a successful PutTile (mode=3):
     S896 plane_mask=00000001
     S8A0 bit0test
     S8A4_import_loop
     S98E_ELEM28_STORE val=00982B70     <-- element+0x28 written
     S9A2_handoff_call
     H_ENTRY
     SAA28_skip_exit
   and then element+0x28 = 00000000 again.  The handoff clears it at
   0x6B8D6 (`SUB A9,A9 ; MOVE A9,*A11(0x28)`) on the path it takes there.

C) THE HANDOFF'S PATH IS CHOSEN BY component_organization
   The handoff dispatches on (*(byte36))[1] (0x6B83A, bit offset 8 = BYTE 1).
   byte36 is the transport context, and its byte 1 = component_organization.
   Measured, after SetTransport(mode=3) + PutTile:
     org=1  transport ok  puttile ok  elem+0x28 = 0098CAB0   (SURVIVES)
     org=3  transport ok  puttile ok  elem+0x28 = 0098CAB0   (SURVIVES)
     org=4  transport ok  puttile ok  elem+0x28 = 00000000   (CLEARED)
     org=2  transport BadMatch        puttile BadAccess
   So XieK_BitByPlane (4) -- the value I had been using -- is exactly the one that
   makes the handoff clear the source pointer.

D) BUT WITH org=1/3 THE EXPORT FAILS EARLIER
   org=1 or 3 with plane_mask=1: elem+0x28 = 0098DB10 (non-zero), then
     export -> err2 and the renderer breakpoint (RP 0x1B296) NEVER fires,
     i.e. the export handler rejects before reaching the render driver.
   org=4: elem+0x28 = 0, export -> err2, RP fires with src_base 0.
   So the two halves of the failure are complementary: the setting that keeps a
   source also breaks the export request path, and vice versa.

E) ALSO MEASURED
   plane_mask must be 1; 0xFFFFFF fails the transport (BadMatch) and PutTile
   (BadAccess) for both org=1 and org=3.

STATUS.  NO client-chosen XIE image.  The 64x64 window is unchanged; src_base has
never been observed non-zero at the renderer.

This is the most precise statement of the wall so far: the import produces a real
source buffer (0098CAB0 / 0098DB10) inside the element, and the only thing that
destroys it is the handoff branch selected by component_organization -- while the
value that preserves it also makes Export bail out before the renderer.
