THE OUT-CELL IS FILLED FROM object->byte36 -- AND THAT FIELD IS NOT INITIALISED
BY CREATE.

A) SOURCE IDENTIFIED (store at 0x6BAAC, breakpoint at 0x3F55D560):
     byte8=1 : obj=0028E050  b36=00290510   STORE src120=00290510  -> out-cell = 00290510
     byte8=3 : obj=003FD7D0  b36=00000000   STORE src120=00000000  -> out-cell = 00000000
   The store's source register A7 IS the object pointer, and it reads
   d@(A7+0x120) = u32(obj, byte 36).  So out-cell := object->byte36.

B) BUT byte36 IS NOT SET BY CREATE.  Creating with every w9 value in one run:
     w9=000000 byte8=0 b36=00000000
     w9=000100 byte8=1 b36=00000000
     w9=000200 byte8=2 b36=00000000
     w9=000300 byte8=3 b36=00000000
     w9=000400 byte8=4 b36=00000000
     w9=000500 byte8=5 b36=00000000
   All zero -- yet an earlier run's w9=0x100 object had b36=00290510.  So byte36
   is whatever the heap block contained; it is NOT a field the create path sets.

C) CONSEQUENCE: whether the import gate at 0x6BAF4 is even REACHED depends on
   uninitialised object memory.  The "byte8=1 fills the out-cell, byte8=3 does
   not" correlation I reported last turn was an artefact of which allocation
   happened to carry a non-zero byte36 -- not a property of byte8.

D) WHAT THIS MEANS FOR THE GOAL: to reach AND pass the gate we need a byte8=3
   object whose byte36 happens to be non-zero.  Create does not control byte36,
   and a 27-combination sweep of the other create fields (positions 2..10, values
   1/2/3, w9=0x300) produced byte36 != 0 in NO case.

So the import path depends on an uninitialised field, and the only lever is heap
layout (what the allocator hands back), not any request parameter.  That is
consistent with a firmware-side uninitialised-field dependency in this path.

Status unchanged: no client-chosen XIE image; src_base 0; window still the
stipple; xshow (core X) is the only verified viewer.
