CAN ANY OTHER REQUEST RAISE byte8 FROM 1 TO 3?  No.

Clean instrumentation: harness/mame/xie_min.lua (ONE breakpoint: the ID store,
used only to learn the object pointer).  All object fields read via rdebug at
the CORRECTED offsets (byte N = ptr + N*8, because ptr is a bit address).

Per slot: fresh connection, create op=2 with w9=0x100 (which gives byte8=1),
read byte8, send <slot> with payload [id,0,...] at that slot's required length,
read byte8 again, then arm.

  slot  len  request        byte8 before -> after   arm
   43    10   BadValue(2)    1 -> 1                 (no reply read)
   44     5   BadValue(2)    1 -> 1                 (no reply read)
   40     4   BadValue(2)    1 -> 1                 (no reply read)
   33     4   BadValue(2)    1 -> 1                 (no reply read)
   24     4   REPLY          1 -> 1                 (no reply read)
   19     8   BadAccess(10)  1 -> 1                 (no reply read)

No implemented request moves byte8.  Only create does, and only through word 9's
high byte:  byte8 = (word9 >> 8) & 0xFF, with byte9 = (byte8==3) ? 3 : 1
(verified: w9 = 0x000,0x100,0x200,0x300,0x400 -> byte8 = 0,1,2,3,4).

Combined with the arm sweep (v12 = 0..6 x v16 = 0..3, 28 shapes):
  byte8=1 -> arm OK,  put = BadMatch (gate compares 1 with 3)
  byte8=3 -> arm ERR8 for every shape, put = BadAccess (unarmed)

HARNESS NOTE: in these runs the arm (and sometimes create) replies were often
not read back -- the reply stream gets misaligned once several requests are
outstanding on one connection.  That does NOT affect this result, because byte8
and the put error codes are read independently (rdebug peek / slot-19 reply on a
fresh connection).  Future runs should read replies with a drain between
requests, or use one request per connection.

CONCLUSION: within the client-reachable request set, byte8 cannot be raised to 3
while keeping the resource arm-able.  The import therefore cannot be driven to
the handoff, src_base stays 0, and EXPORT has nothing to draw.
