INSTRUMENTATION VALIDATED, AND THE byte8 GATE MODEL IS RECONFIRMED.

A) INSTRUMENT VALIDATION (harness/mame/xie_val.lua, 10 breakpoints)
   All 10 reported "Breakpoint N set", 0 "Error in command", and the control
   breakpoint CTRL_REGIDSTORE fired on create.  With this set, a single PUT_TILE
   produces the FULL ordered trace:

     T_ENTRY(0x6B966) -> T_DISP(0x6BA8C) -> T_C2(0x6BA1E) -> T_J1(0x6BA22)
       -> T_B32(0x6BA32) -> T_MERGE(0x6BAA2) -> T_GATE(0x6BAF4) -> S19RET a8=8

   The gate at 0x6BAF4 DOES execute.  My previous turn reported 0 hits for it and
   I withdrew the model on that basis -- that was a dropped-breakpoint artifact,
   and the withdrawal was wrong.

B) THE MODEL, RE-MEASURED (asserting replies, byte(N)=ptr+N*8):

   create w9=0x100 -> byte8=1, byte9=1
      arm1     -> SILENT(ok)
      PUT_TILE -> ERROR 8 ; trace reaches T_GATE ; byte8 after = 1

   create w9=0x300 -> byte8=3, byte9=3
      arm1     -> ERROR 8            <-- arm fails
      PUT_TILE -> ERROR 8 ; trace does NOT reach T_GATE ; byte8 after = 3

   So, with trustworthy instrumentation:
     * byte8 == 1 : arm succeeds, PUT_TILE reaches the gate and fails there
     * byte8 == 3 : arm FAILS, and PUT_TILE does not even reach the gate

   That is the same mutual exclusion reported earlier in the session:
     arm   requires object->byte8 == 1
     import requires object->byte8 == 3 (gate 0x6BAF4)
   and only create (word9>>8) sets byte8; a write watch on byte8 across a full
   cycle recorded zero writers.

C) NET: the model is RESTORED and now rests on validated instrumentation and
   asserted replies.  The blocker is unchanged and re-verified: the two requests
   need opposite values of a byte that nothing but create can set.

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