GATE REACHED WITH byte8 == 3 (first time) -- AND ITS OPERAND IS NOT byte8.

Method: create w9=0x300 (byte8=3), then force object->byte36 non-zero with a
LABELLED DEBUG POKE (not a client-driven state), then PUT_TILE.

  poke byte36 = 0 (control):
     STORE src120=00000000 ; OUTCEL star=00000000 ; (no T_GATE) ; ERROR 10

  poke byte36 = 1:
     STORE src120=00000001 ; OUTCEL star=00000001 ;
     T_GATE a7=00000000 a11=00000003 ; ERROR 8          <-- GATE REACHED

  poke byte36 = p44 (003DBAF0):
     STORE src120=003DBAF0 ; OUTCEL star=003DBAF0 ;
     T_GATE a7=00000000 a11=00000003 ; ERROR 8          <-- GATE REACHED

So the out-cell hypothesis is CONFIRMED: byte36 non-zero => out-cell non-zero =>
the out-cell test at 0x6BAB8 passes => the gate at 0x6BAF4 executes even with
byte8 = 3.

BUT THE GATE STILL FAILS: it compares
     a7 = 00000000   with   a11 = 00000003
and 0 != 3 -> BadMatch.

THIS OVERTURNS MY GATE MODEL: a7 is NOT byte8.
  byte8 = 1  ->  a7 = 1
  byte8 = 3  ->  a7 = 0     (not 3!)
So the "3" the gate wants belongs to some OTHER field, whose value happens to be
1 when byte8 is 1 and 0 when byte8 is 3.  Every earlier statement of the form
"the import gate requires byte8 == 3" was therefore wrong -- the gate does not
read byte8, it reads whatever a7 is derived from.

WHAT IS NOW ESTABLISHED (all measured):
  * object->byte36 is uninitialised by create (0 for every w9 in one run, and
    non-zero for one allocation in an earlier run);
  * out-cell := object->byte36 (store at 0x6BAAC);
  * non-zero byte36 makes the gate REACHABLE at byte8=3;
  * the gate then compares a7 (0 at byte8=3, 1 at byte8=1) against 3 and fails.

NEXT: identify what a7 is derived from at 0x6BAEA..0x6BAF2
      (6BAEA MOVE *A11,A7,1 ; 6BAEC SETF 8 ; 6BAEE MOVE *A13(0xff40),A11,0 ;
       6BAF2 MOVE *A7,A7,0 ; 6BAF4 CMP A7,A11)
where A11 = *A13(0xffa0), and find the field that must equal 3.

Status: no client-chosen XIE image; src_base 0; window still the stipple.
