ARM CHECK TRACED -- and the printed operands CONTRADICT the static reading.

Addresses computed in code this time (my hand-computed ones were wrong by
0x8000, which is why the previous attempt showed no trace at all).

  w9=0x100 (byte8=1): arm1 = SILENT(ok)
     ACBA a11=00000002  b8=00 b12=00
     ACBE a7=FFFFFFE9
     ACC4 a7=00000000
     ACCA a7=00000000
     ACD0 a7=FFFFFFFE
     OKPATH_6AD34            <- success path

  w9=0x300 (byte8=3): arm1 = ERROR 8
     ACBA a11=00000002  b8=00 b12=00
     ACBE a7=FFFFFFE9
     ACD4_BADMATCH           <- rejected

THE PROBLEM: at 0x6ACBA the static reading says
     6ACBA MOVE *A13(0x20),A11,1     ; A11 should be an OBJECT pointer
     6ACBE MOVE *A11(0x40),A7,0      ; then A7 = object->byte8
but A11 prints as 00000002, i.e. not an object pointer, and `b@(a11+0x40)` reads
0 -- so the static reading of this block is WRONG.

Also, A7 at ACBE prints the SAME value (FFFFFFE9) for both byte8=1 and byte8=3,
yet the subsequent branch takes different paths.  So the value I am printing is
not the one that decides the branch, and I have not identified the real operand.

Therefore: the reason arm fails on a byte8=3 resource is NOT established by this
trace.  I am recording the negative rather than inventing an explanation.

WHAT IS STILL TRUE FROM LAST TURN (measured and reproducible):
  * with object->byte36 pointing at a byte equal to 3, PUT_TILE returns success
    (A8=0) -- the import mechanism works;
  * but arm still fails on that same byte8=3 resource, so EXPORT fails, RP never
    fires, and src_base remains unreadable;
  * the goal (a client-chosen image via XIE) is NOT met, and the successful
    import depended on debug pokes, not on client-reachable state.

NEXT would need either a correct reading of the arm check (all registers at
0x6ACBE, not just A7/A11) or the DEC 2.3 definition of what arm validates.
