SLOT 20 (arm) USES THE SAME LOOKUP AS SLOT 19 BUT PASSES A DIFFERENT TYPE BYTE.

Slot 20 (file 0x6AA54) calls 0x7EAB966 at 0x6AAB2 -- the identical lookup slot 19
uses -- but the arguments differ.  Push order at 0x6AA8A..0x6AAAE:

  1  MOVI 0xffff                      (7th from last)
  2  *A9(0x68)   = request BYTE 13    (6th)   <-- slot 19 pushes the literal 3
  3  *A9(0x20)   = payload word 0 = id(5th)
  4  A13+0x40    = out pointer        (4th)
  5  A13                              (3rd)
  6  A13                              (2nd)
  7  client                           (1st)

In 0x6B966's frame that makes *A13(0xff40) = request byte 13, and the gate at
0x6BAEE/0x6BAF4 compares the object byte against it.  Request byte 13 is payload
byte 9 -- the field this client has always set to 1 (p[9], the "mode").

MEASURED (4 breakpoints: ID store, class dispatch, gate CMP, lookup-null):

  arm, w9=0x100 (byte8=1)  -> CLASSDISP=1 cls=02  GATE=0  arm succeeded
  arm, w9=0x300 (byte8=3)  -> CLASSDISP=1 cls=02  GATE=0  arm = ERR8
  arm, w9=0x300, mode=3    -> CLASSDISP=1 cls=02  GATE=0  arm = ERR8
  put, w9=0x100 (no arm)   -> CLASSDISP=1 cls=02  GATE=0  put = BadAccess

So:
 * arm DOES find the resource and DOES reach the class-2 dispatch.
 * arm NEVER reaches the 0x6BAF4 gate -- GATE=0 in every arm case, including
   byte8=3.  Its BadMatch therefore comes from somewhere else, earlier or in a
   different branch.
 * mode=4 gives BadValue(2), so the mode field is validated (1..3 accepted).

Also measured this turn (clean instrumentation, byte N = ptr + N*8):

 * byte40 is 0 for BOTH the class-2 resource (op=2) and the class-1 photoflo
   (op=1).  Earlier reports of a non-zero byte40 are NOT reproducible and are
   retracted.
 * Poking byte40 to a valid pointer (the object's own address) does not change
   arm's result for either byte8=1 or byte8=3, so byte40 is not the arm gate
   either.

STATE: arm fails on a byte8=3 resource after the class dispatch, for a reason
that is neither the 0x6BAF4 gate nor byte40.  The next step would be to bisect
0x6BA1E..0x6BAA2 (the class-2/3 branch) during an arm on a byte8=3 resource,
since that is now the only unmeasured stretch on the arm path.
