THE BYTE-SWAPPED DISPATCH FAMILY IS NOT CLIENT-REACHABLE.

slot 47's table entry is 0x3F54BDA0 (file 0x697B4), a trampoline that swaps
the bytes of request words at bit offsets 0x10/0x18, 0x20/0x28, 0x30/0x38 and
then re-dispatches:  SLL 5 on the minor byte ; ADDI 0xd3f90 ; MOVE *A11,A11 ;
CALL A11.

Breakpoints armed on both the table read (0x3F54BFF0) and the CALL (0x3F54C000),
i.e. the trampoline body itself.  Breakpoint count went 36 -> 38, so both were
installed (no "Error in command").

Requests sent (one connection, after creating a class-2 resource):

  minor47 len3  plain            -> ERR1 (BadRequest)
  minor47 len16 byte-swapped args-> ERR1 (BadRequest)
  minor47 len16 plain args       -> ERR1 (BadRequest)
  control minor19 len16          -> ERR10 (BadAccess)

  TRAMP hits: 0
  ("TRAMP minor=.. target=.." never printed; count = 0)

So minor 47 is rejected with BadRequest before the trampoline body ever runs,
for both swapped and unswapped argument words.  The trampoline is therefore
internal/dead for a client: there is no client-reachable byte-swapped request
family, and the swapped-import route is closed.

Related diagnostic: the default handler 0x64C3C (used by every unimplemented
slot, and by slot 0) is the BadRequest responder -- that is why probing the
unimplemented slots returned ERR1 rather than a length error.

Also note: the trampoline's own self-reference is why it can never be entered
from a client -- it re-dispatches on the SAME minor byte it arrived with
(*A10(0x8) is not touched by the byte swaps, which only cover offsets 2..7),
so a client request for minor 47 would re-enter it forever.

One measurement caveat recorded honestly: in this run the opening `create`
returned no reply (timeout), and the control import then returned BadAccess
rather than the usual BadMatch.  That does not affect the trampoline result --
TRAMP never fired even though the server was still answering requests -- but it
means this particular run's create state is not a clean baseline.
