CORRECTION to evidence/class3_registration_impossible.txt.

That file claimed op=1 never reaches the registering body and that class and
registration are "mutually exclusive".  THAT IS WRONG.  Breakpoints on the two
registration sites show op=1 DOES register:

  bpset 0x3F53D820  (file 0x67B04  MOVE *A9(0x40),*A11(0x0))   ID store
  bpset 0x3F53D9A0  (file 0x67B34  CALLA 0x07F16CA2)            table register

Hits per request (one connection, real sequence):

  create op=2 0040D000 -> REGIDSTORE id=0040D000 obj=003FDC50 ; REGISTER
  create op=1 0040C000 -> REGIDSTORE id=0040C000 obj=0036A950 ; REGISTER
  arm1 / put / arm2 / slot43 / export / arm1(C1) / put(C1) -> 0 hits each

So the registration sites are reached ONLY by create, and BOTH opcodes reach
them.  What differs is WHICH object gets registered -- measured by peeking the
pointer each REGIDSTORE reports:

  op=2  id=0040C002 -> obj 003FD7D0
        id@0=0040C002  class=2  w=64  h=64  b8=01 b9=01 b40=00000000
        slot24 query -> REPLY              (addressable)

  op=1  id=0040C001 -> obj 0032F210
        id@0=0040C001  class=1  w=3259088  h=1  b8=90 b9=AF b40=00790352
        slot24 query -> BadAccess (10)     (registered, but not a drawable
                                            resource -- the fields are not a
                                            resource layout)

So:
 * op=2 registers a class-2 RESOURCE, correctly laid out, addressable.
 * op=1 registers a class-1 PHOTOflo under the client's id.  Its w/h are not a
   geometry, and query returns BadAccess because the found object is not a
   resource of the requested kind.
 * The class-3 object measured earlier (e.g. 0x3FD110, class 3, id@0 = 0) is an
   INTERMEDIATE that is never the registered object.

Therefore the accurate statement is NOT "class 3 is never registered" but:

   the id is registered to a class-1 photoflo (op=1) or to a class-2 resource
   (op=2); no create opcode registers a class-3 object.

Separating photoflo from image was then tested across five configurations
(I = op=2 image id, F = op=1 photoflo id), each with a 64x64 window and the
export render-params breakpoint still armed:

  armF putI expF  white : arm1 OK  put BadAccess(10) arm2 OK  exp OK  src_base=00000000 chg=False
  armF putI expF  black : arm1 OK  put BadAccess(10) arm2 OK  exp OK  src_base=00000000 chg=False
  armF putF expF  white : arm1 OK  put BadIDChoice(14)        exp OK  src_base=00000000 chg=False
  armI putI expF  white : arm1 OK  put BadMatch(8)            exp OK  src_base=00000000 chg=False
  armF putI expI  white : arm1 OK  put BadAccess(10)          exp OK  src_base=00000000 chg=False

Arming the photoflo changes the import error from BadMatch (8) to BadAccess
(10) -- a distinct state, and the only new behaviour found -- but src_base
stays 0 and no window changed in any configuration.
