WHY CLASS 3 IS UNREACHABLE -- the ID is never stored.  This also retracts the
"byte 40 == 0" explanation: byte 40 is a copy-out field, not a gate.

A. slot 8 (create, file 0x668E4) dispatches on request word 0:
     6696C DEC A8 ; JREQ -> 0x66912      op 1 ---\
     66970 DEC A8 ; JREQ -> 0x66912      op 2 ---/  SAME routine 0x7EA78DA
     66974 SUBK 2 ; JREQ -> 0x66902      op 4 -> 0x7EA76CA
     66978 DEC    ; JREQ -> 0x66922      op 5 -> 0x7EA7B46
     6697C DEC    ; JREQ -> 0x66932      op 6 -> 0x7EA7DA8
   So op 1 and op 2 are the SAME create code (file 0x678DA).

B. Inside 0x678DA (file), the only op-dependent value is:
     6799C  MOVE *A9(0x20),A11,1     ; word0 = op
     679A2  SUBK 0x2,A11
     679A4  MOVK 0x1,A11
     679A6  JREQ 0x679AA             ; op==2 -> arg = 1
     679A8  SUB A11,A11              ; else   -> arg = 0
     679AA  ... push arg, w, h, and bytes at request offsets 44/43/42
     679D6  CALLA 0x7F638C2          ; allocator
     67A20  MOVE *A9(0x20),A11,1
     67A24  DSJ A11,0x67AFC          ; loop runs `op` times (op = level count)

C. Measured objects (RESPTR 0x6CC0E gives the allocation deterministically):

   create op=1  ID=0040B301  object=003FCC90
     words: 00000000 00010103 00000101 00000000 00000040 00000040 ...
     word1 = 0x00010103 -> class byte 3
     id@0 = 0x00000000     *** ID NOT STORED ***
     ID string 0040B301 appears NOWHERE in the object's first 0x100 bytes
     slot24 query -> ERR10 (BadAccess),  PUT_TILE -> ERR14 (BadIDChoice)

   create op=2  ID=0040B302  object=003FDA10
     words: 0040b302 00010102 00000101 00000000 00000040 00000040 ...
     word1 = 0x00010102 -> class byte 2
     id@0 = 0x0040B302     ID stored
     slot24 query -> REPLY,            PUT_TILE -> ERR10 / ERR8

   ID position sweep: only payload word 1 is accepted.  Putting the ID in any
   other word makes create fail with ERR14, so word 1 IS the ID field.

D. Retraction of the byte-40 rule.  The shared lookup helper (file 0x66FA0)
   has two variants:
     6707E  MOVE *A7(0x140),*A11(0x0),1   ; non-NULL out-param: COPY byte40 out
     6708C  MOVE *A7(0x140),A7,1 ; JREQ    ; NULL out-param: require byte40 == 0
   The "must be 0" test only exists in the variant with a NULL out-parameter,
   and it raises 0xE, not 0x8.  The 0x8 (BadMatch) we observe comes from the
   table lookup 0x7F174A4 / the class compare, not from byte 40.  So patching
   byte 40 could never have fixed PUT_TILE, which is exactly what was measured.

E. Falsified during this check: the word at parent+48 is NOT a child object.
   At 003FD6F0 it reads as heap-block metadata followed by the terminal message
   "Connection request from IP host 192.168.99.1".  There is no class-3 child.

CONCLUSION: PUT_TILE demands a class-3 resource that is registered by ID, and
the only class-3 creator (create op=1) never stores the ID, so it can never be
found.  The missing piece is the request that both selects class 3 and stores
an ID -- the create routine's `(op==2)?1:0` flag is the only knob found so far.
