PUT_TILE SUCCEEDS -- FIRST TIME IN THIS INVESTIGATION.

The gate operand chain is now fully decoded:

  0x6BAEA  A11 = *(caller frame slot)      -> A11 = the out-cell's ADDRESS
  0x6BAEC  SETF 8                          -> 8-bit field
  0x6BAF2  A7  = *A7                       -> ** A7 := *(object->byte36) **  (deref!)
  0x6BAF4  CMP A7,A11  (A11 = 3)           -> requires the BYTE AT (byte36) == 3

Measured:
  SCRATCH=0x00010100 ; poke *SCRATCH = 0 and 3 ; poke object->byte36 = SCRATCH

  *scratch = 0 :  F2 a7=00010100 star7=00000000 ; GATE a7=0 a11=3 ; PUT_TILE ERROR 8
  *scratch = 3 :  F2 a7=00010100 star7=00000003 ; GATE a7=3 a11=3 ; PUT_TILE SILENT(ok)

So when the byte pointed to by object->byte36 equals 3, the gate compares 3 with 3,
passes, and the import request returns success (A8=0, no error packet).

FULL CHAIN ATTEMPT (both tiles, debug pokes labelled):

  white: create w9=0x300 -> arm1 ERROR 8 -> PUT_TILE SILENT(ok) -> arm2 ERROR 8 -> EXPORT ERROR 8
  black: create w9=0x300 -> arm1 ERROR 8 -> PUT_TILE SILENT(ok) -> arm2 ERROR 8 -> EXPORT ERROR 8

So:
  * the IMPORT now works,
  * arm (slot 20) still fails on a byte8=3 resource (its own check at 0x6ACBE),
  * EXPORT consequently fails, RP never fires, src_base cannot be read.

IMPORTANT CAVEAT: the successful import depended on two DEBUG-ONLY pokes
(object->byte36 and the byte it points at).  Nothing here shows a client request
can produce that state.  It proves the mechanism, not reachability.

WHERE THIS LEADS: exactly the same structure almost certainly applies to arm's
check at 0x6ACBE (`A7 = *A11(0x40),0` then DSJS), which I previously read as
"requires byte8 == 1".  Given that the import gate turned out to be a DEREFERENCED
byte rather than byte8 itself, arm's check should be re-read the same way before
concluding it needs byte8==1.

Status: PUT_TILE success achieved (debug-poked state); no image yet; src_base
still unread because export fails; xshow remains the only viewer that has shown
an image.
