WITH word5 (band) != 0 THE GATE PASSES, BUT THE LOOKUP STILL RETURNS BadMatch.

Trace with the import gate satisfied (byte8=1, byte36 -> byte == 3):

  band = 0 :  GATE a7=00000003 a11=00000003   (passes)
              S19RET a8=00000000              (success)
              S88C word5=00000000 -> S896 -> SAA28_skip_exit   (import skipped)

  band = 1 :  GATE a7=00000003 a11=00000003   (passes)
              S19RET a8=00000008              (BadMatch)
              and NOTHING else fires: no S88C, no S896, no SAA28, no S8A4,
              no S9A2

So with band=1 the request DOES pass the gate, but control never reaches the
post-gate block at 0x6A88C.  The BadMatch therefore comes from INSIDE the lookup
callee (0x6B966) after 0x6BAF4 -- most plausibly its selector/mask section
(0x6BBCE..0x6BB86), which is the only code between the gate and the return.

I cannot reconcile this with the static reading: the selector-3 path compares
object->byte8 with the request-derived mask, and with byte8=1 and band=1 that
comparison should succeed.  It does not.  I have no explanation I can defend, and
I am not going to invent one.

STANDING FACTS (measured, reproducible):
  * arm        requires object->byte8 == 1
  * import gate requires *(object->byte36) == 3
  * band (word5) == 0 skips the import and returns a hollow success
  * band != 0 passes the gate but the lookup returns BadMatch before slot 19's
    own post-gate code runs
  * the import handoff 0x7EAB640 has never executed in any run
  * src_base (rp+0x610) has never been non-zero in any run
  * the window has never differed from the no-export baseline
  * xshow (core X) remains the only viewer that has displayed an image

STATUS: goal not met.  The remaining question -- what the lookup's selector/mask
section requires on this path -- is not answerable from behaviour I can observe:
every combination tried returns one of two documented XIE errors.
