ROOT CAUSE, MEASURED END TO END.

  render setup (render driver, file 0x1B152):
      0x1B256  MOVB *A5(0x20),*A11(0x630)     ; rp byte 198 <- element byte 4
  DDX render (file 0x35EA8):
      0x35F0E  A8 = *A11(0x630)               ; read it back
      0x35F12..0x35F1E  accept 0,1,2,3 only
      0x35EF2  -> "DDX_RENDER: unsupported compression scheme" -> BadValue

  LIVE TRACE (genuine DEC 2.3 client, export to a depth-1 window):
      W256  a5=003FD590  val=55  rp=00291550
      RD198 rp=00291550  b198=55
      -> DDX_BADVALUE_class -> DDX_RET a8=00000002 -> client BadValue

So the chain is exact:
  element byte 4 (uninitialised = 0x55 heap fill)
    -> copied verbatim into renderptr->compression_scheme
    -> DDX_RENDER's very first validation rejects it
    -> export returns BadValue before src_base is even examined.

The render-params block IS written, and the writer is identified; the value it
copies is garbage because the BAND ELEMENT's byte 4 was never initialised.
(The element is 0x3FD590, the same band element whose +0x28 holds the imported
image; its byte 4 should be the compression scheme, XieK_PCM = 0.)

WHY THIS EXPLAINS EVERYTHING SEEN
  * every client-side attribute is correct (transport mode 3, scheme PCM 0,
    levels 2, organisation, drawable, GC, depth 1, geometry, imported data
    verified 0xFF white / 0x00 black);
  * both the hand-rolled client and DEC's own 2.3 client fail identically,
    because the defect is in server-side element initialisation;
  * changing component_organization, destination (window vs pixmap), colormap,
    lut, geometry or client library never changes the outcome -- none of them
    touch element byte 4.

STATUS.  NO client-chosen XIE image has ever been displayed.  The window is
unchanged; white vs black captures differ only by Terminal Manager figures and
cursor.  xshow (core X) remains the only viewer that has shown an image.

NEXT (precise): find who should initialise the band element's byte 4 -- i.e.
which server path sets the element's compression when a Photomap's Sink is
created -- and why that path did not run for a photomap populated via PutTile
with XieK_PCM.
