DEC XIE 2.3 HEADERS FOUND (peer session) AND THE REAL CAUSE OF EVERY BLOCKER.

H) THE AUTHORITATIVE HEADER
  xie/docs/dec/xie23/include/dec/extensions/XieProto.h
     XieK_MajorVersion 2 / XieK_MinorVersion 3   <- matches the VT1200
  Found on DEC OSF/1 2.0 DECstation + 1.3a AXP media by the sibling session and
  copied here with the man pages (84 Xie*.3X, DEC 1992).

  The 2.3 minor table (XieProto.h:99-...):
    1 InitSession   2 TermSession   3 QueryEvents   4 SelectEvents
    5 SetOpDefaults 6 QueryOpDefaults 7 CreateByReference 8 CreateByValue
    9 DeleteResource 10 QueryResource 11 BindPhotomap 12 AbortPhotoflo
   13 ExecutePhotoflo 14 TapPhotoflo 15 AbortTransport 16 GetStream 17 GetTile
   18 PutStream 19 PutTile 20 SetTransport 21 Export 22 FreeExport 23 Import
   24 QueryExport 25 Area ... 33 Crop ... 40 Mirror 43 Rotate 44 Scale
   45 Translate 46 LastRequest

  CONSEQUENCES -- three long-standing misidentifications, all now corrected:
   * slot 20 is **SetTransport**, not "arm/execute".  sz_xieSetTransportReq =
     140 bytes = 35 words, which is exactly the fixed length I measured and
     could not explain.  Every "arm" experiment was really a transport setup.
   * slot 3 is QueryEvents, NOT SetOpDefaults.  My "SET_OP_DEFAULTS(bpp=1)" was
     a QueryEvents reply all along.  SetOpDefaults is minor 5 (20 bytes).
   * ExecutePhotoflo is minor 13 (8 bytes), which I never sent.
   * There is no PutClientData in 2.3 at all; and "QueryTechniques" does not
     exist either.

I) STRUCTS THAT MATCH MY MEASUREMENTS EXACTLY
   xiePutTileReq  (32 bytes = 8 words):
      photomap_id, x, y, width, height, plane_mask, final(BOOL), pad[3]
      -> the word I had been calling "band" (word 5) is **plane_mask**.
   xieExportReq   (64 bytes = 16 words):
      photo_id, drawable_id, gc_id, src_x, src_y, width, height, dst_x, dst_y,
      photo_lut_id, colormap_id, match_limit, gray_limit
   xieCreateByValueReq: resource_type, resource_id (this build wants 12 words)
   Element type numbers (Xie.h): ImportClientPhoto=2, ImportDrawable=4,
      ImportPhotomap=7, ExportDrawable=33, ExportClientPhoto=31

J) THE BREAKTHROUGH: CLIENT-REACHABLE PutTile, NO DEBUG POKES
   SetTransport parameter sweep (plane_mask=1, pixel_stride[0]=1,
   scanline_stride[0]=8, compression_scheme=0, component_organization=4
   (=XieK_BitByPlane)):

     mode=0 -> BadValue        mode=1 -> ok, ctx[0]=1, PutTile a7=1 BadMatch
     mode=2 -> ok, ctx[0]=2    PutTile a7=2 BadMatch
     mode=3 -> ok, ctx[0]=3    PutTile ** ok/silent, a7=3 **   <-- PASSES
     mode=4,5 -> BadValue

   So: SetTransport copies `mode` into transport-context byte 0, and PutTile's
   gate (0x6BAF4) requires that byte to be 3.  With mode=3 the import succeeds
   with NO pokes -- the entire "debug-poke-dependent" blocker is gone.
   object->byte36 becomes 0x00291FB0 from a CLIENT REQUEST (was always 0).
   ctx bytes measured: 01 04 00 00 00 00 00 00 01 00 00 00 ... -> ctx[0]=mode,
   ctx[1]=component_organization, ctx[8]=plane_mask low byte.

K) EXPORT NOW ENTERS THE RENDERER
   With a populated photomap, Export (minor 21, 16 words, w=64 h=64, unbound
   photomap so it is "performed immediately" per XieExport.3X):
      export -> ok/silent ; RP(0x1B296) fires ; ** TYPE5 store fires **
   and with w=0,h=0 the renderer is skipped entirely (silent, no trace).
   This is the first time the renderer has been entered from a client-driven
   request.  BUT: src_base (rp+0x610) is still 00000000, and framebuffer captures
   (dwt/screenshot.py) for a solid-white tile and a solid-black tile are
   effectively identical -- the 1600 differing pixels are the Terminal Manager's
   memory figures and the pointer cursor, not our 64x64 window.

STATUS.  Still NO client-chosen XIE image on screen.  Window unchanged; white
and black identical in substance; xshow (core X) remains the only viewer that has
displayed an image.  What changed this turn is decisive for the mechanism: the
transport layer is now correct and PutTile imports real client data with no pokes.
