TURN RECORD: WHERE EXPORT'S BadValue COMES FROM, AND WHY IT IS NOT THE FLOATS.

A) EXACT REJECTION SITES IN THE EXPORT HANDLER (server, file 0x62D40)
   BadValue (MOVK 0x2) is emitted at 0x62DCA, 0x62E06 and 0x62E3E, each preceded
   by a comparison of a request word against a field of a looked-up object:

     0x62DBC  A7 = *A9(0x80)        ; a request word (16 bytes into the data)
              A11 = *A11(0x80)       ; the same offset in the looked-up object
              CMP A11,A7 ; JRC 0x62DF4   else BadValue
     0x62DF8  A7 = *A9(0xa0) / A11 = *A11(0xa0) ; CMP ; JRC  else BadValue
     0x62E30  A11 = *A9(0xc0) ; JREQ 0x62E3C      ; word == 0
              A11 = *A9(0xe0) ; JRNE 0x62E68      ; and the next word == 0
              else BadValue
   So the rejection is a RANGE/COMPATIBILITY comparison between export operands
   and the resource's own attributes -- not a decode failure.

B) THE XieFloatRec HYPOTHESIS IS DEAD
   Sent match_limit and gray_limit as properly encoded 1.0
   (e_sign=0, m_sign=0, exponent=1, mantissa=0x80000000 -> words
   0x00010000,0x80000000) and as 0.0, full 15 data words, everything else
   identical:
     match/gray = 1.0/1.0 -> export err2, src_base 00000000, RP/TYPE5
     match/gray = 0/0     -> export err2, src_base 00000000, RP/TYPE5
   Identical.  The floats are not the cause.

C) WHY THE LIBRARY WOULD NOT HAVE HELPED EITHER
   The 2.3 client library on disk is MIPS (osf1-2.0-decstation/libXie.a) and
   Alpha (osf1-1.2-alpha/libXie.a).  It builds requests by filling the very
   structs already read out of XieProto.h (photo_id, drawable_id, gc_id, src_x,
   src_y, width, height, dst_x, dst_y, photo_lut_id, colormap_id, then the two
   XieFloatRecs), so disassembling it cannot change the byte layout -- only
   confirm it.  The failing comparison is server-side range checking, which the
   library does not encode.

D) STATE OF THE TWO PATHS (unchanged, re-measured this turn)
   unbound photomap: transport ok, puttile ok, export err2, renderer entered
                     (RP + TYPE5) with src_base = 00000000
   bound photomap  : bind ok, export err1 x6, execute err1 x6, no renderer trace

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

NEXT, concretely: identify which object the export handler looks up at 0x62DBC
(it is not obviously the photomap) and which of its fields at 0x80/0xa0/0xc0/0xe0
is being compared, then set the export operands inside those ranges.  That is a
register/field identification in one handler, not a protocol question.
