TURN RECORD: err2 IS THE RENDER DRIVER'S RETURN CODE.  EVERY EXPORT-SIDE STEP
SUCCEEDS.

Instrumented the whole export tail (slot 21, file 0x62D40) during a working
client-driven sequence:

  ALLOC ret=00321C50      ; 0x6301C..0x6302A allocate the export context: OK
  SETUP_before_render     ; 0x63072 fills it (type 5, photo_id at +0x48,
                          ;   geometry at +0x88/+0xa8)
  CALL_RD                 ; 0x63098  CALLA 0x07E5B152  == render driver 0x1B152
  AFTER_RD a8=00000002    ; <-- the RENDER DRIVER RETURNS 2 (BadValue)
  EPI a8=00000002         ; handler returns that value unchanged

Also confirmed on the same run:
  * the drawable lookup at 0x62F7E (0x7EA3432) SUCCEEDS, so the handler jumps to
    0x62FBC;
  * at 0x62FBC the frame word is zero, so the second lookup (0x62FDC, the one I
    was told to instrument) is SKIPPED -- it is never called on this path;
  * the extent check at 0x62F12 passes;
  * no BadValue site in the handler fires (EX_BADVALUE_1/2/3 never reached).

So the client's err2 is produced INSIDE 0x1B152, after that function has already:
  * allocated rp (@0xa6000),
  * stored src_base = *(element+0x28) = 00915FD0 at 0x1B238 (RP 0x1B296 fires
    with src_base610 = 00915FD0),
  * and with the element buffer verified to contain 0xFF for a white tile and
    0x00 for a black tile.

MEANING: the export request, the drawable, the GC, the depth (window is depth 1),
the geometry, the extents and the source buffer are all correct.  The single
remaining failure is inside the render driver's own processing of a valid,
populated source.

STATUS.  NO client-chosen XIE image.  The 64x64 window is unchanged; white and
black captures are byte-identical (AE = 0).

WHAT WOULD FINISH IT
  The next instrumented range is 0x1B296.. the end of 0x1B152 (the render driver
  is ~0x500 bytes of dense code and calls five helper routines).  That is the
  only place left where the BadValue can be produced.
