SLOT-19 LOOKUP RE-TRACED WITH ASSERTING HARNESS -- and it produces an
INTERNALLY INCONSISTENT result that I could not resolve.

REPRODUCIBLE TOP-LEVEL FACTS (two trials, identical):
  create -> SILENT(ok); arm1 -> SILENT(ok); PUT_TILE -> ERROR 8 (BadMatch)

TRACE (trial 1 and trial 2 identical):
  S19CALL        0x6A7AE   slot 19 calls 0x7EAB966 (file 0x6B966)
  CALLEE_ENTRY   0x6B966   callee entered
  CALLEE_DISP    0x6BA8C   class dispatch reached (class 2)
  CALLEE_C2      0x6BA1E   class-2/3 branch entered
  S19RET         0x6A7B4   a8 = 00000008

So the callee IS entered, DOES reach the class dispatch and the class-2/3 branch,
and returns 8.

THE CONTRADICTION:
  A full scan of 0x6B966 (330 instructions) finds exactly THREE sites that set 8:
     0x6BAFA  MOVK 0x8,A7     (fall-through of the JREQ at 0x6BAF6)
     0x6BB4C  MOVK 0x8,A11
     0x6BB88  MOVK 0x8,A7
  Instrumented with breakpoints at the corresponding points (0x6BAF4 the CMP
  before the first, 0x6BB4A, 0x6BB86) in this run:
     SEL_DISP 0x6BBCE -> 0 hits
     MASK_A_8 0x6BB4A -> 0 hits
     MASK_B_8 0x6BB86 -> 0 hits
     ERR_0A   0x6BAC4 -> 0 hits
     OK_PATH  0x6BBE2 -> 0 hits
  and in the previous run the gate CMP 0x6BAF4 also had 0 hits.

  So the function returns 8 without executing ANY of the three 8-producing sites,
  and without reaching the selector dispatch, the 0xA site, or the success exit.
  Every transition I can see between 0x6BA1E and the exits is accounted for, and
  none of them yields 8.

WHAT I AM NOT GOING TO DO: pick one of those sites and assert it is the source.
The measurement says they do not run.

Conclusion: the same class of instrumented-vs-observed contradiction that has
recurred through this session is present here, and I have not found its cause
(candidates: dropped breakpoints in this MAME build, or the return value being
delivered by a path outside 0x6B966).  Until that is understood, the origin of
PUT_TILE's BadMatch(8) is NOT established, and the byte8/gate model remains
withdrawn.

Status unchanged: no client-chosen XIE image; window still the stipple;
xshow (core X) is the only verified viewer.
