Live-resource scan attempts:

  0x60000..0x180000, filter ID + +36/+44 non-zero:
    hits 0x7FB8A and 0x7FAB2, but byte4 (class) = 0
    -> these are copies/table entries, not the resource

  0x3E0000..0x400000, filter ID + class2 + width=64 + height=64 + +36/+44:
    no match

The live resource pointer is not stable across runs and the field-filtered
scan does not find it reliably.  The earlier known address (0x3FDC50) was a
heap allocation from a previous run; it is not present in this run.

Consequently the class-byte patch test could not be applied to the real
resource, and the real PUT_TILE still returns BadMatch (S19LOOK a8=8).
