As per @Halfak's recommendation in T163995, please lower the "Very likely good faith" filter threshold from precision=.995 to .99.
Also, let's try lowering "Likely have problems" to precision=.45, and see what happens.
Together, these changes should improve our ability to find intersections between "Very likely good faith" and "Likely have problems," as discussed in T163757.
(I've already updated the ORES model properties spreadsheet to reflect this change. )