Underlying type balancing for integral operands

[abgs]

What is the intended interaction between 1 and 3, particularly the transition from underlying type to actual type for the "unsigned operand"? This seems to lead to strange outcomes for code like a_u16 + b_u16 + c_u16 where there are expressions whose underlying type is unsigned while the actual type is signed. The only way I see to interpret 1 and 3 together as written would be as "if the essential type of either operand is unsigned, the result is the actual type of the essentially unsigned operand". Presumably either "either is unsigned" should refer to actual type, or "the result is unsigned" should refer to the essential type of the unsigned operand. The latter seems to produce the most consistent results when applied to the rules in 6.5.