The combination of inner and outer set analysis is not trivial. The announcement of this feature and the description within the help are creating an impression that they are replaceable to each other and simple in the usage but by reading it carefully there are some dependencies and exceptions mentioned - unfortunately without explaining it in detail.
If I remember correctly it states that by any kind of conflict the most inner set statement will win - overwriting the other set statements. I think what you have created is going in following direction:
{
(< [SRT] = {'val1'} >*< [SOCTN]={'val3','val4"} >)
+
(< [SRT] = {'val2'} >*< [SOCTN]={'val5'} >)
*
< [SRT] = {'val3','val4'} >
}
which shows that there is a conflict because [SRT] could have this value OR that values but not both at the same time. If it's applied within a single set statement it will lead to exclusions and maybe even to ZERO as result but by using it on the outer-side and the inner-side it's like a chain of selections - which means at first is the outer set statement = selection applied and then the inner one. If the inner selection doesn't fit to the outer set statement it replaced it.
Beside this the outer set statement isn't wrapped with brackets to a certain aggregation which means it will be applied of all following aggregations - also to this part: count(distinct[SOCTN]).
