Ask the same question the role asks
The target of the test is one of four categories, not the broad good/evil alignment. In the normal grouping, Innocent and Meddler are both good categories. A negative category result therefore cannot remove every good possibility.
A reliable positive result supports the actual Innocent category at the time of the test. It is still not a universal promise of true statements or permanent good alignment: taint and alignment-changing effects are different mechanics.
A correct negative result on a good Wannabe
Suppose a normally good Wannabe at address 9 is showing a revealed Underling. The actual category is Meddler. A reliable Researcher rejecting “Innocent” is correct in that case.
The contradiction only appears if you replace the question with “good?” in your head. Write the actual question beside the answer: “Innocent category? No.” Then use a separate clue for current alignment. A name that sounds reassuring is not a substitute for reading the test.
The source also needs a reliability check
A displayed Researcher can be a disguise, and an Innocent information source can be tainted in the Demo rules. Establish the reliability needed for the conclusion or keep the result conditional.
If your problem is which good-looking role is a disguise, this category test may help narrow the identity but will not necessarily settle the whole case. Choose the next clue according to the field that remains unresolved: category, exact role, current alignment or information state.