A spy fails by being caught, and it is normally only through being caught that spies reveal their methods.
Conclusion
Available data on spies are skewed: you can learn a lot about bad spies but very little about good spies.
Evidence
You only really get data on spies if they're caught, and only bad spies get caught.
Evaluate
Although there are some hints at Conditional Logic here because of the "only through" and "never", the conclusion is not trying to derive anything conditionally, so I think I would resist trying to diagram this and think more about the gist-y principle behind it.
The author is essentially saying that our data set on spies is inherently biased in a certain way based on the nature of collecting data on spies. It's almost like the data we have on spies is a self-selecting sample of data on "bad spies".
By definition of us having data on you, you're a bad spy.
Goal
Because this argument seems to be working more on a conversational level of "unrepresentative sample", I wouldn't be as stringent about forcing the answer choice to have all the same structural similarities.
I basically want to hear about a situation in which our data is skewed because the nature of how we collect the data makes it very likely that we'll see one subpopulation overrepresented.