In terms of what you mean by realism, do you mean what would actually happen as a consequence of actions being done or a copy paste from real life data without considering the in game differences on various factors (policies, history etc)? It seems from the Sestak example, to be the latter scenario in which case, I wouldn't necessarily consider that to be an example of "realism" as I have applied the term for a good while.
Historically speaking, the "quest for realism" meant that in game actions had "realistic consequences or effects" not that the data matched real life for the sake of matching real life regardless of what might make it not be so.
Determining whether a simulated consequence or effect is realistic is only easy when speaking in very general terms: "X should result in an increase in Y and a decrease in Z" or "X policy will result in growth in Y sector of the economy." It is much more difficult to determine if something is
quantitatively realistic. In-game income tax revenues increased by 41 billion dollars between FY2019 and FY2020. Is this realistic or is it unrealistic? Is there an acceptable range of values that should be considered realistic? If the calculated value is too close to the RL value despite the numerous policy differences between game and RL, is this a coincidence born from a balancing out of several competing factors or should we be concerned about the model itself? Usually the only way to initially test if something is realistic (that is, if it would happen in a RL scenario) is to use RL numbers as a guide.
Perhaps the Sestak example was poorly written; it was intended to illustrate the fact that others' preconceptions of what is and isn't realistic are drawn from looking at what happens IRL. Sestak wasn't necessarily concerned that the numbers weren't a copy-paste of RL, he was wondering why the revenues were X amount lower than they were IRL. The answer, of course, was that I was trying to keep the year-to-year growth realistic in relation to the previous year's budget, which was done by a different DGM and had revenues that were much lower than RL. And my assessment of whether the quantitative year-to-year growth was 'realistic' was very generally based on current and historical RL growth statistics, combined with subjective assessments of how the past year's legislation would affect growth (based, once again, on my own RL knowledge and/or research). Oftentimes, comparison to RL numbers is the only way to determine if something is realistic or not. If I publish an income tax revenue estimate that is $200 billion dollars lower than the RL value, people will want to know why. And it is often difficult to isolate the 'why;' are the numbers lower because of some unforeseen quirk of the model (say, something fundamental to the way that I am calculating the numbers), or are the numbers lower due to the implementation of actual in-game consequences arising from the past year's legislation? The only way to establish a concrete starting point is to use your model to try to recreate (approximately) the RL numbers. But this reverse-engineering process can lead to plenty of problems down the road.