If I understood correctly, you mean to calculate payout distribution with current data but with new rules. I see one issue here: different rules introduce different behaviors, so the result of simulation would be inaccurate even probably far from real effects, which makes the simulation less valuable.
By the way, perhaps it's good to try this with comment reward pool first.
Yep, that is why it was my third choice. Much better would be the first two, using some bots and / or real people doing it just to help out.
In fact it could be a blind test where participants don't know which proposal has been implemented in the sandbox. Perhaps everyone can be rewarded for participating, since the actual rewards in the sandbox would not be real.
Simulation has its value. But without real money involved, the results are usually inaccurate (I don't know how far it will be though). You'll know the difference if you've ever played with market simulators, for example stock markets or currency markets or futures.
I agree with you.