It cannot work. Whatever rule you come up with related to inactive stake, you can easily replicate either through bribing, or just might happen by accident.
In addition, the attackers could easily be miners as well.
You cannot cobble together a chain selection rule + some other bizarre ad-hoc rule for detecting history attacks; they have to be one and the same rule, otherwise this will just be to easy to exploit due to complexity.
Remember that between two chains, the suspecious part only appears at the forked place where we need to focus on. That means you have to limit the history point to the part where the main chain fluctuates. You can't control the history point so easily, can you?
In addition, the attackers could easily be miners as well.
Being a miner is not enough. He has to buy as many active miners' accounts as possible. The purpose of mentioning the miners was to explain that there should not be too many quiet periods because the voting cycle of miniers is about an hour. (In case you didn't read through the paper, the voting cycle of the stakeholders is about 100 hours, so that the meaning of "active" in our system is an actual cyclically-publishing activity, which is different than the common sense of the meaning of "active" .)
And the miners are a group of people who are continuously benifiting from the system, not like the long since cashing out stakeholders. They should not be so easy to buy over.