Just so we’re talking about the same thing, my points are:
- I believe a majority-hashrate attacker can use the timewarp attack to permanently increase the block rate beyond 1 per 600s on average, while keeping the hashrate constant, as long as they keep up the timewarping scheme. They can’t keep increasing exponentially or anything like that.
- With a grace-period limiting how much time can go backwards between the last block of a period and the first block of the next period, we limit how much that that increased rate is. If the grace period is G seconds, then the attacker can maintain a block rate of roughly one per \frac{2017 \cdot 600 - G}{2016} seconds, at constant hashrate. I don’t remember why it’s 2017 and not 2018 there (as would be expected if both the off-by-one and Erlang asymmetry are relevant), but I will try to simulate it.
- I have no opinion about when/where to enforce timestamp rules beyond bounding the amount of time the timestamp can go backwards between the last block of a period and the first block of the next period.