There is also afaics a math flaw in ArticMine's analysis. Unless N is very small, then a miner with a significant but less than 51% hashrate is going to win a block in most every N set, and thus they can hit the 2 * MN hard limit every time, gradually ramping the median block size up over time. Thus the spam attack is not avoided, rather it just takes longer. And again I had pointed out that by shorting the coin, they can potentially recover their lost block rewards and profit. And if N is very small, then the likelihood that a miner can win all N blocks with less than 51% hashrate increases. Also it is not clear to me from ArticMine's specification if N is overlapping meaning a FIFO queue? But I doubt that makes any difference to my math point.
Bumping up against the hard limit is probably wastefully expensive for this "attack"; you only need to produce blocks bigger than the (expected, since you don't know in advance) median to cause the new median to shift higher.
However, your attack doesn't work AFAICS. If a (less than 50%) miner spams his own blocks full of transactions (greater than current median, exact values not important as he's not including any other transactions), he can only expand the median relative to his % of the hash rate. For example, if a 33% miner produces blocks all 2x the median, he effectively reduces "real" block space by 33%, so the median should expand 50% to accommodate. His effect after that expansion is likely not going to be significant. This should remain true all the way up to 50% and 2x. After that, a block from the attacker actually captures the median value, and he can do whatever he wants.
N=100 right now BTW.
There are some other possible weaknesses or mal-incentives in the current approach; I think it still has room for improvement.
