I wrote a script that simulates a game of a grabbix-style rush deck. It just plays the cards appropriately (with some special logic for what to cast Immo on) and counts how many turns it takes to deal out 100 damage, if the opponent does nothing. The cards I used in the simulation are unupped. I can run the simulation 10000 times in a few seconds to see precisely how fast on average the deck works, with a fairly low margin of error (accurate to about 0.05 turns).
The next step was to randomly swap out cards and test again if the resulting deck is any faster than the one before. If so, we keep the change and try again. A few hundred random card swaps + autotests later and we are left with the very fastest possible unupped speedbow.
Hover over cards for details, click for permalink
4vh 4vh 4vj 4vj 4vj 4vj 55u 590 590 590 5c1 5f2 5f9 5f9 5f9 5f9 5f9 5f9 5fa 5fa 5fa 5fc 5fc 5l9 5l9 5l9 5l9 5l9 5l9 5ur 8pm
According to my simulation, if the opponent does not respond, this deck can pull off a 5 turn win almost 50% of the time. Average TTW in the simulation is about 5.6 turns. In practice, 6 or 7 turns is typical, 5 occasional.
Note that this deck has no deflagrations. 2 deflagrations make it a lot more viable as a real deck. I usually swap out one lycanthrope and the crimson dragon for the deflags.
I used a variant of this strategy to craft my tourney deck which I used to win the eastern tournament (the first tournament I entered!) on Sunday. Basically just did the random swapping, with the condition that the deck must always have 6 Crimson Dragons.