Quote from Algernone25 »I have always been fervent in stating that I wanted to see proof of any rigging in shuffling before admitting it's there, that I'm not going to accept just anecdotal data and the rants of a salty YT crapposter.
This is a sufficient sample size and the data analyzed in a very well-thought out and consistent manner, and the data is presented in a very clear format that makes the point rather clear. I was wrong, there is a problem with the shuffler.
I'm convinced that there is an issue that's going on, and Wizards needs to address this rather quickly because it's a VERY damning issue for them. Until then, about the only thing you can do is arm yourself with the knowledge. For those not wanting to pour through all the technical details, here's a good tl;dr:
-- Limited Decks will constantly be mana-starved, no matter how many you play. Fill up on stuff low on the curve.
-- 22-23 lands is the sweet spot for not getting issues - Play too many more and you get flooded more often, play too many fewer and you get screwed more often compared to expected value.
-- 3 land opening hands are great. 2-land or fewer hands get starved more often, 4-land or more hands get flooded more often compared to expected value.
-- Taking any mulligan seems to put the land/spell ratio back to where it should be.
--By all accounts this appears to be a common mistake in implementing the shuffler algorithm in which the deck is randomized, but not randomized enough. This is why taking a mulligan fixes it. This is an act of incompetence moreso than malfeasance.
-- There is no data to suggest that Arena gives you more copies of specific cards more often than expected.
I am hopeful that the people in charge of this at WotC catch wind of this and are able to affect a fix of some kind. It shouldn't be that hard, from what the report claims.
The rigging part is sensationalism. The truth is that the developers messed up on the system for the first turn help in order to make best of one more livable. The problem that happens, and I've observed this myself, is that land pockets form in the deck when the assist is in place. There could be any number of reasons for this, but most likely it's not the fault of the actual randomization algorithm. The developers use the same core algorithm for both best of three and best of one, and taking a mulligan fixes the situation.
So with those observations in place, the issue is likely that they randomize the deck first and wrote the starting hand helper class and methods such that they pull cards they need from the randomly shuffled deck to form the seven card hands. Now, lets say it does this for two hypothetical hands, and once it has determined which one is better, it puts the other hand back into the deck: What operation would it proceed to go through to do so and is the client responsible or the server?
Well, for one they definitely do not want to do the randomization on the client side because that could allow someone to rig their own deck shuffle via third party tools, and if people think that throwing anti-cheat software onto it would be a viable solution: how about we don't repeat that entire thing that happened with Ubisoft? So the better choice would be to run the shuffler at the server. Now, this is a server that is getting hammered by millions of connected clients and reshuffles are expensive, so you'd want to minimize how many times you are shuffling the deck. Hence, I believe that when it puts the cards back into the deck, it simply inserts them back where it found them.
So, 60 cards, shuffle the deck, draw 14, optimize both by swapping cards between the hands and the deck, then reinsert the lesser of two hands back into the main deck using the same index positions they were drawn from. Is it any wonder you end up with land pockets when someone does that? Basically, they have to reshuffle the deck again after they reinsert the cards, but likely aren't figuring the deck is sufficiently randomized and want to save on server resources.