Biased AI cascades, and matchmaking

Volrak

Matchmaking is clearly skewed. Maybe it's an experiment to try to address the 30-way ties at the top of the ladder (if so, apparently a failed one).

When it comes to gem cascades, a perception of skew is exactly what I'd expect if there was no skew, because there are many clear psychological factors at play. For example:

• Players almost always win vs the AI. So a big AI cascade resulting in loss is a far more "shocking" and perceptually notable event than a big player cascade, which usually results in win-more or win-anyway.
• Chained player extra turns feel like a series of normal turns; the player evaluates the state in between each move; evaluating the impact of entire chain actually takes effort. Chained AI extra turns, on the other hand, coalesce into one massive cascade event, and players have no choice but to evaluate the event as a whole chain.
• The AI often plays clearly suboptimal moves, and it's natural to question why. When the result is not a cascade, it's a psychological non-event. But when the result is a cascade, as it surely will be sometimes if it's random, then an easy psychological response is to treat the sequence of events as evidence of causality.

There are probably more.

span_argoman

killwind wrote:

Hello guys,

I have also been trying to sort how I get paired up. I see Mersicide, SellerSloan, Quebecers, Largatha, Majincob, Shteev and BaDJer, regularly.

My coalition members are constantly complaining about running into me in events.

So lets try and baseline this. Let me give you my stats. I am an avid Quick Battle player, I constantly either outright win the QB or I place top 10 if im not feeling the grind.

On the quick battle side of the house I have 7400+ wins 940+ lost. Win percentage is 88.7%. I am Platinum tiered in all colors as of 2 weeks ago. I have mastered 333 of 679 cards.

All this data is so that we can find a common denominator as to how I am constantly being paired up with you and my coalition members.

I matched up against you in the recent NoP event quite a fair bit. I see those names fairly often too. Though for me it's more Mersicide, BaDJer, Quebecers and a few other names. I get Majincob once in a while but I haven't seen Shteev in a while. I'm guessing there might be other criteria to the match pool that overly restricts it and makes the match-ups less than random.

Stats-wise,
Platinum in 4 colours (Gold in White), exceeding the Platinum baseline by an average of ~100 for the 4 colours, 221/250 for White
Mastered 393/730 cards
QB: 4,651W / 261L / 94.7%

The response from customer support to Fiddler's issue of no matches found for events was that his mastery was way too high above any other player's (he already had 300+ above Platinum as of the first few events). I guess they tweaked it to enable him to be able to find matches to play in the events, but there is definitely some consideration on the player mastery level.

Our QB records are rather different with you playing a lot more so I'm inclined to think that QB and the Events do not intersect in any way. I guess they might need to tweak their algorithms on matching by mastery level further though.

Morphis

EDHdad wrote:

As for constantly getting paired with the same person, my belief is that you're probably getting paired with people who are on a similar level as you with respect to the event. So if you're grinding the event 1 minute after it starts and go on a 35 game win streak, you're probably getting paired with some other person who has also been grinding since the first minute and who is also on a 35 game win streak. Or some other factor related to the way your are playing the game (for example, a top 10 player might only be paired with other top 10 players). If so, then a person in the #1 coalition is probably destined to frequently be facing the same opponent. I doubt that you have an equal chance of facing the #500 player in an event that you have of facing the #5 player.

I doubt this to be the case.
I started this nope event really bad(2 loss and a third not too far) and still I faced killwind all the time.

Also when I faced not killwind my opponents were usually not that strong and were not in the top spots.

Ohboy

span_argoman wrote:

killwind wrote:

Hello guys,

I have also been trying to sort how I get paired up. I see Mersicide, SellerSloan, Quebecers, Largatha, Majincob, Shteev and BaDJer, regularly.

My coalition members are constantly complaining about running into me in events.

So lets try and baseline this. Let me give you my stats. I am an avid Quick Battle player, I constantly either outright win the QB or I place top 10 if im not feeling the grind.

On the quick battle side of the house I have 7400+ wins 940+ lost. Win percentage is 88.7%. I am Platinum tiered in all colors as of 2 weeks ago. I have mastered 333 of 679 cards.

All this data is so that we can find a common denominator as to how I am constantly being paired up with you and my coalition members.

I matched up against you in the recent NoP event quite a fair bit. I see those names fairly often too. Though for me it's more Mersicide, BaDJer, Quebecers and a few other names. I get Majincob once in a while but I haven't seen Shteev in a while. I'm guessing there might be other criteria to the match pool that overly restricts it and makes the match-ups less than random.

Stats-wise,
Platinum in 4 colours (Gold in White), exceeding the Platinum baseline by an average of ~100 for the 4 colours, 221/250 for White
Mastered 393/730 cards
QB: 4,651W / 261L / 94.7%

The response from customer support to Fiddler's issue of no matches found for events was that his mastery was way too high above any other player's (he already had 300+ above Platinum as of the first few events). I guess they tweaked it to enable him to be able to find matches to play in the events, but there is definitely some consideration on the player mastery level.

Our QB records are rather different with you playing a lot more so I'm inclined to think that QB and the Events do not intersect in any way. I guess they might need to tweak their algorithms on matching by mastery level further though.

Wait mastery makes sense. I mastered quite a few more cards between the last event and this one. Last event I kept getting matched with warcin. Didn't see him at all this one. He for replaced with killwind.

Morphis

Mastery probably should not be the case if by that we mean total mastery points.
Many people on platinum facing killwind have many(or all)colors mastered.
I have only "two"(one actually but black is missing like 5 mastery points).

If instead the matchmaking is based on amount of rares/mythics mastered it could be much more likely since I got to platinum using only those.

To be honest if I would have to guess a reason it's either bugged/not well thought algorithm or is simply based on collection.

Overall I am more inclined to the former.

shteev

bken1234 wrote:

The Killwind thing is frustrating because people play him in both Platnium and Gold.

In the same event?? Or in different tiers in different events?


My mastery level currently stands at 483/727.... 181/255/129/195/165 (all plat). I did just do a bout of mastery before the last event, and met up with Killwind for the first time.

killwind

During the end of the event some of Extinction members were telling me that they were playing against my non-event decks. So this makes me wonder what criteria are they using to create the pairings. What database are they polling for this to occur, is it using an active player database or does it draw from all players available.

I dont know if the Dev's even look at these boards.

Round Robin:
if they are using a round robin type algorithm to create pairings there is a static start point in the round robin algorithm. So whoever sits in the first slot of the round robin will always be checked against whomever comes in for a pairing. it checks against the first spot. If the guy sitting in the first spot is in a game, then it randomly creates a pairing. But what happens is that it constantly does a check against the guy sitting in first spot and checks if he's available. If the dude in the first slot is available he gets paired with the next available opponent as it finishes the check. It will then do this to the guy sitting in the 2nd slot...

This is something I see in the line of work that I do.

BLACK LOTUS

There are going to be arguments that support both sides of this and I've complained about the cascades and horrible support placement for the player and almost perfect placement of AI support placement. The problems extend beyond these problems ... i.e. the activate ability... Whether I'm playing a card with the activate ability or using a plainswalkers' ability the activated gems are placed so that you can't match the activated gem but when the AI casts a card with activate it can make the necessary match to get the activated gem.

One of the other problems I've noticed especially with Koth.. his reshape the land ability. When I use this ability yea I do get to make a match after the gems are destroyed and sometimes I even get to make a red match, however when the AI uses the same ability it sets off a chain of matches and still gets to make a match after the chain has finished. But this is the same for other cards as well like supports or spells that change a gems color. The AI will get gems to change color near other gems of the same color and make a match instantaneously or can make a match on its next move. When it changes a gem color for me it's off in a corner with no other gems of the same color even close to it.

Yes I'm going to get ppl like PQMTG to say I'm "whining" or say how often do I get an awesome cascade or perfect support placement... Yeah you're right I do get the occasional awesome cascade or perfect support placement.... But not near as often as the AI.

I have been keeping track of one stat and that is the starting red gems when I use Koth (5) and the starting red gems when my opponent is Koth (8).

As far as matchmaking I'm facing the same opponents ... Quebecers, Mercicide, Killwind. Etc... And it always seems I'm facing 1 of 4 Plainswalkers .... Kiora, Koth, Obnixilis, or Garruck.

Corn_Noodles

killwind wrote:

During the end of the event some of Extinction members were telling me that they were playing against my non-event decks.

This changed a patch or two ago and it has never been acknowledged if it's a bug or a correction. It used to be that planeswalkers that did not share a color with a node in an event would not show up as opponents.

blacklotus

fyi: I just met Killwind mono-red Koth a an opponent in this latest white/blue/green terror event.

andrewvanmarle

I encountered somefhing similar in the past few events: drawung four of each card in sequence : i play behold the beyond our forgotten creation and i would draw 4 of card A and two of card B and the next two turns two of card B again, my starting hand would look like this too.

It didn't happen every match, but i did have quite a few matches where i died because i only drew "lands".

Buret0

Tin foil hat party in GD!

Fiddler

I assume the event matchmaking is based on mastery level. I see kill wind a lot in my event matches. I have high mastery, over 400 in three colours, and close to it in the other two. Is killwind's mastery similar?

And different events use different algorithms. I assume this because I was only excluded from one type of event because of my mastery level. but that has been fixed.

Nitymp

Surely matchmaking *should* be based on who is in your bracket though. I mean, having sub-brackets within brackets is stupid.

doughnut_hole

I mean....yeah. The AI can be pretty biased at times.

For multiple games, both on the QB and the Terrors in the Shadows event, when the AI gets all the cascades, it can be pretty frustrating. I've noticed that for when the game wishes to break my winning streak after a certain amount of wins (say, 30+), it'll automatically go into "hard" mode where every match is set up so that the board is heavily colored in its favor. Each move is at least three cascades long for the AI, where my moves are just one-single moves (and never in my color). The algorithm for these events seem to set up each turn for the player so that the AI can capitalize on its next move: I move one blue gem because it's the only move I have, and this in turn allows the AI to move the white gem, causing an entire cascade that takes 15 seconds long, unleashing spells, creatures, and supports. After losing these type of rounds, the next game is set up in my favor, where I suddenly make five to eight cascades in one move. I guess it's meant to "balance" out the algorithm.

Moreover, I've noticed that the AI has a lot more duplicate cards these days. Like, when playing a lot of the AIs, if it casts a Scour from Existence, it'll most likely have at least three other Scour from Existence cards in hand. This can get really annoying. I played one game where the AI cast THREE Goldnight Castigators in one game: it cast two in one chain-cascade move, and then produced another just two rounds later. Before that, I played an AI that cast four Undergrowth Champions in one game. And so forth. (3 Crush of Tentacles; 3 Angel of Deliverance; *4* Ulvenwald Hydra; 3 Descend on the Sinful).

Not to say that this is bad or anything, and that the developers should make drastic change; I'm just replying to this thread stating what I've observed. I also don't mean for anyone to take what I've experienced as a generalized phenomenon to add on to more, legitimate complains. Just adding to the "Biased AI cascades" title. Had I a solution, I would state it here. Unfortunately, I don't have any.

Anyway, I'm still a dedicated player and will happily be waiting for the next new set of cards. I would like it if the algorithm wasn't so skewed at times, but...meh. I'm not a game developer, and I don't know the language of gaming/code, so I'll settle for having fun still (besides--if we won all the time, this game would be pretty boring).

<---An "A-okay" face for the developers! You guys never get one.

Morphis

A thing to be considered is that it is even worse in terror than in NoP.
In NoP it's not good in itself cause I'd like more not having such variance on both ends(ai and player).
But in NoP when it happens for the player he can rush and win so at least is advantageous.

In terror you have to wait so unless it happens when you have all objective met(counting what you will cast out of the cascade too of course) it's not that much useful.

Steeme

I'm glad I'm not the only one noticing the shift in AI behaviour over the past month or so.

Before I start, let me declare that these are just my observations, and it should in no way be construed that I am whining about the difficulty of the game.

They have improved the algorithm and in doing so it has ramped up the AI's difficulty level.

This is fine, as we all enjoy a bit more of a challenge, but when you're facing someone like Koth or some other highly tuned grief deck, it gets flat out over-the-top out-of-control.

I only notice these things when they are glaringly out of balance. I've noticed that on a disturbingly high percentage of matches, the AI is just able to consistently gain mana and loyalty sometimes at 2 or 3 times the rate.

This usually happens with a combination of the following:
1) It is able to consistently trigger multi-cascades at-will without any cards to help generate them (ie. color changers)
2) It is able to not only consume its primary colors in mana, thereby maximizing its gain, but at the same time deny me of my primary colors, thereby minimizing my gain (huge mana swing)
3) It is able to do the above, while directly or indirectly damaging my supports, thereby effectively having free support removal

In addition to this, I notice:
1) Before I have played any real cards, or made any real moves, the AI is on "cruise-control"
2) The minute I drop any type of support, no matter what it is, it becomes the AI's new purpose in life to do everything it can to remove the bloody thing
3) The minute I drop any type of real troop that looks like it will be a threat (either by straight up power or by triggered abilities) the AI shifts into "god mode" where it flies off the handle and just starts dumping its entire library onto the playing field multiple-times over per turn

Again, these are my observations. I notice it because I can detect a huge shift in the flow of the match. I think a lot of times (ie. in Quick Battle) it's not very important because there are quite a few weak decks out there. However, when matched up against highly tuned decks, sometimes it's just easier to swipe the application off the screen and move on.

paralistalon

I've been paying attention to cascades since reading this
topic, and my experience has been that I'm just
as likely to get an insane cascade as the AI.

Omega Red

I don't think the AI gets more or better cascades than me but I always found interesting how after a big board shake new stones seem to fall always in sets of the same colour. Like, you use Koth's first ability and then four blue stones fall straight, a prefabricated match basically; then three green, then three yellow, etc. This always irks me as it seems coded deliberately that way to make the game faster and more arcade-like.

Volrak

In the first turn of an event game yesterday, I got 3 extra turns, for a total of 45 mana and 21 loyalty.
A similar thing happened in a game the day before, with at least 24 loyalty and maybe more mana too, but I was too busy playing the game to be bothered to add it up (which I guess is the default for everybody). Obviously, in both cases I had no colour changing supports on the board, being the first turn.

I don't think I'm especially lucky. I think this happens to everyone sometimes but most people don't think twice about it. Yet they do think twice when it happens to the AI, as it surely will sometimes if it's random.

For what it's worth, individual observations can be interesting to read, but neither mine nor anyone else's actually supports any theory of how random the AI is or isn't. To do that you need unbiased data, and lots of it. Feel free to formulate a specific theory you want to test, collect the data, do the stats, and demonstrate lack of bias in the experiment - because that's the only way to get any confidence in an alternative theory. In the absence of that, the simplest theory which can explain the observations must be assumed. (And it goes without saying that random gem generation is the simplest theory which can explain any and all individual observations.)