Established decks and Matchup Percentages

[apple713]

Im not sure if this already exists but I would like to start a compilation of the established decks and their preboard win percentages vs other established decks (eventually to include all decks). I realize that this is hard considering the variety of deck lists and variations however if there are enough people agreeing with the percentages then I'll post them. As people add to this post I'll update our list. One last thing I would like to add to the compilation is top 3-5 cards a deck is typically weak against. If you agree with something someone says quote them and +1 it so I can easily tell what recommendations people are agreeing with or disagreeing with.

Let me know if I should combine decks like blade control and miracle control or rock and Jund

Example (completely fabricated numbers)

High Tide vs.

Sneak Attack - 70% <------indicates that High Tide will beat Sneak attack 70% of the preboard games
GW/x Maverick - 90%
Miracle Control - 25%
Blade Control - 35%
Jund - 60%
The Rock - 30%
Canadian Thresh - 65%
Goblins - 10%
ANT - 50%

1. Chalice of the Void <----hardest card for High tide to deal with
2. Rule of Law
3. In the Eye of Chaos




Sneak Attack


GW/x Maverick


Miracle Control


Blade Control


Jund


The Rock


Canadian Thresh (RUG Delver)


Goblins


ANT (Ad Nauseum Tendrils)




Start Posting or making suggestions on wether this is a good idea or waste of time.

[emidln]

Moon Palace (http://www.mtgthesource.com/forums/showthread.php?4982-DECK-Moon-Palace) vs the field:

Sneak Attack - 87%


GW/x Maverick - 87%


Miracle Control - 87%


Blade Control - 87%


Jund - 87%


The Rock - 87%


Canadian Thresh (RUG Delver) - 87%


Goblins - 87%


ANT (Ad Nauseum Tendrils) - 87%


This thread makes about as much sense as my matchup estimations...

[Shawon]

Isn't this what the Too Much Information column on SCG is for?

[RJM]

Isn't this what the Too Much Information column on SCG is for?

Pretty much this.

[apple713]

Isn't this what the Too Much Information column on SCG is for?

didnt know this existed. This is a lot of good information however, i figured this compilation would be more along the lines of a reference for metagame decisions. So, if you anticipate at your local tournament that 30% of the decks will be RUG delver, and goblins has a 80% win percent vs rug delver (hypothetical), that is the clear choice for the deck you should play at your upcoming tournament. Also the weak cards part would be highly useful when deciding on boarding decisions.

[ThomasDowd]

playing whatever:

versus everything else: 100%

I always draw the nuts

Because I sold my soul to the devil.

Matchup percentages aren't real, you are playing a game where you draw 7 random cards out of 60 every time you start.

just learn how to get yourself out of sticky situations with your deck of choice, sometimes you can't do that and you lose (yeah, you will always hit your 2 outer, and they won't have it). welcome to playing card games.

even the TMI series gives a very rough estimate. The amount of data you would need to make that those statistics relevant is huge, that's not even accounting for human error, ultimately you make the choice of how you play your cards. Also, surprise, people cast cards suboptimally all the time and the amount of hidden information in this game is ridiculous.

figure out how/why you lose, try to fix it, move on.

[TheInfamousBearAssassin]

playing whatever:

versus everything else: 100%

I always draw the nuts

Because I sold my soul to the devil.

Matchup percentages aren't real, you are playing a game where you draw 7 random cards out of 60 every time you start.

just learn how to get yourself out of sticky situations with your deck of choice, sometimes you can't do that and you lose (yeah, you will always hit your 2 outer, and they won't have it). welcome to playing card games.

even the TMI series gives a very rough estimate. The amount of data you would need to make that those statistics relevant is huge, that's not even accounting for human error, ultimately you make the choice of how you play your cards. Also, surprise, people cast cards suboptimally all the time and the amount of hidden information in this game is ridiculous.

figure out how/why you lose, try to fix it, move on.

This is truly awful advice. Even a small amount of data has statistical relevance. Certainty is of course only approached, never reached, but it's foolishness to say we should ignore data because we're not 100% certain to the fourteenth significant figure.

[catmint]

People are in general bad judging their MU%.
The only way how this would make sense is to create a "hidden" questionaire were a high number of people are asked to judge on the different MU% within a short timeframe. Then you can get mean values and deviation to become a picture that might be close to reality. Even if it is not close to reality you get a picture of what people believe to be reality.

Investing that much work you could also simply do what the "too much information" column does and get the actual results.

[apple713]

People are in general bad judging their MU%.
The only way how this would make sense is to create a "hidden" questionaire were a high number of people are asked to judge on the different MU% within a short timeframe. Then you can get mean values and deviation to become a picture that might be close to reality. Even if it is not close to reality you get a picture of what people believe to be reality.

Investing that much work you could also simply do what the "too much information" column does and get the actual results.

The problem with the tmi column is that it just records data across the field in general.

I gues what we need is like tournament reports with the deck someone played and all their matchups and outcomes. Once we have a good number we can start filling in data. Lets say we have 10 people who played rug delver and they report their matchups and we notice that out of those 10 players the y'all played against maverick and 7 out of 10 matches delver beat maverick. That would be worth noting. As more players report then the stats will develop statistical significance. If thy are all hovering around 50% then there is probably no correlation to the matchup.

[Megadeus]

I think it would be too difficult to really do because like some people have already said, player skill is a big reason a deck wins or loses. And we all know of some pretty terrible players that play legacy just because its fun (myself included)

[Star|Scream]

Moon Palace (http://www.mtgthesource.com/forums/showthread.php?4982-DECK-Moon-Palace) vs the field:



UMM, that deck hasn't been updated since 2007 so I highly doubt those matchup percentages.

[Finn]

Read the whole post, McFly.

[EpicLevelCommoner]

Well, the easiest place to start is the mirror match: no matter what, someone will win, and someone will lose, or sometimes both will draw. While there may be some variety in card choices, the way the deck plays generally remains the same (exception being ramp decks, as those enable such a large variety in card choices that it enables more strategies).

So if we use the following formula: W/(W+L), a mirror match would always be 50%.

[Star|Scream]

Read the whole post, McFly.

likewise

[apple713]

Well, the easiest place to start is the mirror match: no matter what, someone will win, and someone will lose, or sometimes both will draw. While there may be some variety in card choices, the way the deck plays generally remains the same (exception being ramp decks, as those enable such a large variety in card choices that it enables more strategies).

So if we use the following formula: W/(W+L), a mirror match would always be 50%.

I wan going to omit mirrors because they should always be 50/50 like you said.

W/ (w+l) is the formula I was going to go with.

I realize there are casual players that may make mistakes however those mistakes will be presented and distributed evenly across all the records so it shouldn't matter. This is assuming that not everyone who plays goblins makes mistakes 90% of the time and that is the reason that they only win 10% of the time va the field.

Just start with some tournament reports and we will see where that takes us

[woremak]

I really don't think you are valuing the individual builds of decks enough.
Consider a miracles vs. zoo matchup, for example. Does zoo have thalia main? Are they boarding in gaddock teeg? Does miracles have access moat post board? Is it RIP/Helm or straight entreat? How many pridemages does zoo run?
Are you going to account for each color splash in goblins as it's own deck? Doesn't anything with a tutor package significantly change its MU% based on the tutors it has access to?

Obviously I'm taking a rather negative approach to this but that's because in an effort to be more accurate or useful than Too Much Information you have to define specifically what you want. I also think the playskill issue is certainly more important in some matchups and with some decks than others, and I don't know how you account for that.

I guess the TL;DR version of this is that while you can certainly create data corresponding to one build of a deck playing against a build of another deck the amount of effort necessary to accurately describe match up percentage in a way that is more meaningful than TMI.

[sdematt]

I honestly think qualitative descriptors for matchups are better then "statistical" matchup percentages, since it really depends on pilot, playstyle, builds, etc.

"Even," "I'm slightly favoured probably," and "I'm probably disfavoured" are much better descriptions of what's happening in this silly game of variance.

-Matt

[EpicLevelCommoner]

I guess another way to go about it is to confirm the data via theory, or, in other words, to make sure they make sense: "How many outs does X have to Y?" "How fast can X beat Y?" "What does X side in against Y?" "How difficult is X to play against Y?" "How inconsistent are both X and Y?"

The last two, inconsistency and difficulty, may not be the most important, but can account for random punts.

[Fossil4182]

didnt know this existed. This is a lot of good information however, i figured this compilation would be more along the lines of a reference for metagame decisions. So, if you anticipate at your local tournament that 30% of the decks will be RUG delver, and goblins has a 80% win percent vs rug delver (hypothetical), that is the clear choice for the deck you should play at your upcoming tournament. Also the weak cards part would be highly useful when deciding on boarding decisions.

What information is lacking from the TMI articles that would prevent the creation of such a table? If the concern is that there aren't enough decks analyzed, I would contact Jesse and Alix Hatfield. They have tons of data and only choose to analyze certain decks and certain match ups. However, there is enough data out there to create the type of table you're imagining.



I honestly think qualitative descriptors for matchups are better then "statistical" matchup percentages, since it really depends on pilot, playstyle, builds, etc.

"Even," "I'm slightly favoured probably," and "I'm probably disfavoured" are much better descriptions of what's happening in this silly game of variance.

-Matt

I think a combination of both is most helpful. If I where to consult a metagame chart to decide what deck to play at an event, I use relative match up %s and and decks a % of the field to make a general decision about a deck or two, then use qualitative data to tune the deck. (However, since I've gone back to being a combo hack, it doesn't matter what my opponent is playing).

However, I'll take what you're saying a step further - The data apple 713 is attempting to amass cannot offer insight into things such as: how should I approach this match-up (play style), or what is my relative skill level with the deck. Even though it may offer some insight into what build someone should construct, that is still heavily influenced by the other two factors so the data isn't necessarily a reliable to provide a predictor of success.

[lochlan]

I agree with Matt. I cringe whenever I see match-up percentages--they're totally meaningless. What's really going on when somebody says "deck X has a 70% match-up against deck Y" is they're attaching an arbitrary value to their assessment that a match-up is favorable. The problem is that it tricks people without critical thinking skills into believing that those numbers actually mean something beyond the qualitative descriptor they're implying. (They don't.)

The only way you could ever have such a number actually mean something is if you took two expert pilots of two given decks and recorded results based on some reasonably large sample size. (Note that it's important for both pilots to actually be experts, as misplays would ruin the sample--assuming we're trying to figure out which deck is actually "better.") You would also need an even division of being on the play or on the draw. And of course you'd have to have separate samples for pre-board and post-board. It would probably also be useful to factor mulligans in somehow. But even then, after this autistic mess of data, you'd only really be testing a given 75 against a given 75, as sometimes a couple sideboard cards can make a big difference in increasing win percentages post-board.

In other words: this thread is not a useful project. If the OP wants to go ahead and try anyway, more power to him, but it should (hopefully) be pretty obvious pretty quickly that nothing of value can be achieved here in terms of his stated goals.

[Admiral_Arzar]

Figured I'd write up actual percentages for your example for the hell of it.

High Tide vs.

Sneak Attack - 40%
GW/x Maverick - 80%
Miracle Control - 35%
Blade Control - 55%
Jund - 60%
The Rock - 60%
Canadian Thresh - 55%
Goblins - 80%
ANT - 60%

1. Counterbalance <----hardest card for High tide to deal with

[EpicLevelCommoner]

I agree with Matt. I cringe whenever I see match-up percentages--they're totally meaningless. What's really going on when somebody says "deck X has a 70% match-up against deck Y" is they're attaching an arbitrary value to their assessment that a match-up is favorable. The problem is that it tricks people without critical thinking skills into believing that those numbers actually mean something beyond the qualitative descriptor they're implying. (They don't.)

The only way you could ever have such a number actually mean something is if you took two expert pilots of two given decks and recorded results based on some reasonably large sample size. (Note that it's important for both pilots to actually be experts, as misplays would ruin the sample--assuming we're trying to figure out which deck is actually "better.") You would also need an even division of being on the play or on the draw. And of course you'd have to have separate samples for pre-board and post-board. It would probably also be useful to factor mulligans in somehow. But even then, after this autistic mess of data, you'd only really be testing a given 75 against a given 75, as sometimes a couple sideboard cards can make a big difference in increasing win percentages post-board.

In other words: this thread is not a useful project. If the OP wants to go ahead and try anyway, more power to him, but it should (hopefully) be pretty obvious pretty quickly that nothing of value can be achieved here in terms of his stated goals.

Feel good about that? Using autistic to describe such a "mess"? That's great: it's not like there is any possibility of someone with Autism or Asperger's trying to get into competitive Legacy . . . oh wait.

Anyhow while I do believe that taking into account all the variables would indeed create a clusterfuck of data, I believe straight-up "unfavorable", "slightly favorable", "even post-board", etc. descriptors are even more useless by themselves. That's why I suggested a confirmation of match-up percentages for consistency, difficulty, G1 and G2 outs and clocks, etc. Even though the 87% meme in the Moon Palace MU numbers is, well, nothing but a meme, the actual descriptors are quite useful in confirming (or in this instance, denying) the MU%.

[apple713]

I honestly think qualitative descriptors for matchups are better then "statistical" matchup percentages, since it really depends on pilot, playstyle, builds, etc.

"Even," "I'm slightly favoured probably," and "I'm probably disfavoured" are much better descriptions of what's happening in this silly game of variance.

-Matt



The thing about statistical descriptors is that they can be interpreted in qualitative ways. Lets say that qualitative descriptors are
probably disfavored, slightly disfavored, evenly matched, slightly favored, probably favored

their statistical equivalents could be 20%, 40%, 50%, 60%, and 80% respectively.

Im not looking to get exact numbers because that would be next to impossible with all the play styles and variations and player errors to take into account. I really just want to get something along the lines of this example i gave for all the decks. This would be a good starting point.





I think a combination of both is most helpful. If I where to consult a metagame chart to decide what deck to play at an event, I use relative match up %s and and decks a % of the field to make a general decision about a deck or two, then use qualitative data to tune the deck.





This is what i want to use the data for and how I would use it.


Anyone know how to contact Jesse and Alix Hatfield to see if they can help with this project?

[Parcher]

Jesse is in a full-time graduate program, with a concurrent TA job. Alix has a new girlfriend. Neither have much time for Magic right now. Though if they were fed all the available data by Jared at SCG, et al, Jesse has formulas set up to process it pretty handily. And Alix could turn that into an article just regarding data pretty easily if all the rest was done for him.

[apple713]

Jesse is in a full-time graduate program, with a concurrent TA job. Alix has a new girlfriend. Neither have much time for Magic right now. Though if they were fed all the available data by Jared at SCG, et al, Jesse has formulas set up to process it pretty handily. And Alix could turn that into an article just regarding data pretty easily if all the rest was done for him.

So what your saying is we should ask "Jared" from scg for the data? I can write a formula and sort data.... It's what I do for work. And we can write the article here.

So who is "Jared" from scg, and how do we contact him?

[apple713]

Anyone know who Jared from scg is that might have this info. Maybe how to contact him

[Tammit67]

Anyone know who Jared from scg is that might have this info. Maybe how to contact him

I assume Parcher means Jared Silva.

[apple713]

I assume Parcher means Jared Silva.

Anyone have an email for Jared silva?

[Tammit67]

Anyone have an email for Jared silva?

I think your best bet would try him through Facebook or starcity's site

[pavlaugh]

I've tested RUG and Sneak & Show quite a bit (50 games), both with other competent pilots and just playing against myself. Loser played first, normal mulligans except for a couple of free 5-card mulls, and no sideboarding. After the first 20 games, I changed the decklist for Sneak & Show to remove petals, adding land and more search/counters.

Decks were close to even with Sneak & Show slightly favored, winning 11/20 (55%) and 17/30 (57%).