Concept
This is a project to utilize a large data resource, specifically www.cubetutor.com draft picks, to create a statistical ranking of the best cube cards by color and CMC. This is a good data resource, because it has a 100k+ cubes and hundreds of millions of draft picks. For example, just the ~3500 cards considered in the project were picked over 123 million times on www.cubetutor.com. Because the sample size is so high, we can have increased confidence that the results reflect the actual preference of cube drafters, rather than just random luck.
This project takes inspiration from the annual power rankings last held in 2016, which I have found to be an awesome and useful reference.
The main difference is the power rankings were based completely on polling of mtg salvation regulars. In contrast, these power rankings are at least initially based completely on cubetutor ratings (later, I joined the data with gatherer information which allowed me to include gatherer user rating as a feature as well as adjust cube count for the number of reprinting of each card) in order to reduce the amount of parsing work required and increase the sample size of the data.
Voting is now CLOSED - the results have been updated to reflect the voting and the netted rankings from all the data (cubetutor, gatherer and mtgsalvation votes).
Goal
Use data to comprehensively rank the top cards overall, by color and also by CMC within each color.
Features, Feature Statistics and Pre-processing
We are currently using four features:
1) Ln(Pick %): Pick % is the percentage of times the card was selected by a drafter in a pack. The mean pick percentage is 13.6% and the card with the highest pick percentage is Black Lotus (86%), followed by Ancestral Recall (73.7%) and Sol Ring (73.5%). Anecdotally, pick percentage corresponds very strongly to cards considered to be the best. The distribution of pick percentages is extremely condensed with the vast majority of cards lying in the [10%, 15%] range while a few outliers like Black Lotus are about 16 standard deviations above the mean. Because of this, the distribution has extremely high skewness and excess kurtosis - in other words, it is very far from normally distributed. To mitigate this, I take the natural logarithm of the Pick %, which helps to condense the outlying numbers. On a natural logarithm scale, Black Lotus is only 9 standard deviation above the mean, and the skewness and kurtosis is significantly lower (distribution closer to normal). Lastly, I standardized the Ln(Pick %) to have zero mean and unit standard deviation. This standardized score is given 60% total weight on the final score.
2) Cube Count / Expected Cube Count: Cube Count is the number of cubes in which the card is included. The most included cards are Evolving Wilds (37881), Lighting Bolt (36037) and Counterspell (32355). While a valuable feature, this feature does not correspond as strongly to the "best cards", since cards like Black Lotus (9220), Ancestral Recall (7427) and Sol Ring (17078) may be included in fewer cubes due to availability, price
or style (e.g. powered or unpowered). However, this feature does have value, primarily because the Pick % is much more reliable when the number of cubes including the card is large. If you created a ranking based only on Pick %, then Pet Project would be ranked #24, with 28.72% pick percentage and a cube count of 545. Furthermore, Kindle would be #34 with a 24.6% pick percentage and 2001 cube count, which is clearly ridiculous. Lastly, I found there was an extremely strong linear relationship between #of reprints and cube count - in fact a linear regression of # of reprints versus average cube count for cards with that # of reprints has an R^2 of 94% (see attached picture)! In contrast, # of reprints has zero correlation to Pick % or Gatherer User Rating, so there's little reason to believe that # of reprints is a feature indicative of cube card quality. This suggests that by controlling for the influence of reprints (availability and price proxy) on Cube Count, we can improve the relevance of the Cube Count feature. You can see the relationship in the attachments Lastly, I standardized the (Cube Count / Expected Cube Count) feature to have zero mean and unit standard deviation. (Cube Count / Expected Cube Count) was given a 15% weight.
3) Ave Cube Bonus: Cube tutor also maintains 360, 450 and 720 average cubes, broken out by unrestricted, peasant or pauper. Although less granular than the "Cube Count", this tells us the popular cards and has the additional advantage of excluding pauper and peasant cards that would not make the list in cubes where money is no issue. For example, Murder is excluded from the 360 and 450 lists even though it is in the top 300 cards by cube count. I gave cards 1 point for being in each list (so cards in none of the lists have 0 points and cards in all of the lists have 3 points). Ultimately, this feature is still 80% correlated to the "Cube Count" feature and also has some quirks (like Library of Alexandra doesn't even make the 720 list). I standardized this feature to have zero mean and unit standard deviation and gave it only a 10% weight.
4) Ave Gatherer Rating: This is the average gatherer user rating across all versions of the card. For versions of the
card that had no user rating, the rating was defaulted to 4.0 (which is is average user rating). This feature turned out
to be 31% correlated to Ln(Pick %) and clearly tends to be higher for good cards than bad cards. While not a particularly
strong feature, feature diversity in general can help improve our ensemble and eliminate some random variation. Once
again, this feature was standardized to have zero mean and unit standard deviation, and given a 15% weight on the final
score.
For more information on feature correlation and statistics, see the attachments.
Weaknesses / Challenges
A numeric ranking is only as good as the features available from the data.
In this case, the features are decent but there will always be outliers and the ranking will never be perfect.
For example, I think Channel is ranked much too low (#87) but this is because of its relatively meager pick % of 18.66%. Part of this could be due to more generic cards / universally playable cards (like Karn or Wurmcoil Engine) having much higher pick percentages than more powerful, but narrower cards like Channel.
Another big challenge is that unlike, say, a Linear Regression that aims to explain the sale price of houses, there is no actual numeric target label that we are trying to explain. Instead, we are trying to match something inherently subjective and undefined - how "good" magic cards are in cube setting. The best we can do then, is try to find decent proxies of goodness in cube. Currently, the strongest proxy seems to be pick % in cube. If we believe there's a good degree of correlation between pick % and the strength of cube cards, the best way to identify potentially valuable features is to find other, somewhat distinct features that still have a significant degree of correlation to the strength of cards in cube.
Cards Considered
I selected the top 2000 cards based on both Pick % and Cube Count. Because there is some overlap, this works out to 3,444 cards.
Furthermore, I filtered out cards with a cube count of less than 2300, because this leads to a lot of false positive cards with small sample size, like Kindle. This left me with 2111 cards total.
For cards that missed the "720 average cube" list entirely, I allowed myself to filter out the following cards that were in the top 1000 and which seem to be complete outliers. For cards that make the 720 average cube list, in contrast, there is no outlier filtering allowed:
Furthermore, there were a few cube worthy cards outside of the CubeTutor 720 Average Cube List that are simply criminally underplayed, which I added back in for consideration:
Cube Tutor Top 300:
These are the top 300 cube cards according to this methodology.
I've attached images showing how the rankings broke out for the top 50 cards.
I created an editable spreadsheet so that anyone can vote on any of the over 2,000 cube cards in the rankings.
To vote, type your username in the top row of a new, unused column and save your scores for the cards you care about in a 0-10 numerical range. You can vote on as many or as few cards as you wish. Once voting closes, the average user rating will be used to adjust the numerical ratings to get combined ratings. This way we will have three difference rating systems: 1) cubetutor/gatherer numeric data only, 2) user ratings based, 3) combined scores (combining #1 and #2, i.e. using the user ratings to adjust the cubetutor/gatherer numeric ratings)
VOTING RATINGS GUIDE: 10: Unequivocal P1P1 over everything else in the game. These cards are broken and make other great cards look silly. For me, the only 10s were Black Lotus, Ancestral Recall and Sol Ring) 9: Power Nine Level Cards - these are in the top 20 most powerful cards in powered cube. For me, a few examples were Moxen and Mana Drain. 8: A cut below the power 9 in level, these cards tend to be the strongest in their color and often still make P1P1. Some examples for me were Vampiric Tutor, Lightning Bolt and Wurmcoil Engine. 7: The are solid role performers that typically make even the smallest powered cube lists (360). You are usually never unhappy to play these cards. Some examples for me were Blade Splicer, Arbor Elf, Shriekmaw, Flametounge Kavu and Treasure Cruise 6: These are cards you would play at 450 but not at 360. 5: These are cards you would play at 540 but not at 450. 4: These are cards you would play at 720 but not at 540. 3 or below: You would not play these cards even at 720.
How will voting change the results?:
Voting will close 3/31/2019 assuming there are enough votes. Otherwise, voting may be extended another quarter.
Once voting closes, the average scores across all the cards that received votes will be standardized to have zero mean and unit standard deviation, and then we add in the average cubetutor z-score for the set of cards that received votes. The reason we add this mean is that people will tend to vote mostly on the best cards which already have positive cubetutor z-scores and we don't want to bring down the mean of these good cards due to selection bias.
To determine the "combined scores", we treat the numeric rating based on cubetutor/gatherer data as a prior and update it based on the number of votes received - cards that received many votes will see their scores adjusted significantly while cards that receive few votes will not have their scores adjusted much.
Specifically, because the statistical significance of the average user rating scales roughly with the square root of the number of voters, each card's combined score will equal:
where n is the number of votes, z_cubetutor is the numeric cube tutor score, z_ave_user_rating is the z-scored average user rating, and z_cubetutor_mean_score_of_cards_receiving_votes is the average cubetutor z-score for the set of cards that received votes.
For example, if 16 users vote on that specific card, the combined score will come 50% from the cubetutor ranking z-score and 50% from the user ratings (which have also been standardized into a z-score). If zero users vote, the user ratings count for 0%, if 4 vote it counts for 42% and if 20 vote the user ratings count for 75% of the total score.
Definitely interesting to see how it differs from the opinions here : )
Some things I consider weird (that can probably be explained due popularity):
- Doom Blade but no Go For The Throat, Damnation, or Toxic Deluge;
- Emrakul, the Aeons Torn is in the list : )
- Splinter Twin is there but Kiki-Jikki is not;
- There's a signet(!?!) in the list;
Compulsive Research I don't think is too crazy as a top 100 card - that gets picked highly by cube drafters like LSV, Reid Duke and NumotTheNummy (I watch a lot of cube draft).
Signets (especially the blue ones) also place highly, and deservedly so in my opinion. Dimir Signet is #94, Izzet Signet #106, Azorius Signet #130, Simic Signet #158. Non-blue Signets are a little lower, but still make the top 360. Again, cube drafters like LSV, Reid Duke and NumotTheNummy take the signets very highly, and the fact that all the signets place highly suggests that Dimir Signet at #94 is hardly a fluke.
Emrakul, the Aeons Torn being in the list at #76 is not surprising. This is a bomb card in the MTGO Vintage Cube which supports re-animator strongly with cards like Entomb, Buried Alive, Shallow Grave, Corpse Dance. Also, it's pretty much the best creature for Sneak Attack and Through the Breach. Once again, top cube drafters take this very highly. For example, LSV calls Emrakul a very high pick in this article: https://www.channelfireball.com/articles/the-ultimate-guide-to-cube-archetypes-blue/
For the cards mentioned not in the top 100, here are the current placements:
1) Go For the Throat is #153
2) Toxic Deluge is # 343
3) Hero's Downfall is #251
4) Damnation is #181
5) Kiki-Jiki is #111
I absolutely agree that all these black cards are WAY above Murder.
Murder in top 100 is a crime, while Toxic Deluge might actually deserve a spot in the top 100.
This is not an issue with the algorithm, but a data quality issue.
Fundamentally, Murder just has a higher pick rate at a large sample size.
The issue, of course, is that Murder was a high pick in block cubes or something or the sort, so it is picked highly over much weaker competition.
Unfortunately, there's no way to correct for this right now, so there are 3 ways to improve the quality of results:
1) Get some more data. If there's another website with data like cubetutor.com, combining the data or creating an ensemble of results could eliminate a lot of quirks like this.
2) "Obvious errors" like Murder can be manually removed (can have a list at top mentioning the cards removed)
3) Use the cubetutor results as a "prior" (e.g. pick the top 40 cards of each color) and then do community runoff polls to determine final placement.
I'm not arguing that Downfall should be in the top 100; only that it makes no sense for it to be missing when there's a strictly worse card on there. Maybe C/Uubes are having a big impact on the rankings? IDK.
This is also why pacifism is above Venser, Shaper Savant for example.
The different card quality in block, core or pauper drafts is the issue, and there's currently no way to separate out those cube picks.
This is currently the biggest issue with the rankings - I posted three potential solutions earlier with #1 being my favorite if anyone else knows another data source
Yes, I'm planning on doing polls but probably one a day to space out the work.
First I have to join this data with card info data so I can sort it by color and cmc.
Then, I'd like to do polls by color and combine those results with the numeric scores for a final result.
However, I haven't found a good polling solution yet - ideally it would allow bulk uploading of each card name as a separate "question" that voters could give a numeric score (e.g. 9.5 for black lotus, 1.0 for murder)
So I found at least one way to improve the rankings.
Although the "top cards" section of cubetutor is based on picks from all cube environments, cubetutor also has "average cubes" of size 360, 450 and 720 broken out by price unrestricted, pauper and peasant.
Although Murder is in the Top 360 cards according to "Top Cards" by both Pick % and Number of Cubes, it is excluded in both the 360 and 450 price unrestricted Average Cubes, appearing in only the 720 price unrestricted average cubes. Clearly these average cubes are using more granular data that partitions cubes by type. Although this raw data is not available, I think the presence of cards in these Average Cubes list could be another feature (with the most points for being in the 360 list, less for 540 and the least for 720).
Emrakul, the Aeons Torn being in the list at #76 is not surprising. This is a bomb card in the MTGO Vintage Cube which supports re-animator strongly with cards like Entomb, Buried Alive, Shallow Grave, Corpse Dance. Also, it's pretty much the best creature for Sneak Attack and Through the Breach. Once again, top cube drafters take this very highly. For example, LSV calls Emrakul a very high pick in this article: https://www.channelfireball.com/articles/the-ultimate-guide-to-cube-archetypes-blue/
No, they meant Emrakul, the Aeons Torn, since the reanimation effects they mentioned are Shallow Grave and Corpse Dance; two of the reanimation effects that CAN actually hit the OG Eldrazi titans.
Hey IMorphling89, this is really fun data to go through. Could you expand each section to include the top 75-100 perhaps? So that we can get a more complete overview of the hierarchies in each color. You yourself mentioned Armageddon missing from the top 30 white list (just), and where are things like Fractured Identity? lol
Fractured Identity, shockingly, fails to meet the minimum requirement of being in 2300 cubes (2267) which is mentioned in the original post. Actually, it's pick % (11%) is pretty low too - it doesn't even rank in the top 2000 cards by either cube count or pick percentage. As a result it's excluded from other lists like the "720 average cube", for example.
Even if it did make the list based on pick percentage, it would still be filtered out, along with a lot of false positives like Kindle that are properly excluded, because it is included in <2300 cubes.
To get cards like Fractured Identity that slipped through the cracks, I'd need to expand the analysis to even more than the 3500 cards initially examined. I do have to say that I'm shocked this great card is so underplayed; hopefully it will see more play in a year from now.
I will most likely just post a link to the full spreadsheet at some point which would show you where every card is ranked.
It's actually fairly easy to see why Fractured Identity is underplayed - it's a new card from a commander set. It's not even in the MTGO cube it flies that under the radar. Most people don't fanatically curate cubes like we do here, and even so you really need to play it to understand how good it is. Not at all surprised to see it underplayed.
For what it's worth, I think you'd probably just need to go through by hand and throw out cards you think are there because of peasant/pauper cubes (pacifism, murder, etc) - I think you mentioned that you might need to do this.
My Cubes - The Busted Cube. A fully functional, almost 100% custom cube. The project started out by asking "What if other colors got cards on the power level of Mana Drain,Ancestral Recall, and Time Walk?" Draft and enjoy!
Yeah I pretty much already did this at this point. There is a reasonably long list of excludes in the original post.
FWIW I currently control for # of reprints in the cube count since this is a strong relationship.
If there was a source that listed the print count by common, uncommon, rare and mythic I could pretty easily calculate the total circulation of every card and control for that, which should boost the ranking of low circulation cards like Fractured Identity a lot.
Hi guys, just wanted to say that as discussed earlier, voting is now open!
This is your chance to vote on which cards you think are best and influence the rankings!
Link to the voting spreadsheet below, as well as how I plan to use the user ratings to influence the rankings.
VOTING (OPEN):
I created an editable spreadsheet so that anyone can vote on any of the over 2,000 cube cards in the rankings.
To vote, type your username in the top row of a column and save your scores for the cards you care about in a 0-10 numerical range. You can vote on as many or as few cards as you wish. Once voting closes, the average user rating will be used to adjust the numerical ratings to get combined ratings. This way we will have three difference rating systems: 1) cubetutor/gatherer numeric data only, 2) user ratings based, 3) combined scores (combining #1 and #2, i.e. using the user ratings to adjust the cubetutor/gatherer numeric ratings)
How will voting change the results?:
Voting will close 3/31/2019 assuming there are enough votes. Otherwise, voting may be extended another quarter.
Once voting closes, the average scores across all the cards that received votes will be standardized to have zero mean and unit standard deviation. To determine the "combined scores", we treat the numeric rating based on cubetutor/gatherer data as a prior and update it based on the number of votes received - cards that received many votes will see their scores adjusted significantly while cards that receive few votes will not have their scores adjusted much.
Specifically, because the statistical significance of the average user rating scales roughly with the square root of the number of voters, each card's combined score will equal:
where n is the number of votes, z_cubetutor is the numeric cube tutor score and z_ave_user_rating is the z-scored average user rating.
For example, if 16 users vote on that specific card, the combined score will come 50% from the cubetutor ranking z-score and 50% from the user ratings (which have also been standardized into a z-score).
How do we determine what an 8 is vs a 10? What metric are other voters using? Is it completely arbitrary and subjective? Right now, there's a 6.25 next to Serum Visions, and nothing next to Dig Through Time. What does that mean? Are we supposed to put 10s next to every card we like and nothing next to ones we don't like? Not sure how to vote in this system, and how it generates meaningful data...
Hi, great questions. Let me try to answer them as best as possible one at a time. If you have some thoughts on how to improve the rankings or the voting I'm certainly interested :).
How do we determine what an 8 is vs a 10? What metric are other voters using? Is it completely arbitrary and subjective?
I choose voting from 0 to 10 as this is usually intuitive for voters. However, if this is too arbitrary I could try to add a ratings guideline. Ultimately everyone has their own subjective opinions of what is best and the goal is to let as many people vote on whichever cards they want to, and ignore cards they don't care about. Voting is definitely a more subjective process than the pure numbers based ranking, but I think the community can provide meaningful ratings and contribute to improving the rankings. Plus, it's fun to poll people and weigh in on rankings like these :).
I can certainly try to provide a metric, which I'll include in the original post.
Ratings: 10: Unequivocal P1P1 over everything else in the game. These cards are broken and make other great cards look silly. For me, the only 10s were Black Lotus, Ancestral Recall and Sol Ring) 9: Power Nine Level Cards - these are in the top 20 most powerful cards in powered cube. For me, a few examples were Moxen and Mana Drain. 8: A cut below the power 9 in level, these cards tend to be the strongest in their color and often still make P1P1. Some examples for me were Vampiric Tutor, Lightning Bolt and Wurmcoil Engine. 7: The are solid role performers that typically make even the smallest powered cube lists (360). You are usually never unhappy to play these cards. Some examples for me were Blade Splicer, Arbor Elf, Shriekmaw, Flametounge Kavu and Treasure Cruise 6: These are cards you would play at 450 but not at 360. 5: These are cards you would play at 540 but not at 450. 4: These are cards you would play at 720 but not at 540. 3 or below: You would not play these cards even at 720.
Right now, there's a 6.25 next to Serum Visions, and nothing next to Dig Through Time. What does that mean?
The ratings you are looking at are my ratings. The column entry says IMorphling89, indicating these are my ratings. The numbers you are seeing are just my subjective ratings. Most human beings probably don't want to vote on every card, so I just voted on the cards I care about. If you want comprehensive ratings for every card considered (~2500) then you'll find that in the original post based on the cubetutor and gatherer data.
Each user is free to type their name in the top of a new unused column and vote on as many cards as they wish.
Are we supposed to put 10s next to every card we like and nothing next to ones we don't like? Not sure how to vote in this system, and how it generates meaningful data...
Each user should vote only on the cards they care about. Cards you don't want to vote on, just leave the voting blank.
Thus, everyone can vote on as many cards as they want. Obviously we can't require everyone to vote on every card.
I've divided the cards by color to facilitate this a bit.
Once the voting ends, there should be many votes on many cards. As described above, we can use this combine this data with the initial (comprehensive) cubetutor z-score ratings to get aggregated ratings that also take into account these user ratings.
How do we do that? For for the aggregate ratings, we start with the cubetutor ratings as a prior, and we adjust from the prior based on the number of votes. Cards that get no votes won't have their rating adjusted at all. Cards that receive many votes may have their scores adjusted substantially if the community rating differs significantly from the cubetutor rating.
Specifically:
1) Cards that receive no votes will not have their ratings adjusted at all.
2) Cards that are voted on will have their ratings rise or fall, but on average they will stay the same. This is important because people will probably tend to vote on the best cards while the worst cards will not receive any votes at all. Hence, if you just z-score the community ratings you'll pull down the cards voted on
3) Each cubetutor rating is a z-score, which means it tells you how many standard deviations above the mean that card is. For example, Karn Liberated is 3 standard deviations above the mean, Fact or Fiction is 2 standard deviations above the mean and Elvish Mystic is 1 standard deviation above the mean.
4) We calculate the z-score of the community ratings, and then we add in the average cubetutor z-score for the set of cards that received votes (this avoids the issue mentioned in #2 - we sidestep the selection bias of people only voting on the best cards.)
5) We generate a weighted aggregate ranking. The weight on the community rating grows with the number of votes the cards received. For example, for cards that receive no votes the community weight is zero, for cards that receive 4 votes the community weight is 44%, for cards that receive 20 votes the community weight is 75%. The formula is stated in the original post.
Good work! I recommend you put privacy protection on so people can only edit a blank column, or something more advanced.. As of right now, I can tinker with everyones ratings. So 1 malicious person can wipe out everything, unless you have a plan for this..
Good work! I recommend you put privacy protection on so people can only edit a blank column, or something more advanced.. As of right now, I can tinker with everyones ratings. So 1 malicious person can wipe out everything, unless you have a plan for this..
Yup, was just editing your ratings to check if this was possible (I undid my changes..).
That helps a little bit, thanks for the answers. I think it's a bit odd that players can throw low rankings onto cards they just want to lower average ratings for... I think it would be nice if players were capped on the number of votes, or capped on only voting for cards they like. It should be a place to campaign for your favorite cards, not to be able to intentionally lower the score of cards other people like.
That helps a little bit, thanks for the answers. I think it's a bit odd that players can throw low rankings onto cards they just want to lower average ratings for... I think it would be nice if players were capped on the number of votes, or capped on only voting for cards they like. It should be a place to campaign for your favorite cards, not to be able to intentionally lower the score of cards other people like.
It's a bit of a twist on previous power rankings contests but I'm willing to give it a go (I don't see anyone else filling the void). I for one intend to rate cards seriously and objectively, whether I like them or not, and I would expect most users here to do the same. Even in a rankings list, you could hurt cards by not listing them at all. The scope for "downvoting" here is greater, but I don't think it should be much of an issue given the integrity of the people on this forum - as far as I can tell. It's just a bit of casual fun after all.
I plan on only voting for the top cards, and going from there. If we want to find out what the 20 best cards are in a given color, the random 5 and 6 voting won't matter much. Plus, it's going to be blended with the super arbitrary CT data, so it doesn't matter much anyways.
Concept
This is a project to utilize a large data resource, specifically www.cubetutor.com draft picks, to create a statistical ranking of the best cube cards by color and CMC. This is a good data resource, because it has a 100k+ cubes and hundreds of millions of draft picks. For example, just the ~3500 cards considered in the project were picked over 123 million times on www.cubetutor.com. Because the sample size is so high, we can have increased confidence that the results reflect the actual preference of cube drafters, rather than just random luck.
This project takes inspiration from the annual power rankings last held in 2016, which I have found to be an awesome and useful reference.
The main difference is the power rankings were based completely on polling of mtg salvation regulars. In contrast, these power rankings are at least initially based completely on cubetutor ratings (later, I joined the data with gatherer information which allowed me to include gatherer user rating as a feature as well as adjust cube count for the number of reprinting of each card) in order to reduce the amount of parsing work required and increase the sample size of the data.
Voting is now CLOSED - the results have been updated to reflect the voting and the netted rankings from all the data (cubetutor, gatherer and mtgsalvation votes).
Goal
Use data to comprehensively rank the top cards overall, by color and also by CMC within each color.
Features, Feature Statistics and Pre-processing
We are currently using four features:
1) Ln(Pick %): Pick % is the percentage of times the card was selected by a drafter in a pack. The mean pick percentage is 13.6% and the card with the highest pick percentage is Black Lotus (86%), followed by Ancestral Recall (73.7%) and Sol Ring (73.5%). Anecdotally, pick percentage corresponds very strongly to cards considered to be the best. The distribution of pick percentages is extremely condensed with the vast majority of cards lying in the [10%, 15%] range while a few outliers like Black Lotus are about 16 standard deviations above the mean. Because of this, the distribution has extremely high skewness and excess kurtosis - in other words, it is very far from normally distributed. To mitigate this, I take the natural logarithm of the Pick %, which helps to condense the outlying numbers. On a natural logarithm scale, Black Lotus is only 9 standard deviation above the mean, and the skewness and kurtosis is significantly lower (distribution closer to normal). Lastly, I standardized the Ln(Pick %) to have zero mean and unit standard deviation. This standardized score is given 60% total weight on the final score.
2) Cube Count / Expected Cube Count: Cube Count is the number of cubes in which the card is included. The most included cards are Evolving Wilds (37881), Lighting Bolt (36037) and Counterspell (32355). While a valuable feature, this feature does not correspond as strongly to the "best cards", since cards like Black Lotus (9220), Ancestral Recall (7427) and Sol Ring (17078) may be included in fewer cubes due to availability, price
or style (e.g. powered or unpowered). However, this feature does have value, primarily because the Pick % is much more reliable when the number of cubes including the card is large. If you created a ranking based only on Pick %, then Pet Project would be ranked #24, with 28.72% pick percentage and a cube count of 545. Furthermore, Kindle would be #34 with a 24.6% pick percentage and 2001 cube count, which is clearly ridiculous. Lastly, I found there was an extremely strong linear relationship between #of reprints and cube count - in fact a linear regression of # of reprints versus average cube count for cards with that # of reprints has an R^2 of 94% (see attached picture)! In contrast, # of reprints has zero correlation to Pick % or Gatherer User Rating, so there's little reason to believe that # of reprints is a feature indicative of cube card quality. This suggests that by controlling for the influence of reprints (availability and price proxy) on Cube Count, we can improve the relevance of the Cube Count feature. You can see the relationship in the attachments Lastly, I standardized the (Cube Count / Expected Cube Count) feature to have zero mean and unit standard deviation. (Cube Count / Expected Cube Count) was given a 15% weight.
3) Ave Cube Bonus: Cube tutor also maintains 360, 450 and 720 average cubes, broken out by unrestricted, peasant or pauper. Although less granular than the "Cube Count", this tells us the popular cards and has the additional advantage of excluding pauper and peasant cards that would not make the list in cubes where money is no issue. For example, Murder is excluded from the 360 and 450 lists even though it is in the top 300 cards by cube count. I gave cards 1 point for being in each list (so cards in none of the lists have 0 points and cards in all of the lists have 3 points). Ultimately, this feature is still 80% correlated to the "Cube Count" feature and also has some quirks (like Library of Alexandra doesn't even make the 720 list). I standardized this feature to have zero mean and unit standard deviation and gave it only a 10% weight.
4) Ave Gatherer Rating: This is the average gatherer user rating across all versions of the card. For versions of the
card that had no user rating, the rating was defaulted to 4.0 (which is is average user rating). This feature turned out
to be 31% correlated to Ln(Pick %) and clearly tends to be higher for good cards than bad cards. While not a particularly
strong feature, feature diversity in general can help improve our ensemble and eliminate some random variation. Once
again, this feature was standardized to have zero mean and unit standard deviation, and given a 15% weight on the final
score.
For more information on feature correlation and statistics, see the attachments.
Weaknesses / Challenges
A numeric ranking is only as good as the features available from the data.
In this case, the features are decent but there will always be outliers and the ranking will never be perfect.
For example, I think Channel is ranked much too low (#87) but this is because of its relatively meager pick % of 18.66%. Part of this could be due to more generic cards / universally playable cards (like Karn or Wurmcoil Engine) having much higher pick percentages than more powerful, but narrower cards like Channel.
Another big challenge is that unlike, say, a Linear Regression that aims to explain the sale price of houses, there is no actual numeric target label that we are trying to explain. Instead, we are trying to match something inherently subjective and undefined - how "good" magic cards are in cube setting. The best we can do then, is try to find decent proxies of goodness in cube. Currently, the strongest proxy seems to be pick % in cube. If we believe there's a good degree of correlation between pick % and the strength of cube cards, the best way to identify potentially valuable features is to find other, somewhat distinct features that still have a significant degree of correlation to the strength of cards in cube.
Cards Considered
I selected the top 2000 cards based on both Pick % and Cube Count. Because there is some overlap, this works out to 3,444 cards.
Furthermore, I filtered out cards with a cube count of less than 2300, because this leads to a lot of false positive cards with small sample size, like Kindle. This left me with 2111 cards total.
For cards that missed the "720 average cube" list entirely, I allowed myself to filter out the following cards that were in the top 1000 and which seem to be complete outliers. For cards that make the 720 average cube list, in contrast, there is no outlier filtering allowed:
Furthermore, there were a few cube worthy cards outside of the CubeTutor 720 Average Cube List that are simply criminally underplayed, which I added back in for consideration:
Link to Full Card Rankings and Data
https://drive.google.com/file/d/15BShDHNTGlUlKc6rzZe89ItJ1Xrb_kDx/view?usp=sharing
Cube Tutor Top 300:
These are the top 300 cube cards according to this methodology.
I've attached images showing how the rankings broke out for the top 50 cards.
By Color and CMC
0-1:
2:
3:
4:
5:
6+:
0-1:
2:
3:
4:
5:
6+:
0-1:
2:
3:
4:
5:
6+:
0-1:
2:
3:
4:
5:
6+:
0-1:
2:
3:
4:
5:
6+:
0-1:
2:
3:
4:
5:
6+:
VOTING (OPEN):
I created an editable spreadsheet so that anyone can vote on any of the over 2,000 cube cards in the rankings.
To vote, type your username in the top row of a new, unused column and save your scores for the cards you care about in a 0-10 numerical range. You can vote on as many or as few cards as you wish. Once voting closes, the average user rating will be used to adjust the numerical ratings to get combined ratings. This way we will have three difference rating systems: 1) cubetutor/gatherer numeric data only, 2) user ratings based, 3) combined scores (combining #1 and #2, i.e. using the user ratings to adjust the cubetutor/gatherer numeric ratings)
VOTING RATINGS GUIDE:
10: Unequivocal P1P1 over everything else in the game. These cards are broken and make other great cards look silly. For me, the only 10s were Black Lotus, Ancestral Recall and Sol Ring)
9: Power Nine Level Cards - these are in the top 20 most powerful cards in powered cube. For me, a few examples were Moxen and Mana Drain.
8: A cut below the power 9 in level, these cards tend to be the strongest in their color and often still make P1P1. Some examples for me were Vampiric Tutor, Lightning Bolt and Wurmcoil Engine.
7: The are solid role performers that typically make even the smallest powered cube lists (360). You are usually never unhappy to play these cards. Some examples for me were Blade Splicer, Arbor Elf, Shriekmaw, Flametounge Kavu and Treasure Cruise
6: These are cards you would play at 450 but not at 360.
5: These are cards you would play at 540 but not at 450.
4: These are cards you would play at 720 but not at 540.
3 or below: You would not play these cards even at 720.
LINK TO VOTING SPREADSHEET:
https://docs.google.com/spreadsheets/d/15xe5pY2niCsEisv5Nu-vK-NppLiGG2Pu9BkbipExZ9Q/edit?usp=sharing
How will voting change the results?:
Voting will close 3/31/2019 assuming there are enough votes. Otherwise, voting may be extended another quarter.
Once voting closes, the average scores across all the cards that received votes will be standardized to have zero mean and unit standard deviation, and then we add in the average cubetutor z-score for the set of cards that received votes. The reason we add this mean is that people will tend to vote mostly on the best cards which already have positive cubetutor z-scores and we don't want to bring down the mean of these good cards due to selection bias.
To determine the "combined scores", we treat the numeric rating based on cubetutor/gatherer data as a prior and update it based on the number of votes received - cards that received many votes will see their scores adjusted significantly while cards that receive few votes will not have their scores adjusted much.
Specifically, because the statistical significance of the average user rating scales roughly with the square root of the number of voters, each card's combined score will equal:
z_combined = z_cubetutor * (1-sqrt(n / (n+16))) + sqrt(n / (n+16)) * (z_ave_user_rating+z_cubetutor_mean_score_of_cards_receiving_votes)
where n is the number of votes, z_cubetutor is the numeric cube tutor score, z_ave_user_rating is the z-scored average user rating, and z_cubetutor_mean_score_of_cards_receiving_votes is the average cubetutor z-score for the set of cards that received votes.
For example, if 16 users vote on that specific card, the combined score will come 50% from the cubetutor ranking z-score and 50% from the user ratings (which have also been standardized into a z-score). If zero users vote, the user ratings count for 0%, if 4 vote it counts for 42% and if 20 vote the user ratings count for 75% of the total score.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
The one thing that sticks out like a sore thumb to me is Compulsive Research in the top 100?
My High Octane Unpowered Cube on CubeCobra
Some things I consider weird (that can probably be explained due popularity):
- Doom Blade but no Go For The Throat, Damnation, or Toxic Deluge;
- Emrakul, the Aeons Torn is in the list : )
- Splinter Twin is there but Kiki-Jikki is not;
- There's a signet(!?!) in the list;
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!
Thanks for the initial feedback :).
Compulsive Research I don't think is too crazy as a top 100 card - that gets picked highly by cube drafters like LSV, Reid Duke and NumotTheNummy (I watch a lot of cube draft).
Signets (especially the blue ones) also place highly, and deservedly so in my opinion. Dimir Signet is #94, Izzet Signet #106, Azorius Signet #130, Simic Signet #158. Non-blue Signets are a little lower, but still make the top 360. Again, cube drafters like LSV, Reid Duke and NumotTheNummy take the signets very highly, and the fact that all the signets place highly suggests that Dimir Signet at #94 is hardly a fluke.
Emrakul, the Aeons Torn being in the list at #76 is not surprising. This is a bomb card in the MTGO Vintage Cube which supports re-animator strongly with cards like Entomb, Buried Alive, Shallow Grave, Corpse Dance. Also, it's pretty much the best creature for Sneak Attack and Through the Breach. Once again, top cube drafters take this very highly. For example, LSV calls Emrakul a very high pick in this article: https://www.channelfireball.com/articles/the-ultimate-guide-to-cube-archetypes-blue/
For the cards mentioned not in the top 100, here are the current placements:
1) Go For the Throat is #153
2) Toxic Deluge is # 343
3) Hero's Downfall is #251
4) Damnation is #181
5) Kiki-Jiki is #111
I absolutely agree that all these black cards are WAY above Murder.
Murder in top 100 is a crime, while Toxic Deluge might actually deserve a spot in the top 100.
This is not an issue with the algorithm, but a data quality issue.
Fundamentally, Murder just has a higher pick rate at a large sample size.
The issue, of course, is that Murder was a high pick in block cubes or something or the sort, so it is picked highly over much weaker competition.
Unfortunately, there's no way to correct for this right now, so there are 3 ways to improve the quality of results:
1) Get some more data. If there's another website with data like cubetutor.com, combining the data or creating an ensemble of results could eliminate a lot of quirks like this.
2) "Obvious errors" like Murder can be manually removed (can have a list at top mentioning the cards removed)
3) Use the cubetutor results as a "prior" (e.g. pick the top 40 cards of each color) and then do community runoff polls to determine final placement.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!
It's almost certainly this.
This is also why pacifism is above Venser, Shaper Savant for example.
The different card quality in block, core or pauper drafts is the issue, and there's currently no way to separate out those cube picks.
This is currently the biggest issue with the rankings - I posted three potential solutions earlier with #1 being my favorite if anyone else knows another data source
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
First I have to join this data with card info data so I can sort it by color and cmc.
Then, I'd like to do polls by color and combine those results with the numeric scores for a final result.
However, I haven't found a good polling solution yet - ideally it would allow bulk uploading of each card name as a separate "question" that voters could give a numeric score (e.g. 9.5 for black lotus, 1.0 for murder)
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
Although the "top cards" section of cubetutor is based on picks from all cube environments, cubetutor also has "average cubes" of size 360, 450 and 720 broken out by price unrestricted, pauper and peasant.
Although Murder is in the Top 360 cards according to "Top Cards" by both Pick % and Number of Cubes, it is excluded in both the 360 and 450 price unrestricted Average Cubes, appearing in only the 720 price unrestricted average cubes. Clearly these average cubes are using more granular data that partitions cubes by type. Although this raw data is not available, I think the presence of cards in these Average Cubes list could be another feature (with the most points for being in the 360 list, less for 540 and the least for 720).
Here's the price "unrestricted" cubes:
http://www.cubetutor.com/viewcube/492
http://www.cubetutor.com/viewcube/893
http://www.cubetutor.com/viewcube/495
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
I'm assuming you meant Emrakul, the Promised End, since Emrakul, the Aeons Torn is nigh impossible to utilize in Reanimator. However Emrakul 2.0 isn't in the list, so I'm confused.
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!
For blue, I provided the top 60 cards because blue has about 3x as many cards in the top 300 as the other colors (no great shock...).
Next step is to also break it out by CMC and if there's interest do polls to further refine things.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
Even if it did make the list based on pick percentage, it would still be filtered out, along with a lot of false positives like Kindle that are properly excluded, because it is included in <2300 cubes.
To get cards like Fractured Identity that slipped through the cracks, I'd need to expand the analysis to even more than the 3500 cards initially examined. I do have to say that I'm shocked this great card is so underplayed; hopefully it will see more play in a year from now.
I will most likely just post a link to the full spreadsheet at some point which would show you where every card is ranked.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
For what it's worth, I think you'd probably just need to go through by hand and throw out cards you think are there because of peasant/pauper cubes (pacifism, murder, etc) - I think you mentioned that you might need to do this.
Regular 450 unpowered cube (with some custom cards) - 450 Unpowered
Yeah I pretty much already did this at this point. There is a reasonably long list of excludes in the original post.
FWIW I currently control for # of reprints in the cube count since this is a strong relationship.
If there was a source that listed the print count by common, uncommon, rare and mythic I could pretty easily calculate the total circulation of every card and control for that, which should boost the ranking of low circulation cards like Fractured Identity a lot.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
This is your chance to vote on which cards you think are best and influence the rankings!
Link to the voting spreadsheet below, as well as how I plan to use the user ratings to influence the rankings.
VOTING (OPEN):
I created an editable spreadsheet so that anyone can vote on any of the over 2,000 cube cards in the rankings.
To vote, type your username in the top row of a column and save your scores for the cards you care about in a 0-10 numerical range. You can vote on as many or as few cards as you wish. Once voting closes, the average user rating will be used to adjust the numerical ratings to get combined ratings. This way we will have three difference rating systems: 1) cubetutor/gatherer numeric data only, 2) user ratings based, 3) combined scores (combining #1 and #2, i.e. using the user ratings to adjust the cubetutor/gatherer numeric ratings)
LINK TO VOTING SPREADSHEET:
https://docs.google.com/spreadsheets/d/15xe5pY2niCsEisv5Nu-vK-NppLiGG2Pu9BkbipExZ9Q/edit?usp=sharing
How will voting change the results?:
Voting will close 3/31/2019 assuming there are enough votes. Otherwise, voting may be extended another quarter.
Once voting closes, the average scores across all the cards that received votes will be standardized to have zero mean and unit standard deviation. To determine the "combined scores", we treat the numeric rating based on cubetutor/gatherer data as a prior and update it based on the number of votes received - cards that received many votes will see their scores adjusted significantly while cards that receive few votes will not have their scores adjusted much.
Specifically, because the statistical significance of the average user rating scales roughly with the square root of the number of voters, each card's combined score will equal:
z_combined = z_cubetutor * (1-sqrt(n / (n+16))) + sqrt(n / (n+16)) * z_ave_user_rating,
where n is the number of votes, z_cubetutor is the numeric cube tutor score and z_ave_user_rating is the z-scored average user rating.
For example, if 16 users vote on that specific card, the combined score will come 50% from the cubetutor ranking z-score and 50% from the user ratings (which have also been standardized into a z-score).
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!
I choose voting from 0 to 10 as this is usually intuitive for voters. However, if this is too arbitrary I could try to add a ratings guideline. Ultimately everyone has their own subjective opinions of what is best and the goal is to let as many people vote on whichever cards they want to, and ignore cards they don't care about. Voting is definitely a more subjective process than the pure numbers based ranking, but I think the community can provide meaningful ratings and contribute to improving the rankings. Plus, it's fun to poll people and weigh in on rankings like these :).
I can certainly try to provide a metric, which I'll include in the original post.
Ratings:
10: Unequivocal P1P1 over everything else in the game. These cards are broken and make other great cards look silly. For me, the only 10s were Black Lotus, Ancestral Recall and Sol Ring)
9: Power Nine Level Cards - these are in the top 20 most powerful cards in powered cube. For me, a few examples were Moxen and Mana Drain.
8: A cut below the power 9 in level, these cards tend to be the strongest in their color and often still make P1P1. Some examples for me were Vampiric Tutor, Lightning Bolt and Wurmcoil Engine.
7: The are solid role performers that typically make even the smallest powered cube lists (360). You are usually never unhappy to play these cards. Some examples for me were Blade Splicer, Arbor Elf, Shriekmaw, Flametounge Kavu and Treasure Cruise
6: These are cards you would play at 450 but not at 360.
5: These are cards you would play at 540 but not at 450.
4: These are cards you would play at 720 but not at 540.
3 or below: You would not play these cards even at 720.
The ratings you are looking at are my ratings. The column entry says IMorphling89, indicating these are my ratings. The numbers you are seeing are just my subjective ratings. Most human beings probably don't want to vote on every card, so I just voted on the cards I care about. If you want comprehensive ratings for every card considered (~2500) then you'll find that in the original post based on the cubetutor and gatherer data.
Each user is free to type their name in the top of a new unused column and vote on as many cards as they wish.
Each user should vote only on the cards they care about. Cards you don't want to vote on, just leave the voting blank.
Thus, everyone can vote on as many cards as they want. Obviously we can't require everyone to vote on every card.
I've divided the cards by color to facilitate this a bit.
Once the voting ends, there should be many votes on many cards. As described above, we can use this combine this data with the initial (comprehensive) cubetutor z-score ratings to get aggregated ratings that also take into account these user ratings.
How do we do that? For for the aggregate ratings, we start with the cubetutor ratings as a prior, and we adjust from the prior based on the number of votes. Cards that get no votes won't have their rating adjusted at all. Cards that receive many votes may have their scores adjusted substantially if the community rating differs significantly from the cubetutor rating.
Specifically:
1) Cards that receive no votes will not have their ratings adjusted at all.
2) Cards that are voted on will have their ratings rise or fall, but on average they will stay the same. This is important because people will probably tend to vote on the best cards while the worst cards will not receive any votes at all. Hence, if you just z-score the community ratings you'll pull down the cards voted on
3) Each cubetutor rating is a z-score, which means it tells you how many standard deviations above the mean that card is. For example, Karn Liberated is 3 standard deviations above the mean, Fact or Fiction is 2 standard deviations above the mean and Elvish Mystic is 1 standard deviation above the mean.
4) We calculate the z-score of the community ratings, and then we add in the average cubetutor z-score for the set of cards that received votes (this avoids the issue mentioned in #2 - we sidestep the selection bias of people only voting on the best cards.)
5) We generate a weighted aggregate ranking. The weight on the community rating grows with the number of votes the cards received. For example, for cards that receive no votes the community weight is zero, for cards that receive 4 votes the community weight is 44%, for cards that receive 20 votes the community weight is 75%. The formula is stated in the original post.
360 card powered Chicago cube:
https://cubecobra.com/cube/overview/e7r
2020 Numerical Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/817969-2020-numerical-cube-power-rankings
2018 CubeTutor Power Rankings:
https://www.mtgsalvation.com/forums/the-game/the-cube-forum/cube-card-and-archetype/803301-cubetutor-power-rankings-2018-by-color-and-cmc
Last Updated 02/07/24
Streaming Standard/Cube on Twitch https://www.twitch.tv/heisenb3rg96
Strategy Twitter https://www.twitter.com/heisenb3rg
Yup, was just editing your ratings to check if this was possible (I undid my changes..).
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!
It's a bit of a twist on previous power rankings contests but I'm willing to give it a go (I don't see anyone else filling the void). I for one intend to rate cards seriously and objectively, whether I like them or not, and I would expect most users here to do the same. Even in a rankings list, you could hurt cards by not listing them at all. The scope for "downvoting" here is greater, but I don't think it should be much of an issue given the integrity of the people on this forum - as far as I can tell. It's just a bit of casual fun after all.
My 630 Card Powered Cube
My Article - "Cube Design Philosophy"
My Article - "Mana Short: A study in limited resource management."
My 50th Set (P)review - Discusses my top 20 Cube cards from OTJ!