so interesting to see the continued removal of mono-red aggro cards.
Well, with recent updates it seems the trend is for a slight preference for combo focused cards, and here we see that continue with cards like Sylvan Safekeeper and Bazaar of Baghdad both making into the cube for the first time since I've been running this algorithm. Other changes of note are the removal of Goblin Guide which is long overdue; I'm quite happy about that as it always seemed to be a 15th pick. Additionally, Daretti, Ingenious Iconoclast leaves the cube for the first time since I've been curating it.
Funny, the other thing I'm excited about with this update is that I get to remove cards that make more game objects!
reran the algorithm - here are the updates.
It's only been a month, but I felt like seeing where things were so I ran the algorithm. It looks like folks have been tinkering with their lists a good deal as there were some interesting changes. We got our first BLB and DSK cards in Jacked Rabbit and Fear of Missing Out. We also have one surveil land making it into the list with Thundering Falls. It's interesting that Shelldock Isle has been edged out in favor of a surveil land. I guess the conditions for triggering shelldock may be considered a bit slow these days.
No significant changes this update. Most interestingly, there are no cards from any of BLB, DSK, or FDN in this update.
The one interesting change is actually the removal of Regrowth, which has been in the cube since I began curating it. I find this interesting because its always a big deal when one of the LEA cards leaves the vintage cube because it gets powercrept out. @aquaone believes that many alpha cards will be removed from the data generated vintage cube over the next few years. In particular we have a running bet where he believes that by 2026, counterspell will be removed from the cube. Time will tell.
Just a small update before we draft this again. Not the final lock in list for CubeCon - will write more on this shortly.
This cube is being featured the week of June 30th-July 06th. So here is an update to ensure that folks get to enjoy a fresh update to the cube. Re-ran the algorithm just now and sampled 51 cubes updated within the past 90 days and meet the other sampling criterion. MH3 cards continue to filter in to the cube this time, Phelia, Exuberant Shepherd, Emperor of Bones, and Nadu, Winged Wisdom.
Enjoy!
A couple of MH3 cards are starting to pop up. Looking further at the data, other MH3 cards popping up in vintage lists but not with enough frequency to be in the cube yet are Emperor of Bones, Ghostfire Slice, Phelia, Exuberant Shepherd, Ocelot Pride and Six. I'll be updating in the coming weeks and months to see how the post mh3 landscape continues to change.
Bribery is leaving. Interesting because this was always considered an unfun card to play against. Now, it seems that it won't be good enough in a world of game ending 3 and 4 drops.
Outside of this a lot of x/1 dorks are leaving in and
- If I had to guess, I suppose that would also be due to the fact that these cards get an increasingly narrow window to shine. Compared to some hasty 3 drop that can swing games when topdecked, these are likely to be dead draws if not in your opening hand.
My goals with this cube have always been to make an environment that is representative of what cube designers are doing and is simultaneously up-to-date with newer cards. I'm generally happy with the output, but am always looking for methods to improve the process. In a prior blog post @djsterlingsilva brought up an article on how the MTGO Vintage cube is moving back toward a Platonic Ideal, and asked if I was using the MTGO Vintage cube as a data point in sampling. It's been some time since I had tried sampling MTGO vintage for generating a 360 card cube, but it's been a few years and my algorithm and sampling process have undergone quite a few changes in that time. So, I generated a cube sampled with all the data I would normally pull down but also added sampling for some popular cubes in the community that down meet my criterion.
This update is the result of the combined sampling of the 360ish card vintage cubes plus these select cubes. These five data points are fairly heavily weighted in the outcome we see here as they are popular and are frequently updated. I'm happy with the result and I think it does do a better job than simply sampling just the smaller lists. Going forward I'll be adding these cubes to the sample.
This is only 24 swaps to the cube, but so much changes with this update.
Since CubeCon 2023, there have been several methodological changes to how the cube is generated - the biggest being changes to the sampled data in updates like this and the one a few months prior where we started pulling data from the AWS bucket. This has lead to a lot of changes in a short amount of time, though note that a majority of those changes are not brand new cards printed.
I ran the algorithm today and am updating for May 2024. Overall, the trend of adds looks good and contemporary with what I see in drafts. We finally get the inclusion of the LTR landcyclers - interesting only island and swamp cyclers make the cut. In the last update I was excited to see Zurgo Bellstriker go; in this updat we see the continued trend of replacing mono-red cards with more value oriented bombs in the swap of Eidolon of the Great Revel for Gut, True Soul Zealot
A few changes that feel akward to me are the removal of Celestial Colonnade and Palantir of Orthanc - which both seem like fantastic cards. I'm also not excited about the inclusion of Abrupt Decay which seems like a very mediocre card. This could just be due to sampling as I note that in the current update I sampled a slightly smaller number of cubes than in the last couple of updates; I'm hoping that after MH3 we'll see an uptick in cube designers updating their cubes and thus give me more data points for sampling.
I updated the cube sampling process to automatically sample from Cube Cobra AWS bucket data - this removes the last manual process whereby I had to review the cubes to sample quartelry. Now, the cube can be generated with a faster turn around time and the sampling process will automatically adhere to coded parameters without introduction of human error in the sampling process.
It's time for a quarterly update to the cube. Here's a quick bullet point list of my thoughts
Winter is winding down here in the Pacific Northwest, time to bust out the Spring Factory.
I love me a classy LEB card, but 7ED Birds is 🤌🏻
Speaking of Beta cards, I updated the Pearl.
I tweaked the quarterly update cadence to align with roughly when CubeCon lists must be submitted.
An interesting thing about this update is that Time Vault made the cut. Sadly, the only ways of untapping the vault in the cube are Pestermite and Zealous Conscripts. Bad card inclusions will happen from time to time when we are aggreatating data. In cases like this the vault is part of a 2 card combo where vault itself is a unique effect only printed on a single card while the untap effect can be a myriad of things including Voltaic Key, Manifold Key, Mirage Mirror, Scheming Fence and Kiora, Behemoth Beckoner to name a few. Because there are so many pairings with vault, they tend to cancel each-other out when doing frequency analysis and none of the other halves of the vault combo make it into the cube. So, cube curators are putting Time Vault into their cubes, but the untap effects are varied enough that they get washed out of the analysis. I'll keep the vault in this cube update, but I ran the algorithm with a blacklist consisting solely of Time Vault and based on that will be playing Lodestone Golem in it's place.
Outside of that change, I'm happy to see both Lightning Helix and Bloodghast go as they were too narrow.
Today is the lock in date for main event cubes for CubeCon 2023. This also servers as an early Q4 update to the cube. I'm particularly excited about this update as I've done a large amount of work to tweak the card weighting system.
When I first started generating these cubes, cards were included based solely on frequency of appearance in cube designer's lists. Over time I've incorporated a range of variables to weight individual cards. I won't go into too many of the details on how I weight each card so for now here is a high level bullet point list of things that go into weighting a card.
If you are curious to know more you can message me or chat with me in person at CubeCon!
One of the reasons for the above weighting update is that I have always felt I struggled a bit to simultaneously capture my two cube design goals of
I always felt I was able to capture goal #1 a bit better and that goal #2 was slow to catch up (e.g. I think it took more than a year for Solitude to make it into the list). With this update it seems we get closer to accomplishing both goals simultaneously. Here we can see the exclusion of non-relevent but popular cards smokestack and braids, cabal minion and then the inclusion of new hits such as the one ring and orcish bowmasters despite low play rates in sampled cubes of 32% and 39% respectively.
I'm truly looking forward to sharing this cube with everyone at CubeCon and having some great design discussions too!
I've been spending quite a bit of time working on improving both the sampling methods used to generate the cube as well as finding good ways of curating the input data. The 5.9 release has changed search on Cube Cobra, which means that how data is pulled for this cube has been updated to be a bit more of a manual process where I need to review what's being added to the input algorithm more closely.
The latest update has reduced the number of cubes sampled to 11, resulting in a broader frequency cutoff. This caused many cards with a frequency of 5 to be at the threshold of inclusion/exclusion in the cube. Previously, the sorting logic for cards at this threshold was based on alphabetical order, meaning that all cards starting with "A" with a frequency of 5 would make it into the cube regardless of their usefulness to the cube community. To improve this, I manually reworked the secondary sorting criteria to be based on ELO score descending. This way, we prioritize the cards with the highest frequency in the data set and then sort by ELO rank when there is a large band of cards in the cutoff range. These changes are reflected in this blog post.