Wrapping Up the JMP Gladiator League

Written by J. David Smith
Published on 19 August 2020

Recently, I wrote a bit about the Gladiator League and how it came to be. Last week, the inaugural Jumpstart (JMP) league ended, with the Amonkhet Remastered (AKR) league taking its place on Thursday. I wanted to take a moment to look back at the league, the results, and the evolution of the bot during the almost-month of the first league.

League Results

Let me begin by setting the stage. The JMP League ran from July 16th until late on August 12th—a day shy of a full 4 weeks. During that time, 166 players completed 416 league runs totalling 1,305 matches. An average of 48 matches and 15 completed runs per day. Not bad for the first run, and—honestly—much higher than expected. The Gladiator community embraced the Sharktocrab with open arms.

Unsurprisingly, In natural systems, count data tends to be exponentially distributed, with a few individuals having large counts and the rest having very small counts.

Those of you more familiar with statistics will note that the chart is flipped from a typical presentation of count distribution that would make this more obvious. the number of league completions each player has is pretty heavily skewed. The typical player that completed at least one run completed 2 runs, but 13 players completed at least 10 runs and two players completed over 20!

Number of matches played during each hour of the day. This chart automatically uses your computer's local time.

The times at which people play are fairly Amero-centric, but not to the level that I was expecting. This is a bit surprising: the original Gladiator player pool largely came from Wheeler's community, and he is on Pacific Time (GMT-7/GMT-8). The peak hours are much more spread than one would expect if there were only players in North America—which I'd guess is due to a pretty sizable contingent of European players.

In the same vein, we can see a HUGE spike in matches played in the first 3 days of the League. That tapered off, remained fairly stable until the final week of the League. I am curious if that last week of low play is due more to AKR's impending release or tiredness of the League. To answer that, we'll need to wait until we have a bit longer between set releases.

Breakdown of color representation in each week of the league. A deck is listed as having a color if it has at least 5 cards of that color, and each deck can contribute to multiple colors (so each box does not necessarily sum to 1). All decks that completed all 5 matches are included. Decks are assigned to a week based on registration date.

I tried doing all color pairings, but it was honestly illegible.

One of the hot topics during the league's run was, put simply, Green. Many players (myself included) are of the opinion that Green cards were massively pushed relative to other colors in recent standard sets. People are—understandably—concerned that this may translate to a similar level of relative strength in Gladiator.

Breakdown of color representation in each week of the league, filtered to only include decks with at least 3 wins.

Before going further, I want to stress that League data is not a good source for win rates. While it is appealing as one of the only sources of structured match data on Gladiator, there are a couple of reasons to be very careful when looking at League winrates. First: the League is very deliberatly not competitive. Many players are not trying to play the best deck, only the best version of their deck. This has major implications because power is meta-dependent and the League is very unlikely to have a meta developed around pure power in the same way that Standard may.

Second–and this is crucial—we do not have enough data to draw serious conclusions for all but the most egregious outliers. 1,300 matches may sound like a lot, but with a high-variance format in a high-variance game in a meta that isn't centered on trying to win at all costs this data just isn't sufficient to draw strong conclusions. I am not saying that we should abandon hope, just that we should be careful.

This league is actually a great example of how metas evolve and change—and why we shouldn't be too quick to draw conclusions. We can see above that in the first two weeks of the League, decks with Green scored 3+ wins at a much higher rate than any other color. Many of these decks are built on Craterhoof Behemoth, one of the hottest cards from Jumpstart. A great example of this is Roseface's hoof list—the very first 5-0 of the league.

Hoof decks are explosive, and will run over anyone trying to durdle around without interacting. Later in the league, we saw an uptick in the amount of interaction played Check out DemonDrinkingTea's Mardu list for a good interaction-heavy example from later in the league. Note that on top of the removal package, it plays 4 board wipes (not counting Ugin).—especially board wipes, which are incredibly difficult for hoof decks to deal with. While SupremePhantom's (multiple) Bant lists all play Green, they also play (most of) the on-color board wipes. Interaction is valuable across the board, and board wipes happen to be very good in a format built from several years of creature-centric Standard sets. These factors combined with hoof no longer being the hot new toy to bring Green back in line with other colors.

A Combo 5-0

There is one deck in particular that I want to shout out, because I believe it is a first for the format. UFO6300 5-0'd the league with a dedicated combo deck. Historically, No pun intended. we haven't had access to actually good combos in this format. Pre-Ikoria, the only combos that I know of involved 3 card loops with either Teshar or Naru Meha that were expensive, hard to assemble, and easy to disrupt.

UFO's list uses Emergent Ultimatum to assemble something just shy of a 1-card combo. You find Omniscience, Peer into the Abyss, and Ob Nixilis, Hate Twisted. If the opponent chooses to shuffle Omniscience, you cast Ob Nixilis and then Peer into the Abyss and dome your opponent for 40-50 with Ob's static ability. If the opponent chooses to shuffle one of the other pieces, then you use Omniscience to dig for it—if you have a tutor in hand you can guarantee that you'll find it and win immediately. They killed me on stream with this list. I definitely picked the wrong option the first time.

This is the first combo I've seen do well I don't count "ramp into Omniscience and then durdle for 15 minutes" as a combo., and I'll be interesting to see how it and other combo decks evolve in the format.

Changes & Updates

As much of the bot JMP card-identifier issues notwithstanding. had been built prior to the start of the JMP league, I was able to focus on adding new features during the League. Below, I want to highlight a couple of them. You can check out the complete changeset for the JMP league here.

The API

The first addition I made was the introduction of a GraphQL API. While I still have more improvements to do to this on the technical end, It still dispatches far, far too many SQL queries for my liking. it is relatively complete—and I actually used it to obtain all of the data used in this post!

Deck Viewer Improvements

One of the oft-understated maxims of computing is that names are hard. In that vein, displaying cards in the deck viewer has been a fairly consistent pain point. During the JMP League, I made rendering more consistent (it will now always show the front face of a card). Additionally, I added a decklist export button! This seemingly-simple change actually had huge ramifications, because you can only access the modern clipboard API when connected over SSL. As a result, I had to update the League site to use SSL. This feature still needs improvement, as it currently relies on the oft-flaky TappedOut card tooltips to rename card entries.

Color Indicators on the Standings Page

One of my favorite things about the Penny Dreadful Standings page is their gorgeous color indicators for each deck, giving you an immediate look at what colors the list contains. This league, I implemented a nigh-copy of it for the Gladiator league. When I say nigh-copy, I mean it. The implementation on the front-end is very similar, and I shamelessly stole their colors, which happen to work well with the existing site colors.

Screenshot of color indicators on standings page.

Upcoming Change: League Start & End Time

One thing that I'm not happy with from the current League is how the rollover from JMP to AKR worked. Internally, I use UTC (GMT+0) dates for everything—including League start and end dates. This meant that the JMP League ended at 5pm on August 12th for players on the West Coast of North America. Rather awkward, I think.

For the next League, this will be changed to be more inline with Arena's update schedule. We haven't worked out the exact time-frame yet, but you can expect the AKR league to end late the night before Zendikar Rising releases and for the ZNR League to start shortly after Arena downtime.

Wrapping Up

On the whole, I'm very happy with how the JMP League went. I have more improvements to the League bot planned for this month, and am looking forward to seeing how the league and the format evolves over time.

If you have any questions, feel free to reach out to me on Discord. The code for the charts contained in this post can be viewed here.