by Doc – Owner, Founder, Still Thrilled About His Pokemon Trainer Amiibo Being #2 In Amiibots
The Real Amiibo Meta
Compared to your average trainer, I have always taken a very different approach when conceptualizing the “amiibo meta”. Most amiibo trainers look at recent tournament results, see which trainers are submitting certain amiibo more frequently and which amiibo are getting relatively farther in bracket than they’d expect, and then make pronouncements on “the meta” as if it’s League of Legends, Pokemon Go or another frequently-updated game.
This is, obviously, the wrong way to conceptualize the amiibo meta. Amiibo AI is set in stone unless Nintendo patches it, and over time amiibo trainers will innovate ways to manipulate the amiibo AI into being a better and better fighter. In other words, the preferences and quirks will be with us forever, but we’ll train better amiibo over time regardless. This is all assuming that information is shared freely (which doesn’t happen in practice) and that tournaments are free of integrity flaws (which sometimes doesn’t happen in practice, though it’s rare), and that equal representation of amiibo is the norm (which never happens). We see incremental innovation in specific amiibo all the time, and sometimes larger experiments and innovations happen as well.
Because of this inevitable innovation against the backdrop of unchanging AI, the best way to analyze the “amiibo meta” is not as a fluid, temporal state of things. The best way to analyze it is to look at the theoretical endpoint: the 20XX, if you will. In a state of 20XX, every amiibo is trained with the maximally-effective training styles. It plays in a way that has the highest-possible win rate across all possible amiibo opponents. It is, with all its AI preferences and quirks, the best possible amiibo of that character.
Doing something like this in practice would be theoretically impossible, and would take tens of thousands of matches because of the inherent RNG in amiibo behavior.
So to analyze the overall amiibo meta, we have to take our best understanding of the theoretical “20XX” of that amiibo, and collate all the roster’s best amiibo together to analyze it. It’s a thought experiment, not an observation of recent events.
The Problem with Tournament Brackets
Most trainers rely on tournament brackets as the defining operator of the “amiibo meta”. This is because we’ve simply always used tournaments, and Amiibots is a pretty new invention.
However, tournament brackets have the unfortunate side effect of forcing out non-generalist amiibo opponents in the early rounds. If your Zelda happens to counter a few high tiers but loses to most low tiers as most Zeldas do, then she’ll sometimes be eliminated by low tiers in the early rounds before the low tiers themselves are eliminated, and only high tiers remain. Because Zelda is a non-generalist, she gets eliminated before we can see her performance against higher tiers.
Overall, the problem with tournament brackets is that only the highest-tier amiibo get a large number of matches, and the lower-tier amiibo get eliminated before any solid observation can be made. Unless tournaments are tier-restricted (which also soils the data by limiting participation against higher-tiers), low-tiers just don’t get much time in the light.
Amiibots Contributes to 20XX
Enter Amiibots. Amiibots is a 24/7 automated amiibo fighting stream on Twitch, in which a few hundred matches between user-submitted amiibo are ran each day. Trainers submit their amiibo files to the Amiibots system, and the Amiibots system matchmakes them using relative skill to fight on stream. It then spits out data cards such as this:
Amiibots makes a rudimentary, primitive form of 20XX possible by providing hundreds of unfiltered matches for amiibo to participate in. Amiibots allows trainers to pit their amiibo against any number of other amiibo at most tier ranges, and the matches happen automatically. It’s not playing against “all possible amiibo opponents”, but it’s a good start to one, and it bypasses the problem with tournament brackets. Basically, it gives us better data to analyze the amiibo meta.
From this information, I posit the question: should Amiibots be considered a better representation of the amiibo meta?
I haven’t run this idea by anyone yet, so I can only guess as to the various opinions that will be stated. I think there’s several solid arguments on both sides, so I’ll lay out four of them here:
- Amiibots gives every amiibo a fair shot at fighting opponents, and this creates better data
- Amiibots doesn’t require a tournament host
- Amiibots rulesets don’t change, so the data is consistent
- Submission trends don’t impact amiibots because so many amiibo are in the system
- Amiibots matchmaking soils the data by pitting amiibo of equal status against each other, instead of a truly random assortment
- Amiibots data can be biased by Tier 3 Twitch subscriptions, if the subscriber wanted to mess up the system
- Amiibots matchmaking sometimes fluctuates or has mathematical issues, such as the rating inflation when Min Min released.
- Amiibots doesn’t have the Best-of-3 system that tournament brackets usually do, and matches are often up to random behavior. Best-of-3 provides a more long-term comparison of the quality of each amiibo.
I’m not of either mind on this, and I see tremendous value in both Amiibots and tournaments being considered the “main meta”.
What do you think?