by Doc – Owner, Founder, Really Should Be Writing His Admin Law Final Right Now
Ever since the blockbusting success of my narrative-documentary, “The History of the Amiibo Metagame“, a lot of new trainers have flooded into the amiibo community since last summer. I’ll pat myself on the back for a moment before moving on.
*pat pat pat*
When these new trainers show up, they always ask the same question: do I have a good amiibo too? Typically they’ll seek out a more experienced trainer who specializes in their character and ask for that trainer’s opinion, and whether they should reset it or keep improving. Naturally, there’s a large disparity between the number of experienced trainers who eat, sleep and crap amiibo training and the number of new trainers who want their amiibo examined. This creates a problem where some trainers are familiar with how that amiibo can be improved, and other trainers are left to their own assumptions, not kept up to date with the advancements of the amiibo meta.
Fortunately I planned for this disparity, which is why the Amiibo Doctor training guides section encourages readers to reference multiple training guides when available. Amiibo Doctor guides exist to crystallize the best amiibo training methods at the time of writing the guide, allowing new trainers to base their amiibo on successful predecessors, giving them a reliable foundation on which to build. That way, if experienced one-on-one feedback isn’t available, trainers can at least get an idea of what to aim for when trying to improve their amiibo, closing the disparity to some degree and keeping them from falling too far behind in the meta.
But what are the “best” amiibo that we should be building our foundations on? How should we determine what the “best” amiibo is?
The Best amiibo
The community has never really settled on how to define what a “best” amiibo is. Suppose an amiibo takes first place in a supermajor tournament, but gets crushed in every other tournament it enters for the rest of time. Should we call that a “good” amiibo, or just lucky? What if it places 8th in two supermajor tournaments but then takes near-last in everything else? Is it “good” now?
This question often arises when it comes to asking around for people to write amiibo training guides. When I first started offering high-achieving trainers money to write an amiibo training guide, I imagined that most amiibo trainers would be jumping at the chance to make a few bucks off their long history of tournament placements and their reputations as the best trainer of that character. Instead, most trainers turned it down, admitting that they don’t feel qualified to write a guide for this amiibo. I would sometimes try to remind them of their many victories with that amiibo, but they’d point to another amiibo of the same character who once had a high placement in a larger tournament or used to have a higher Amiibots rating and suggest I ask that person instead. I observed this pattern across most experienced trainers, all of whom felt like they didn’t have the best amiibo no matter their achievements.
I think those trainers’ feelings were flatly wrong, and that the community ought to establish a consensus on how to determine what a “better” amiibo actually is. This post is my best attempt at doing so.
Better amiibo
There’s a few factual premises I’m working off of in coming to the conclusions in this article. New trainers sometimes doubt one or more of these, but I assure you that they’re all tried and true premises.
- You can’t choose your amiibo opponents in existing tournament formats. If you’re entering an amiibo tournament/Amiibots, you don’t know whether you’ll face Mario or Marth in the next match, and it’s too late to change how you trained it.
- High tiers by definition have fewer bad matchups against low tier opponents. A high tier that develops a lot of bad matchups against its subordinates will rarely remain in its position for long.
- RNG (“Random Number Generation”) is a major influence on every amiibo match because it determines most of the amiibo’s decision-making.
Your Opponents
Consider the implications of #1. Knowing that #1 is true, we, the trainer, have to be ready for anything when training an amiibo for tournament, because we can’t adjust our training strategies after starting the tournament. Let me give you an example.
Many new trainers ask why Incineroar is so broken, and why doesn’t anyone train amiibo to specifically counter Incineroar’s Alolan Whip. Experienced trainers often respond with a story about how they in fact have an amiibo designed to counter that, but it plays so poorly against non-Incineroar opponents that it would be eliminated in every bracket before it ever went up against an Incineroar in the later rounds of the tournament. Basically, there’s plenty of specific ways to beat Incineroar, but the strategy that will beat Incineroar doesn’t beat anything else. Amiibo that are trained to beat specific bad matchups will not have good matchups against the other competitors.
Therefore, amiibo shouldn’t be trained to counter certain opponents, they should be trained to maximize their performance (win rate) across the set of all possible opponents, with a large set of opponents. That is to say, the best amiibo are the ones that perform as well as possible against a very wide variety of amiibo opponents from many different amiibo trainers. Strategies that are made to work against generally all opponents work best.
Character Matchups
#2 kinda screws with that. It’s true that high tiers don’t have a lot of bad matchups. However, they do have bad matchups. Amiibots has started tracking this kind of data as well at the Amiibots statistics tool, which you should take a gander at. I’d show you, but frankly I can’t figure out how to work this thing.
Bad matchups are important because they tend to exist no matter how the individual amiibo are trained. Suppose King K. Rool (for purposes of this post) beats Link 80% of the time in tournaments. Let’s also reasonably assume that both the K. Rools and the Links are trained by competent trainers and that the different Links and K. Rools have different playstyles to some degree.
Even when multiple trainers are submitting multiple variations of the same character, that 80-20 matchup will still persist across King K. Rool – Link matches. It may be 75-25 or 90-10 for some specific pairings of amiibo, but generally you’ll see something like the negative matchup persisting across all King K. Rools and Links. This is because the matchup arises out of something more than just a K. Rool playstyle being great at beating Link. It would arise because one character has a mechanic, ability or AI that the other character simply can’t overcome, given the limitations of amiibo AI.
Mii Gunner is a good example of this. Mii Gunner amiibo used to employ a strategy where they would only camp on one end of the stage and spam their “Miissile Gunner” missiles, and knock away the opponent when they got close. The opposing amiibo had no real recourse against this given the limits of amiibo AI’s ability to recognize the opponent’s strategy. If your amiibo didn’t have a reflector move that it could intelligently use, a Mii Gunner opponent was a highly unfavorable matchup no matter how you trained your amiibo.
If an amiibo trainer is taking bad matchups into account, he has to train his amiibo to maximize its performance across the set of all possible opponents but also understand that sometimes you’re just going to get crushed because of the other guy’s character pick, and that doesn’t make your amiibo “worse”. It’s just the luck of the bracket.
RNG
#3 is the monkey wrench that gets thrown into the gears of the amiibo meta. If we were in a scenario where RNG didn’t impact amiibo gameplay (meaning that you could reliably get exactly the same match by starting with exactly the same conditions), then you could just train an amiibo, put it against a very, very broad gauntlet of a large number of competently-trained, diversely-trained amiibo opponents, and draw conclusions about its skill based on its performance in that gauntlet. Then you could make adjustments, run your amiibo through the gauntlet again, and see whether it performs better than it did last time. That’d give you at least some idea of whether your amiibo was “better” than it was, because you’d have a stable set of opponents to mark your amiibo’s improvement against.
But RNG is still a thing. So if you were to try that gauntlet a second time with exactly the same conditions, your amiibo would perform differently than it did last time because the decisions that each amiibo made would have different randomly-chosen outcomes. You might train an amiibo that performs very well against the gauntlet in its first attempt! But one could reasonably argue that its quality performance was only due to good RNG, and it could do poorly in its second attempt to validate that argument. Therefore, RNG makes single instances of success or failure irrelevant.
So how do we demonstrate that an amiibo is better or worse?
Drawing Logical Conclusions
Well, we first have to accept that it’s at least somewhat dependent on the opponents the amiibo will face. If your amiibo beats every amiibo opponent it’s ever faced, then for all intents and purposes it’s the “best” amiibo around for lack of competition. But that’s really just a false positive, because the most achieved amiibo of all time still lose at least sometimes, so it’s more reasonable to conclude your amiibo just isn’t getting tough opponents.
We can curb those false positives by making sure that our amiibo is facing a large number of amiibo opponents from a diverse set of trainers. When each trainer trains differently, that creates as tough of a set of opponents as possible to measure our amiibo against, and the best amiibo would have to win most often across all those opponents. It’s sort of like how you can be the best wrestler in your state, but you’re not the best wrestler in the country until you’ve faced all the best wrestlers in the other states.
As a result, the best amiibo needs to maximize their performance across the set of all possible opponents, with a large set of opponents. So looking for the amiibo that statistically wins most often across a large set of competently-trained opponents is a good start.
Then we have to accept that the character of the amiibo has its bad character matchups, so comparing amiibo cross-character isn’t really valid due to having different bad matchups. If all Links share a set of bad character matchups but King K. Rool has a different set of bad character matchups, you can’t fairly say that your King K. Rool is better than my Link. It’s comparing apples to oranges. So getting crushed because of the other guy’s character pick doesn’t make your amiibo “worse” compared to the other guy’s character pick. It’s a statistical fact of life for all trainers of that character, so we should only compare amiibo of the same character to determine who’s the best relative to the other amiibo of that character.
Finally, due to the presence of RNG, we should be careful about drawing conclusions about our amiibo based on small or single circumstances. It may have just been good RNG, so RNG makes single instances of success or failure irrelevant. To draw as reasonable a conclusion as possible, we need to put our amiibo in a lot of fights against a lot of opponents to determine where to rank its overall winrate relative to other characters. In other words, we’re using the Law of Averages to remove RNG from the equation as best we can, and gauge the amiibo’s performance from there.
How do we do that?
Amiibots has:
- Objective mathematical metrics to rank amiibo relative to each other
- Visual proof of all matches so trainers may observe their amiibo and improve it
- A character-specific, ruleset-specific leaderboard to rank the best amiibo by character
- ~2500 matches running daily to amass as large a data set as possible (with each ruleset having its own 24/7 stream)
- A potato gambling game to fuel my gambling addiction
I don’t know of any better solutions to these issues, except for the gambling addiction. Amiibots doesn’t solve all our problems, as RNG is still present (but that’s somewhat curtailed by the objective metrics if you let it do a handful of games) and it relies on trainers to submit their amiibo to Amiibots in the first place, but I earnestly can’t imagine a better solution for getting this kind of data short of hiring full-time amiibo tournament organizers. We can’t beat RNG, and if a trainer wants to argue that his amiibo that’s not on Amiibots, or hasn’t been on Amiibots in a while, is better than the ones currently on Amiibots, then that argument should be ignored. That’s akin to saying “my amiibo is better but you can’t see it”, and that argument doesn’t fly.
What about traditional Amiibo Tournaments?
Well, I’m not really sure what to make of them. Historically, bracketed amiibo tournaments were the only form of amiibo competition so deciding the relative value of a tournament placement was confusing and subjective. Some trainers would consider a 16th place result in a 1,024 amiibo tournament to be more impressive than 1st place in a 256 amiibo tournament, and some wouldn’t. Some trainers would look at the characters that amiibo beat, or who trained the amiibo that lost along the way. A few trainers would sit off in the corner, not entering tournaments at all, and crankily complain that their amiibo were the best ever and everyone else is an idiot, but those guys are mostly found in my Youtube comments section.
I don’t see a consistent way to weight the importance of tournament placements, due in no small part to the incredibly wide variety of amiibo tournament rulesets. Amiibo tournaments vary in the number of participants entered, the individuals who enter them, the characters permitted to enter, the stages permitted in the tournament and any other such factor that can be altered by the tournament organizer. This is also why the community has never had a consistent way to measure the importance of a tournament placement in deciding whether an amiibo is “good” or not. There’s no consistency in the first place. If I were to only consider tournament placements for tournaments that legalized every amiibo except the universally-banned Incineroar, then I would only be able to determine the skill of top tier amiibo, because all the mid and low tier amiibo would be crushed in bracket. If I were to only consider amiibo tournaments of a certain size, I’d still have no way to draw equality between placements in tournaments that legalize top tier amiibo versus placements in tournaments that don’t. There’s no functional funnel for this kind of thing, and that’s before addressing the issues relating to RNG and performance.
So unfortunately, I don’t have an answer for you on how to weigh tournaments. If you win a tournament, that’s awesome! Your amiibo probably has some hutzpah to it. But I can’t tell you much more than that.
What do you think?
Amiibo Doctor 