Howdy

My name is PaperLuigi3; I'm somewhat active on the amiibo subreddit. I've been applying scientific method to amiibo for about the last eighteen months. I solidly believe that the only way forward for amiibo training is through vanilla tournaments (no custom moves, no equipment, no stat boosts); after all, the customs metagame quickly died off once people realized that Little Mac with +200 attack is unstoppable regardless of his powerups. To this end, I've been compiling data and performing experiments as well as constructing an ebook on "amiibo theory" - the study of how amiibo operate and what affects their operations.


I've probably spent more time on amiibo than most of the people on this forum save for darling Cloud himself; by my calculations, I've performed over 5,700 individual battles on both the 3DS and Wii U systems. My amiibo resume is... extensive.

Ask me anything.

First of all, welcome! I really have no life at all to be on here this late. What you have to say does seem impressive, but how come you never picked up tournaments themselves? We've banned Little Mac, so we still have a customs meta. 

Hey! Welcome to the forums! I hope we'll get to know more info (like how the Final 3's A.I. will work (research from cpus))!
@Blakers Ya think you don't have a life? I'm posting this at Midnight (r.i.p. my sleep...eh whatever I can handle going sleep-less!)

Unfortunately there's really no accurate way to predict the AI of the final three, even using CPU opponents. See, in Brawl and Smash 4, CPU opponents actually shift over time based on the moves that players using that character have previously used. So everyone has different CPU opponents.

On top of that, there's an incredible depth to the learning process of amiibo that just isn't there in CPU opponents. Some could say that a hand-trained amiibo is entirely different from its CPU counterpart.

The only thing I'm confident in predicting is that Corrin will very likely have trouble using his/her side B. In practice, amiibo tend to have trouble using multi-part moves, take for example ZSS or Sheik. Both have difficulty using their flip jumps

Agreed. Some amiibo have two-part moves programmed into them, such as Ness' and Lucas' Up-B, and Mario's FLUDD (if you train him properly). As to whether Corrin will be able to pull that off, it's a different question entirely.

The real question is if Bayonetta will be able to do her normal combo - the Side B to Up B to Side B to Side B to Up B - you know the one. If she can pull it off successfully for at least two hits, then that would put her in a dominating position against at the very least Bowser, Ganondorf and to a lesser extent Little Mac. Little Mac, with or without CHC and EPS, will still crush Bayonetta so long as he uses a simple counter against her smash attacks, but against most of the other cast members, she'll do fine.

Looking at this discussion, it's time for some shameless promotion! Check out my article on the Final 3 if you haven't! https://amiibodojo.com/2017/04/14/cloud-corrin-and-bayonetta-where-you-can-get-them-and-training-speculations/

I've read it, and I don't see much of a problem with your predictions, inasmuch you didn't try to use some kind of quasiobjective standard to make your claims. You just said "This is what I think, and this is kinda sorta why", and I enjoyed that. Too many people are trying to find some way to predict the general behavior of these last three amiibo, but that's fundamentally impossible to do: amiibo, although they have the same bone, their flesh develops differently. It actually has a "range", so to speak, and if you were to chart out these ranges (after applying a mathematical standard derived from the variables obtained through experience - another algorithm I'm working on!) it would look like a normal distribution. Most amiibo would be less than one standard deviation from the norm; they would be somewhat better or somewhat worse than the average amiibo, but not by much. The tournament winning amiibo would all be at least 1.1 standard deviations towards the positive, and the exceptions would be at least 1.1 standard deviations to the negative.

This is why predicting amiibo doesn't work: if you buy Cloud and train him, you have influenced his standing on that normal distribution. You may be a phenomenal trainer and have pushed him three standard deviations to the positive, thus putting him in the top 0.15% of Cloud amiibo (this is, of course, assuming that all Cloud amiibo in the world are equipped with the exact same custom moves and equipment - which they won't, which is another reason I disagree with non-vanilla amiibo training). So if you are a phenomenal trainer, then you have a very inaccurate picture of what the Cloud amiibo's behavior is.

So if you then decide you want to get a picture of how an average Cloud amiibo will operate, you'll need to hack a significant number of pre-trained Cloud amiibo (again, assuming every single Cloud amiibo has the exact same equipment so the numbers are objective), plug in the variable values obtained from each amiibo through the algorithm (that I'm still constructing and working out the kinks on) and chart that on a statistical normal distribution. 

Once you do that and have a complete data set, only then can you see if your Cloud amiibo matches what the picture of Cloud actually is: by hacking into the 504 KB on his chip and plugging them in, you can see if it is accurate or not. And if it's not exactly in the middle, on the top of that hill of the normal distribution, then your experience with a Cloud amiibo is inaccurate because it's not exactly the average so you have experienced an inaccurate picture of Cloud.

But Cloud's not even out yet and that's all a bunch of work and I don't have the money nor the copious amounts of time to do that and post the results, so nobody should claim to have a standard to predict what a future amiibo fighter will act like. Because of that, thank you for not claiming that.