Just a few notes based on some recurring comments I've seen:

I think a lot of people are misinterpreting how we got movesets, saying things like "x pokemon was missing x move". Every combination of moves with at least 1 damaging move available for each pokemon were tested at each trainer. We were not trying to find the BEST moveset, we were only determining if it could sweep. As mentioned in the video, if a non perfect moveset is tested and it sweeps 3 times in a row, it moves on from that pokemon because obviously a better moveset will still sweep. If a "worse" moveset is tested and it can take out 4/5 pokemon, then another "better" moveset still can only take out 4/5 pokemon, it still keeps the first moveset to get that high score. It generally simply means "better" movesets didn't actually perform any better in that fight. So when you see pokemon with "weird" movesets, this is why. Remember only the top score moveset is kept, EVERY moveset was tested unless a pokemon got a perfect score for some moveset in which case the mon gets a perfect score and the rest are skipped.

I need to acknowledge that the AI is not perfect, we're still finding bugs and there's very likely situations I haven't considered when coding it where it just does not know the correct thing to do. If you've tried writing AI for pokemon yourself, you probably know it isn't an easy task, and requires many many iterations to get right. Currently known bugs include confusion moves being spammed, underestimating double/triple hit moves and dream eater looking at the wrong pokemon for sleep checks. Dragonair was also accidently given access at misty rather than surge and venusaur was accidently given earthquake, slightly boosting their rankings. If you find any more situations like this, please message us in the discord server, as this is where I'm most likely to see it :)

The trainer’s AI is improved from the base game, and is the same as the player AI. This leads to results like charmander being unable to sweep brock. I can see this being a controversial decision but think about it this way, it takes the randomness down dramatically because the AI is not using completely random moves. This makes it feasible to actually run the simulations, because we require significantly less repeat battles. I'd argue without running thousands of battles with a purely random AI you cannot get scientifically accurate results, and this is computationally infeasable. Also generally, if charmander cannot beat brock with decent AI and another pokemon can, of course the one that can is clearly the better pokemon. It is escentially taking into account the worst case scenareo for the original random gen 1 AI, so if they can beat good AI it shows they are consistent. It is even across all pokemon, so the ranking holds.

Our ranking does not take into account grinding time. This is a very fair point, while dragonite can sweep all of blue’s teams, it’s very difficult to get it to a level on par. Our ranking goes by level cap with no stat exp, so in theory dragonite at level cap is more equivalent to a dragonite 10 levels lower with stat exp. This means the ranking is still somewhat accurate, although the most rigorous way to do it would probably be to calculate the total exp you get for each trainer in the game after you catch it and figure out what level you would be. I hope people can understand that this would be extremely time consuming and the process for this video was extremely long already, hence why we did not go down this route. The way we modelled it is as if someone is using infinite rare candies, which I think still has value to a lot of the community.

Blue’s team is split into three and each one is weighted equally. I agree this is an oversight and does weight pokemon slightly to how well they perform in the champion fight. In future we will make sure to normalize fights like this down to one score.

Our ranking does not take into account valuable TMs being used up on the pokemon that maybe could have been used elsewhere. This may be something we can improve on in the future, but I'm not 100% certain on how we'd go about modelling this. If anyone with some math/stats knowledge has any suggestions for how this may be modelled please post them in the discord server!

A bug has also come to my attention where all mons on a trainer's team were set to level cap. Since this is even across all fights it should not effect the results majorly, however it has now been fixed for whenever we do this again.

Also just a note that the code, full results and a more detailed description of the process can be found linked in the description for those interested :)

- Craig/cRz Shadows (Programmer)

What about a calculation of "if every pokemon could be your starter pokemon" ?

Jrose11 has entered the chat

From the charts, it looks like all three Blue variants are scored fully... so this tells us which Pokemon would do best at beating the gyms, the Elite Four, and then Blue THREE TIMES IN A ROW. In other words, it's skewed to pokemon who can beat Pidgeot (x3) / Alakazam (x3) / Rhydon (x3) / Executor (x2) / Gyrados (x2) / Arcanine (x2) at level 61-65. It would be interesting to see how the rankings changed if each Blue encounter was weighted by 1/3.

I'm a little confused as to why some Pokemon are listed in their pre-evolved forms (Ponyta, Caterpie, Pikachu) while some are listed as evolved (Weezing, Muk, Beedrill)

MY BOY SANDSHREW I AM SO PROUD OF MY BOY

"We see a big jump for Dragonite and a stellar performance for Articuno"
Articuno: went up 50 places, yet the dragon is considered having the 'big jump'

3:28 Basically, the argument to put Ponyta much earlier, maybe near Bill(25 route).

Of course a Water and/or Psychic Type would DOMINATE Gen 1 Kanto. 🐐

Fascinating. One flaw with this metric that immediately leapt to my head is, it doesn't account for the practicality of being able to get each tested moveset for a Pokémon in the early game. Some would require an unrealistic amount of grinding, and some would be flat-out impossible to obtain because they require TMs not available until the late game. Still, this seems as good a method for evaluating Pokémon usefulness as I've ever seen someone come up with.

articuno, zapdos, we need a third legendary bird

best i can do is a diet charizard 🤷🏻‍♂️

i love the jjk references😭

Screw Pokémon master, this is a Pokémon PHD thesis

Make Trasform a priority move.

#justiceforditto lol

The random mix of whether you use the name of the first form or last form and the pictures not matching is driving me crazy! But I appreciate the effort you put into this

Just because of this video, I started Pokemon Red on my old Gameboy Color again today, paid the 500 bucks for Magikarp and now, just before Misty, have my own Gyarados. I'm 32 and I'm over the moon.

I remember buying the Magikarp outside Mt Moon in my first playthrough. I was confused as Splash did nothing, but anyway tried to level it up just to see if it learned something useful. I was glad when it learned Tackle, only to be dissapointed again when it did almost no damage either. But I kept trying, and when evolved I was shocked by how much the same Tackle and now Bite was doing. Ahhh the good old times, of course it ended on my very first hall of fame 😁

My friggin head tryna deal with the switching names/forms being called out @_@

Great content but that definitely did my head in

4:50 as a kid I routinely fought Erika as the 7th gym leader because I choose squirtle as my starter and Blaine was far easier, and I would rush to Fuscha to get Surf asap.

Worth noting that Venasaur did so bad because the sim didnt know about the toxic/leech seed glitch

"Imagine your 8-year old self leaving on their first Pokemon journey." Actually, I was four. They let a four-year old go on a Pokemon journey unattended aside from my Chikorita.