Obviously the correct way to find what the best card is to see what percentage of your winning runs contain a particular card. Ascenders Bane is the best card in slay the spire

Computer, what is the best card pick? SKIP No, I mean actual cards. DEFEND

So what we can extrapolate from this data is that callipers could be useful here

Boss relic idea: Gain one energy, you can no longer skip card rewards.

New player: what are the best cards? Jorbs: do you have 100 hours?

Content like this is what makes you the best StS contest creator in the world. You're one of the best players, but this high-effort analysis here is so much better than what anyone else in the community puts forth. Exceptionally great. Thanks!

This dude must be the worst person to ask 'where do you want to eat?'

Jorbs, you gotta do a few runs where you have to choose cards solely by their ELO rating and see how far you can go.

I expected this to be 50 minutes of Jorbs saying 'Its wraith form' Im dissapointed

Viewer: "Is X a good card?" Jorbs: "No, but yes, but no."

Northenlion: I rate cards from 1 to 5 *Jorbs laughs in spread sheets*

One thing this analysis really sheds a light on is what cards are "conditional good". Cards with abysmal ELO and pick rates but around 60-70% win rates. For a beginner, or just an enthusiast, to try to figure out what else must have been going on in the run for those cards to be so complimentary is a great way to understand the synergies in the game. I know, there is no such thing as a synergy though :)

"I'm salivating because these are all good cards." - Jorbs 2019.

Great video as always. The point that having access to a lot of data doesn’t necessarily mean we are drawing useful conclusions is an important thing to understand in the Information Age. That’s what makes this game so great to me... you can’t just autopilot through and expect great success. You have to thoughtfully engage with the game. If you didn’t have to do that every time...eventually the game would get old.

No wonder singing bowl is rated so high with Silent, what else are you supposed to do with all those cards that don't have intangible, retain, poison or lose strength written on them ?

The rain sounds /white noise is soothing lol, I fell asleep multiple times while watching because I haven't slept much last few days lol. I did finish the video same day though! Very interesting data!

2:58 HE DID IT HE SAID THE WORD GRAB THE STICKS

Really appreciate the in depth look at skepticism of stats, looking at context etc as ways of dealing with bias and why things are performing etc. It's something that has absolutely driven me crazy about mtg stat usage, I will probably link this to some of those people on how to look at groups of statistics and what they mean. And like in any balanced deck building game of course it's "it depends" and context needed outside pick 1. And in slay the spire with the whale you don't even get that. Great job making a seemingly clickbait title into a in depth discussion on the how and why of stats.

Is it just me, or did anyone else value Jorb’s pick percentage far more than any other metric?

Wow, Jorbs has a sub 40% win rate! That makes me feel much better about myself! Jorbs- "My win rate is clearly much higher when i'm trying my hardest" ... Cool.

I lack the words required to express how much better you went about this than those reductive and innumerable """tier lists"""

Really nice to listen to jorbs with the rain in the background

One metric that I think is important for this analysis is how many times you've already picked that card in a particular run. I would think that maybe the second Noxious Fumes, for example, would not be as strong as the first Noxious Fumes. Probably the first two or three Piercing Wails are very strong, but the fourth or fifth would probably be significantly weaker, and so by grouping the fifth Piercing Wail with the first, this data might be implying that Piercing Wail is weaker than it is.

Great guide to metrics! As someone who's not usually into statistics and elo and whatnot it was nice having them all lined up and explained so thoroughly. You really did a good job making this video accessible. The ability to upgrade cards at campfires does screw around with the pick rates for upgraded versus unupgraded cards a bit though, doesn't it? Especially for cards you don't want many of that you find early on in their unupgraded forms.

KZfaq exclusive content... truly, a gentleman and a scholar. Thanks for all the content, Join Ribs!

It would be interesting to see ELO for each act. I would assume that dagger throw would have a higher act 1 ELO then piercing wail, but that would change by act 3

This is a public service announcement: Elo is the last name of Arpad Elo, who devised the rating system used for chess. ELO is Electric Light Orchestra, a rock band who are not involved in rating systems for games.

@22:50 While you were counting down the 3 ways a card can end up at 100% win rate, I was thinking exactly that: There's the fourth possibility that a card is only good (and thus picked) in certain deck "archetypes". You could already tell just by looking at the stats for Heavy Blade early on in the video. I see you mostly pick it when you have strong strength scaling unlocked. And since strength scaling, at least the way you play it, is a strong archetype with a high win percentage, it makes sense that Heavy Blade is a really strong card... in such a deck. Most of the time it's either very good OR pretty bad.

its funny how you can do all these calculations and data charts and in your head its already all calculated in one go and gives a simple feeling like good or bad

"We"ll take thirty minutes..." Look at the lenght of the video : riiiiiiiiiiiiiiiiiiight... =D Love your content, Jorbs! Cheers!

Watch at 1.5x speed, you can go higher if you’re comfortable

The rain sound in the background is so soothing lol

Skip is obviously the best card, pick it every time you see it.

Very important stat that didn’t seem represented here is % of winning decks that contain card. A look at that combined with the card’s win% in some way would seem very valuable. Would get rid of the situational picks, and would emphasize cards that are ‘required’ to be successful.

I love when Jorbs makes these types of videos!

Omg @ 40:30 I just noticed the cat! Its so adorable. Cat also wants to learn and improve at Slay The Spire

Would definitely be interesting to distinguish your first pick from your second, third, or beyond. Piercing Wail is a common, so it'll show up a lot. I bet you take one Piercing Wail more often than you take three.

Great stuff! Thanks. Having watched this, I've realized I'd really like it if you've created a separate KZfaq playlist for the runs that you've tried hard in and really concentrated on winning (no matter the outcome). That's the most interesting content for me. Like the time when you were going for a A20 streak.

This should be done annually.

I'd like to point out that choosing to not pick Offering+ has a 100% winrate as well. The only thing we can understand here is that Offering+ is a game-winning blessing whether we use it or not.

What I take away from Jorbs video, and might be a bit of my confirmation bias but, in games or in life, it matter little what the big shinny stuff is being offer. They are good and helps a lot, but it is the small things you do everyday that does much more for you overall. And I think that it is neat that it is somewhat proven by data in a game.

I think something beneficial you could make would be a list of cards that are synergistic together. Ones that when building your deck you prioritise if you have another card in its synergy group. Also I'd love to hear about relics and the gold economy in the game. Thanks so much for making this video it was really really informative.

I think another way to determine which cards are "the best/worst in slay the spire" would be to look at different roles a card can fill and then within those categories how efficient and effective certain cards and/or groups of cards are at what they do.

Something I think you could try is using what cards you choose to upgrade at campfires as a way of comparing upgraded vs base level cards. The power jump between Defrag and Defrag+ vs the jump between Reaper and Reaper+ is very different, which is why I see you upgrade Reaper a lot less often compared to Defrag or Biased Cog.

I honestly think ALL the cards are trying their best!

Plus of course the issue that you play on A20 and in the Beta version of the game - I suspect most people asking simply "what is the best card" play on A0 or certainly less than A5 on the 'normal' version. Different cards will be "better" or "worse" as you're literally playing a different ruleset.

This was a super pleasant video to draw to. Thanks, Jorbs!

Jorbs, thanks for this really interesting and insightful video. I think what’s missing, though, is a semi-analytical approach instead of a purely datat-driven one. You already know about synergies like Panacea + Wraith Form or Spot Weakness + Limit Break; I think a better analysis would be evaporated eluating a lot of those hype interactions based on things like how likely they are to occur and how well they are expected to do against certain fights/overall. Cards that are strong enough to stand alone, like Wraith Form, can also be judged by the same metrics.

The lessons of this video are largely the same as for many other disciplines, including programming: 1. The answer to almost all general questions is “it depends” 2. Correlation is not causation

I have no idea if anyone will ever see this, but there's a bit of discrepancy in the SpireChat playlist: This is episode #58, and episodes 55-57 aren't in the playlist. (Those would be the Northernlion card rating video, the video on Busted Crown, and the video on Equity)

I just found his streams and been watching a couple of his youtube videos. Is he like a data/computer scientist? when talking about these kind of things he just seems so incredibly knowledgeable. is this just enthusiasm for things like data and AI or does he have a formal education in the field?

i hope his cat is thinking about how he's talking to no one

If you assume knowledge of StS, this would be a fantastic first lecture of an introductory statistics course

The biggest problem is the Relic synergy with cards picked......that is the biggest data that is hard to show here....

The other argument on a card is how many you already have. A card could be awesome, but if you already have 2, a third would be crazy.

12:00 Ironclad has Feed, Reaper, and likes to take Bites sometimes, so often has meta scaling, so it makes sense to me that he’d prioritise hallway fights over Silent who has no meta scaling.

I would love to see this again with 2020's data!

Me seeing this in recommended: "Yeah I guess I'll listen to another spireside chat on my way to work, I don't really remember this one?" Me seeing the upload date: 🤤🤤🤩🤩🤩🤩🤩🤤🤤🤤🤑😤🤤🤤🤤🤤😇😇😱😱😱🤓

We already have access to a trained neural network - his name is Jorbs

Would love one for relics too! amazing video, enjoyed it

This is so accurate it's surreal! Can you do a series on how to analyze game data from other game? Like how to build a deck in magic or which hero to pick in a moba style game using data.

I think also keep in mind skip shows up every time so like 40 ish times per run on average, you may not see a rare one time and even uncommons. So while it shows the importance of skip it is hardly indicitive of being the best choice. And it's pick rate is low relatively to other strong cards(~20%vs 50plus)

An interesting idea would be to play a game purely picking cards based on the Elo values, showing that the context of the deck is what really determines a cards value

Thanks alot for this one.I love Content like that.

Yet another comment by me, but this is a separate albeit correlated topic: does the ELO rating take in account if the run you took the decision in was won or not? I'm assuming the latter and that means that rating is a very good answer to the question "what do you think is the best card in the game?", but for the more general question of best card in the game or in general to get some interesting rankings on them should take into account that maybe you don't play optimally, and the only way I see this done is giving less value of the victories in the card being picked if the run was a losing one, since there's an higher chance that it was a mistake and should have been valued as a loss(not that we can prove that it was a mistake). I feel like another effect of such a system where cards you choose in losing runs would be valued less is that cards chosen because they were the "least worst" option would be valued less, so cards that have just been chosen because they were okish in an unlucky run where your deck was worse than okish would weight less, but I guess that's already dealt with in a way by the elo's system of increasing less if opponents were rated lower.

So basically it's 42. Great video.

Lesson, SKIP and CLAW are both OP

I think win percentage when offered is a little more accurate than win percentage when picked. It still has some problems like flash of steel only showing up in act 3 or if you play poorly, but I think it’s a decent enough measure overall.

Zephyr waits for claw to be integrated into this spreadsheet, because it is clearly superior to reaper in all measures and has a win percentage of 101.

Next time you do this I think it should be split by classes, it would make it cleaner to look at and account for things like your lower defect winrate.

I'm actually really surprised by how low the ELO is on Flex. It seems like a very straightforwardly good card, especially if your deck is small.

Elo rating surprised me as a metric but I feel it is by far the most accurate one

You should make one of these for the most difficult enemy in the game

"We're gonna take 30 minutes and..." *looks at 50 minute video duration* Riiiiiight...

My take away from this is that Singing Bowl is the best card. Kappa

Jorbs can you thank Zephyr for his insights for me

The stats grad vibes are strong here 🦾

In basketball they use "score differential per second the player is on the floor". If you calculated the per-fight average result (win, loss, including damage) of each card on each floor, you'd have a natural way to compare cards. i.e. decks with card X win floor 10 fights at N%; card Y on the same floor wins at M%. This still conflates mixed strategies but at least would make individual cards more comparable.

This is my favorite video of yours. Would you consider a follow-up that uses data to analyze the equally-complicated question of which upgrade is most valuable in StS? Oh and for what it's worth, when people ask me what card is best I usually mentally frame it as 'if all cards were offered, which would I pick on floor one?' Then they get the troll answer of Omega :)

I'd be super interested to see an n-grams approach to this data, where pairs of cards are given the same treatment as these are!

I hope Jorbs returns to this topic in December, with only runs from December 16th of last year through whenever he does the video, and compares numbers between 2019 and 2020 (with a big asterisk on Watcher 2019) before giving an aggregate

A lot of thoughtful analysis. Judging from Jorbs' comments, it seems like the best way to look at these data is to find out how this or that card compares across metrics. Also, a little disappointed that the table doesn't split the cards by category. Jorbs has mentioned before that one should pick cards that improve your chance to win in various scenarios. I'm considering downloading the data to split cards by whether they're damage cards, defense/mitigation cards, or utility cards. This approach still has a huge weakness in that some cards get better if you have other cards to go with them, but I still think it could be useful. Is crippling poison the same as deadly poison? Steam is the same as steam barrier, right? What is undo?

Oh! He said it! MORE CLAW

you should collect data on how often you play a card and how often you win fights while playing that card

Defining 'best' by Act probably would help here like mentioned. Or even by floor.

Wow, Spire Chat Poj.

For a more on topic comment, shouldn't win rate for this table have been divided by the class's win rate for this to be a fair comparison between classes? I guess if the point was to find the best card in general this is accurate, but in we wanted to branch in(as we did) the relative value of cards and find a ranking for how often you should/do pick it when shown the relative effect on winrate should be more important than the absolute effect. For an example let's say you had an 80% winrate on watcher, in that case no matter the specific card in the win%(picked) column watcher cards in general will show up much more often than other classes', and we show at the start how there's a relatively relevant gap in the winrate of classes. EDIT: oh you talked about that point right after I paused to make this comment, welp. Still it was in favor of the ELO metric, which is fine but I feel like the suggestion I made for the other column is still relevant.

Incredibly useful and interesting video. Love it

biased cognition + has to be best card all around...I know many would argue that adrenaline is the best,but the thing is,adrenaline doesnt win games by itself,biased cognition does..Reaper is close,really close to being the best,but it can be a dead card unlike biased cognition where you almost always want to pop it

I wonder if wraith form will make the list

i thought all your content was youtube exclusive, youtube jorbs?...

I feel like combining unupgraded with upgraded and look at times it was upgraded with picked upgraded as a sub category on a card. As where upgrades go seems about as relevant.

Reminds me of: "You're both pieces of shit, I can prove it mathematically! Wait, this has been a long time coming anyway, let me grab my whiteboard..." Only difference being that Jorbs is a lot more composed than Rick usually is :D

Wonder what would Alpha AI from Google come up with if left to play this game few billion runs... That would be pretty fascinating to see what proves statistically significant in the very long run and somewhat easier to analyze post-factum then a game of Go or Chess where half of the 'why Alpha did something' is guesswork.

My sub box was hiding this from me. Rude. jorbsGlare Nice Christmas gift btw.

Ogbomosho to xXMomentumXx as well.

please bring this series back bro

you should calculate how many battles you won with the cards, and then weight the results along a curve to account for you winning every battle in a run except the last. Also calculate which floor and act you picked the cards. So the difference between picking a card act 1 and act 3 is displayed,

It’d be really interesting to see what cards were picked first in the run, which ones were skipped, etc

Isn't it the fact that there is no 'best card' because Combinations matter the most. So you might have a card that makes a 'not the best' card better than the 'best card'? I mean if you were given a 5 Card 'hand' to try to create 'perfection' for a Class, you could. But if you missed on just ONE of those cards in your hand, another combination of 4 Cards could be vastly superior.