I love how humanity went from pythagoras creating theorems for understanding triangles to young adults figuring out the statistical significance of an rng manipulation theory for a game.
To be fair, they probably did the same thing back then but we just don't hear about it. Just like how future generations won't hear about this, but instead will know all of the things we made that had lasting significance.
I kwow that this is a joke, but what do you think scientists and mathematicians are doing in our generation? Things haven’t really changed, it’s just that Science and Mathematics are more accessible to the general public and therefore are used for trivial tasks like this one.
Actually humanity hasn't changed at all. In despair they pray to a higher being/system for salvation. And that's also what happens here. People are frustrated by RNG and then a prophet appeared and promised them salvation as long as they believe. Rest is confirmation bias.
bet you havent encountered people calculating breast size of gacha waifus based on number of pixels or successfully identifying/tracking a criminal/assaulter based on cctv footage of his shoes or tracking down someone with an arrest warrant based on pizza boxes seen in his Twitter post while arguing with a grown child
Application security expert here who has some experience exploiting PRNG mechanisms. The thing that bothers me about this theory is that it relies on some weird implementation choices on behalf of Mihoyo. It implies that the state of the RNG generator only changes by user action, which would mean that either: A) Mihoyo is querying RNG client-side for progression. or B) Each individual player has their own RNG state serverside. RNG manips would be almost as bad for a Gacha game as it would be for an online casino, and this isn't the publisher's first game in the space. I have to imagine they at least know the basics of preventing RNG attacks. With that in mind, B would be a wild choice. You would have to do it specifically, and it only opens you up to RNG manipulation attacks. A would be even more insane and would essentially mean that a skilled attacker could control the RNG themselves. The fact that there isn't (to my knowledge) a thriving underground market for RNG hacks suggests that this isn't the case. (Yes- there is anti-cheat, but there's no such thing as foolproof anti-cheat). And also, just several of the techniques and data here just don't line up with how I would expect a likely RNG attack for this system to work. I can theoretically explain multiple parts of it individually, but holistically its difficult to explain what might be happening in code. Of course, just like the statistical analysis, this doesn't conclusively prove anything. RNG manipulations are some of the most inscrutable and arcane attacks out there, and even when they work, usually aren't 100% consistent or reproducible. They're some of the most fascinating attacks in speed-running, and if you want an example of an amazing one, look up "How We Solved the Worst Minigame in Zelda's History" by Linkus7.
From a programming perspective, I definitely agree. Speedruns have some really interesting manipulation tech, I remember for pokemon emerald, you could initiate the first pokemon battle at a specific frame to get really high IVs on your starter due to the way PRNG reset in that game. I don't really get why the study decided that blocking into hours/days made it easier to analyze because even if the timestamp is used as part of the seed, you would use something like a UNIX timestamp. To be fair, there are some weird techniques out there for PRNG manipulation, including the Linkus video you mentioned.
I had a wild thought a while back of trying to get a machine learning algorithm try to guess what I would get from rolls from gacha. I'd be betting that they use PRNG and not some sort of physics based RNG mechanism, and the limited info available to the client would be enough to reverse engineer it. Do you think such a technique would be feasible?
@@TarZBlammo yeah I also think it's not an easily doable approach, I'd need to get a few thousand people to help me with the dataset, at which point it would be hard not making HoYo aware of such an endeavour. Then they could just slightly change the seeding algorithm and it'd be all wasted effort.
@@davidebic technically speaking even physics based RNG isn't really a thing on a macroscopic level since on our level everything is deterministic, on scales that quantum effects are significant then you get true and unpredictable randomness with a known distribution
It's so nice to see real statistics applied to something like this, loved the video! I think it would actually be very interesting to run a study on substats taking into consideration all the issues you mentioned. I doubt it would be possible though, since you need very high sample sizes and there is little incentive for players to submit stats if they can't see the results until the study is complete. Have you heard about Star Rail's 50/50 being closer to a 56/44? Some people are saying that the sample size is too small to reject the null, and I'd love to hear your take on that.
Thanks for the compliment! As for the substat rolls themselves, I think there would be way too many issues, especially given the necessary data collection involved. As for the 56/44, I don't think the sample size is an issue given the 15 mil+ data points that the study claims. If anything, the bigger "issue" is that people that post their pulls may want to brag about their luck, biasing the sample group. Reading more into it, the creator did reduce this potential source of bias as much as possible and did multiple statistical tests. On top of this, I am more inclined to believe the rate is actually 56/44, given that there is a legitimate programming reason why these particular percentages can appear (banner character being included in the pool of standard characters when losing). Generally when significant bias happens from a RNG algorithm, it is mainly due to programming error rather than a fault of the algorithm itself (unless you pull a no-man sky and use a really crappy one).
@TarZBlammo Yeah, like he said. I heard that after you lose your 50/50, you still have a 1/8 chance of getting the limited 5 star, and a 7/8 chance of getting one of the other 7 standard 5 stars.
I saw a study pointing out that although it’s closer to 56/44, you have a lower curve of getting pity in Star Rail, so the mathematical expectation of both games getting and winning the 50/50 is about the same.
I was really interested in learning how to do this method until I remembered that I'm a psychology student that was forced to study statistics Once I realized that the statistics samples were too low to prove anything and that there's a good amount of confirmation bias here, it made it clear to me that this is too much work for a low chance of it actually being legit
Is it really confirmation bias if there's real proof that it works? The specific channel on his discord is full of S+ relic rolls every single day, i have no idea why people are so antagonistic towards him when dude actually share how he got his relics to other players, at least try it yourself before dismissing the method. I got my Acheron to 80/200 crit ratio thanks to this so i'll keep abusing this until hoyo make changes lol
@@Amalia19 ah yes, because all the people who got dogshit relics are gonna excitedly post screenshots of their dogshit relics to contrast the S+ relic rolls lmao
@@LilacShowers Yes we also put our double crit relics that got ruined into the sheet, to make sure no one rolls on that specific time in the future since the string pattern in that specific time is shit I don't understand the agression, is it because you can't believe all that relic you got could be better? How many times do you even get a good relic, once a year? lol we're getting that in daily basis, it's your loss anyway so i'm not gonna bother trying to convince you anymore
@@Amalia19 that entire comment shows zero understanding of statistics lol even if the method entirely worked you'd still get counterexamples because it's still rng even if mitigated and also while i fully believe at least some of you do put your data that contradicts the trend into the sheet (albeit seemingly with the mistaken impression that it must mean those are "bad days" instead of just unlucky data lmao) we weren't even talking about the sheet, we were talking about the discord channel, where you were bragging about people posting lots of S+ relic rolls every day, because most people would not in fact go and post their bad or mediocre rolls in that channel lol (which was my point before) i don't think i was even particularly aggressive in my other comment; i mean, i could accept "dismissive", but pretty sure you're just the one being like super defensive, which... i don't really understand, because if your claim that this method really worked out for you is true then you wouldn't have much reason to be huffing copium about it and i'm not sure why you'd lie about that; maybe you've actually just gotten a few lucky relics in an ocean of shit that you've convinced yourself that nah the method works it's just that "that specific time is shit" so you're doubling down to deal with the cognitive dissonance that you seemingly just so happen to keep rolling relics at "shit times" lol anyway go off i guess ...come to think of it if you can just randomly get "shit times" how does this even help lmao like if the theory is true it sounds like it just adds another layer of uncontrollable rng
I saw one of those videos a couple months ago, and with how odd programming errors can appear in all kinds of places in games I was open to the possibility, then I saw the sample sizes for the substat combinations and the disclaimer that the patterns actually "weren't 100%" and noped out, these people are reading patterns into white noise.
Im on the dumb side of the bell curve that's also sure that there isn't a way to manipulate it cuz of one thing. There'd be a lot more perfect relics if it was a thing.
Either way, after seeing that video, I stopped dumping all my xp into a single relic at a time and instead upgrade 1 line for separate relics. I’ve definitely been getting better results.
@@ZeeZeeVee because doing that does make sense. If you have, say, 20 relics with 3 substats, choose the best one among them and roll it to +15, you could get lucky and get a really good piece, or roll on wasted substats. If you instead roll all pieces to +3, you'll have a smaller, better pool of new "promising artifacts". Since you now have 20 artifacts with 4 substats, you can discard the worst one, and then roll to +6 on the remaining ones. Doing this repeatedly basically culls the worst artifacts and leaves you with the best ones of each step, instead of going all in on a single piece (who could have been discarded when getting them all to +3 or +6 due to bad rolls). Doing it this way is more tedious and labour intensive, and most of all: an exercise of patience, since you don't roll the first good piece you see, instead wait until you have a bunch of promising pieces and then do a rolling session. You could have to wait more than a month to have enough pieces to "guarantee" at least a good one
I saw a post from someone who's EXTREMELY unlucky grinding for Trace Materials, they asked the customer service about is not getting any 4 star rarity is a bug. They replied with something like "I feel sorry for you but that is not a bug, just RNG" 😭 RNJesus really just does that
I was noticing some of the things you said in your video with what I was experiencing with my account. It's not like it's going to hurt to follow the relic manip guides cause there really is nothing to lose right - but I got some of my best artifacts just randomly upgrading to 15 more than I do with the guides. Like you said, I feel like the data amount is indeed low enough that you can never really tell. So really good video my guy! Like you emphasized, it's not really bad that people are studying this because after all - everybody wants them sweet builds. Again there is really nothing to lose when following the relic manip guide or not.
Feel the same, I think the more data they get its going to look like everything with 25%, making it just random so guess the best way is just keep farming and trying, also don't need perfect crazy build just to flex with every single character
@@asoret facts man... We don't really need the perfect 80/200 "good" looking builds to clear in the first place. Just keep farming if you're a beginner and the same with endgame players and you will get results whether or not you follow the relic manip guides. I feel the same way that it just looks like to me that if they took a million player's data (hypothetically) then it would in deed look like it's distributed equally
@@fooo0p Man I saw an Acheron who had like 95/260 Crit ratio on just combat start so it was a permanent thing not a temporary Bronya Ult or something like that, though 12% Crit Rate did come from Fu Xuan but that's something with a 100% uptime by default. It's been eating up the inside of my head making me farm better stuff for my 82/200 Acheron (82/240 once Sigonia planar set stacks up). I am blinded by jealousy everyday
@@KuroiShiAnimu damn you on sigonia set? unless is PF then it's not really optimal but holy you got that great build. I am on 46/220 but I do have e2 and izumo (that 12% atk is not to be underestimated - it's 4 substats worth and that 12% crit is permanent with at least one more nihility so it doesn't need stacking like sigonia). So without buffs it's going to look like 76/220. With a permanent fu xuan and sparkle (with s1). I basically have 100% crit rate then.
@@KuroiShiAnimu But I want to point out that it's really nothing to be bothered by - those great builds you see are probably refresh tier builds with whale investments. You could even clear MoC and other end game content with a three star lightcone with acheron. But if you want to vertically invest to get to those same type of builds then you could just farm forever... though without whale things (such as refreshing trailblaze power) it's not really recommended.
This was an excellent and interesting to watch analysis, and I appreciated the overall neutral tone of letting the numbers/statistics speak for themselves. After it was mentioned at 12:35 how useful a bimodal plot would be to show evidence for there actually being "special" UIDs, it occurred to me that even breaking the numbers down by weekday and patch number is possibly unsupported by the available data. I'm guessing that lumping all the data together like that shows values even closer to a true random distribution, and that perhaps that's why they decided to break it down by day/patch/region to try to (perhaps subconsciously) overfit what little data they had. Regardless, I'd love to see (or be a part of!) a much larger data gathering attempt without problematic self selection biases such as the "special" UID notice.
Thank you for proving and explaining the confirmation bias. I thought a similar thing when I saw that a “Special UID could cause what they call a shit day.” That’s like doing an experiment with a rat in a maze and when the rat doesn’t find the cheese you just don’t record it because “The rat is having a bad day”.
Something i'd be interested to see is if there was a bias, not for the slot that gets upgraded, but the stats themselves. I would think, if there was a bias for specific substats upgrading, then it'd be based on the stat itself rather than the slot the substat is in
Precisely my doubt on this whole thing. Why would hyv make a particular line more likely to be upgraded when they can just increase the weight of a particular substat like say, flat Def? That was what I thought when I heard "RNG manipulation" too. This just seems too weird to be true.
Yes it's been confirmed substats like break effect, defense, hp, and attack are more likely to be upgraded than others. Just like how energy regen ropes have a significantly lower chance to be rolled on main stat
Love it. Sorry I saw this late. Been busy. But yes everything you said here is correct. The shit uid thing was a made up word we used so it’ll be up to the player if they want to upgrade further or not. We even mention it more on the sheet how to be aware of it now. Dividing it even more and getting everyone’s uid…. At that point yep. I don’t got time for it. It’ll have to be up to my discord. Which it does seem they are working on multiple things. I try my best to not be biased actually. All data inputted is recorded if the players entering are trusted(has been there long enough etc) (yes we fear players messing with us and giving us purposely bad data) You don’t really need all out high level relic stuff yes. But of course everyone hates getting bad flat rolls right? The sheet was originally meant for that. Dodging stats you don’t want. You can try to use it to try to snipe substats. But honestly even with data that’s hard to do. Our eyes aren’t looking directly at the server, all we can do is record what we see in game. The most legal way of looking at it. I do understand your point of view and yes statistically what we do is not optimal. And it does become biased over time. Can’t really blame us for that though. Limited resources and all. Not all of us are gonna burn materials for what is needed. Players tend to get picky. Let alone even getting them to help with the data. It’s like accidental bias. We overall let the player decide to use our data or not also. Nothing really to lose cause you are gonna be rnging your upgrades anyway. Thanks for the video man. Hope I saw it sooner.
Thanks for watching the video. It is good to know that there is at least some kind of screening process regarding participants. However, I still think that anyone participating in the data study shouldn't be able to see the results in real-time as that does interfere with results. I would even go as far to say as this is a experimental flaw. I know you and your team try not to be biased, but things like confirmation bias happen on a subconscious level, without intent to do so. Though I understand if this change can't be made since part of the reason you can convince people to give data is by providing them results that they can use to roll during a patch. It is what it is, I hope you and your team have luck in the future getting more people to participate!
@@TarZBlammo not gonna lie. I’m surprised how much we data with the support we recently got. It’s kind of biased to certain days still. But at least all of the days are above 1k+ tests. Huge improvement over 2.0 patch. Only time will tell how it goes from here.
I wouldnt say nothing to lose, what you lose is relic exp on your checker and junk relics to test roll and credits and time on leveling these checks, and time on waiting for specified slight improvement days
9:15 I have serious doubts Hoyo uses Unity RNG for pulls or artifact upgrades. First, it needs to happen on the server. And second, particularly pulling RNG must be foolproof or they could get into serious legal trouble. I definitely hope for their sake that they're using scientific quality RNG. Also, why would Hoyo implement complex strategy to upgrades if just using RNG is much simpler. If there is a bias, it's due to programmer error.
c#'s rng algorithm has been scientifically verified and is more than good enough for applications like this. As for rng regarding pulls, there aren't exact legal parameters for it as generally, the community verifies rates through websites that store players' pull data and variation within 0.1% seems to be common for the current wish data recorded so far. As for your complex strategy comment, I completely agree, I don't exactly know why anyone would force themselves to create a completely new rng algorithm when there are already existing ones.
This video showed up well before any of the videos touting it as THE reason relics roll poorly, and I'm glad I gave it a watch as it's potentially saved me a lot of wasted time, energy, and effort were I to have fallen for the manipulation hearsay that's out there
I respect the lack of anger towards the SUID stuff, I don't think I could've contained myself if I made a video on this When I tried it really felt like I just wasted a few hours of my life trying to understand some bullshit theory with copout elements that go "yeah uh if it doesn't work you're just on a bad time lol" But even if we discard all the statistics, it makes no sense for this to exist anyway. If it existed at a time, Hoyo would have patched it out already as this kind of exploit would be bad for them, and they have way more data than we could ever gather
Spoiler: No, lmao. This is like the most basic common sense people should had. Imagine if this was real, one month later people would have their perfect relics, nobody would play the game anymore. Implying, of course, that this wouldn't be treated as bug and fixed by Mihoyo asap.
1. The relics system in Genshin and HSR is universally hated by everyone, if everyone have free good relics, people would actually play it more often. Do you think people only play the game to farm relics or something? 2. The 5 stars 50/50 drop rate is literally bugged in HSR and is actually 57/43 in your favor and they have yet to fix it, so there is a precedent of them just not fixing this
@@sammysaito529The 50/50 thing is most likely intended. "Wow my luck in this game is so good! I lose 50/50s way less!" And more people play it when they hear people say that
@@Boarbatrice Not really, people hate 50/50 no matter what, this stuff only starts to matter in higher sample sizes, aka if you play long enough, otherwise it's still effectively 50/50 for you. Also it's not necessarily intended, as OP already said in the video, the numbers check out if you assume that the Rate Up character is part of the "permanent" cast on their banner. Wuthering Waves seems to have the same issue, I hope these never get fixed though
@@sammysaito529 The 56/44 is 100% intended, if it's a bug they would have fixed it (duh). When you lose 50/50, the rate-up character is included with the standard line-ups. It's basically just a weighted coin-flip to make the players feel better about their pulls, they don't want people quitting because they lose five 50/50 in a row.
@@sammysaito529 1. That's the point. To keep people playing. 2. That's 400% hearsay. It's some weird rumor "some guy in China analysed millions of data points". Ok, where is the data?
That guy does not take criticism very well. Many people have tried to point out to him the issues with his data such as the low sample sizes and poor haphazard methodology. He's just not interested in listening. He seems more interested in building a community of yes-men around him. Basically things like the special UID are things he tacks on to make his theory increasingly unfalsifiable. The way the rates supposedly change every day of the week, depending on how long your clear time is, resets each patch, and can jut be randomly different for some people. Every single one of these works to make the theory unfalsifiable. I don't think he intends this, but he has a very strong bias towards his theory and interprets all new data as fitting the theory.
Agreed. He loves to use big words to convert your average people into yes-men, typically those people who just do nothing but regurgitate whatever they find online from certain creators.
@@broccoli4487to a degree, do think it works, since the data founded is good, and it can and does match, but the idea that it changes per patch is kinda un-justified. (CCTM is pretty valid, since data matches: 7 days of testing, with each day doing a run of traces mats at 1:30 M a day, and estimating to be 80-90 4 star trace mats + another 7 days, half at 1:00 M and the other at 2-2:30, and it gives half as much, to close to 5%) Theirs def validation to his claims: CN of GI 3 years ago talked about a method like this, it just never went anywhere because people kinda gave up on the project. (Or it went underground, idk) Also, he kinda kicks u out of the server after awhile, so I 100% understand the skepticism.
Such a good explanation for simple statistics, better than my classes at least. And a reaaly good critical analysis of simple prospective investigation. A nice way to get into statistics
Thx for making this video to save me the trouble of testing it. I figured it had to be BS, but I am not knowledgable enough about statistics to break things down well like you did.
Man this was a great refresh for my psychometry class. It's the first of your video that I watch but if not you should definitely make a series teaching statistics through gachagames. Homework made fun !
Not exactly. I guess we need to reword that. It was simply just a word we used to throw out there. The outcome is still the same if you think about it. Except the %s are flipped. That’s it like instead of the highest % being common for you then it would be the the bottem % that is common for you and the highest % on the sheet being least common for you. If you think it this way. All you would have to do is aim for the lower %s.
@@BigBoiPnoy I believe its a much safer assumption, that if your data is sound and a trend is established that "special UIDs" are just outliers statistically and nothing more as you have no way to prove they even exist, if the data collected isnt establishing a trend that paints a much worse picture that special UIDs are being used as a scapegoat to feel better about wasting your time trying to crack a complex algorithm with only 1 month of limited data
@@dankmeimes4274 on the recent updates to the sheet. The discord editors and moderators debunked SUID. It’s strictly something that happens during the time of day you are upgrading. Basically certain hours of the day have reverse outcomes that contradicts the overall data. But if a player waits 1-2 hours it matches data after. So basically with this knowledge, overall in a 24 hour period, there’s more cycles of one pattern over the other. And that’s why on data. No matter how much we record. It’s not a perfect 25% each line.
@@BigBoiPnoy this can just straight up come down to low sample size. Reroll 100000000 relics on 1 acc then we can start talking about statistical significance.
After going through college classes specifically for statistics in studies, I instantly saw all those flaws in the creators’ study lol Hopefully a better made study and data collection will be done jn the future
So glad you mentioned the "special UID" thing. It seems that every time the data doesn't distribute in the way that he wants it to he creates a new subset to explain it. All with no actual testing, of course. I think its pretty clear that he's p-hacking.
This kind of reminds me of the Spotify shuffle thing. People were complaining that Spotify’s 100% randomized shuffle didn’t actually feel randomized as sometimes you would get the same two or more songs in a row. Spotify wound up making their shuffle not a true shuffle to make it *appear* more random. What if that is something like this?
It could be, but I don't think haphazard data test such as the one in the video would be able to prove that. I mean they don't even publicly place on their sheets what kind of significance they have, so the statistical rigor really isn't there
Standard alpha is .05, in class the lowest we’ve done is .01 I’ll consider doing this upon a bias sample. I did not expect to learn stats from a Stair Rail video lol
I actually was sort of willing to give this the benefit of the doubt... until I saw the bit about "special UID days". Creating a special "cheat" category that's specifically designed for data that doesn't fit your hypothesis and then asking your data collectors to self-classify is COMICALLY against every fundamental of good data gathering. There are ways to deal with outliers in statistical data, and this is aggressively not it.
Tbh, never knew about that from when I used to do Relic manipulation, but hearing about it now, makes it sound like a scam, ngl :/ (but tbh, kinda glad I did it, and got stuff from it: My QQ is top 1% in the states, My firefly is top 5%, my Himeko is like top 20%, and my Fuxuan is like top 5% It did help, and kinda glad for it.)
from my experience reseting string its really does work, i usualy got every substat that got up after using the reset things i didn't encounter that much anymore
I have little experience in statistic but this is insane... you went quite deep to actually analyze the bias theoretically... honestly too much variables to be able to "manipulate" and yet the probability is for each sub stat is mostly close for every sample... at the end this is "probability", tendency/bias could have been an accident...
Such thing was believed to be true in other gacha Fate Grand Order as well but not about relics, about gacha results that it's based on seed and the time when you roll
So here's how the algorithm ACTUALLY works: If you have three good substats, you get three to four upgrades on the one you don't want. You're welcome, problem solved.
I loved the video and agreed with pretty much the entirety of it ! Even with my limited statistics knowledge, I know how flawed the data collection methods used for the sheet are, among other issues. However, one thing that I find compelling which keeps me from outright dismissing Bernard’s videos is the fact that on a lot of character artifact leaderboard websites, he does appear to be very highly ranked with many characters, competing with whales who constantly spend jades to refresh their stamina (which Bernard claims he doesn’t do). If he’s able to go toe to toe with such people by using his method, can you provide any insight as to how we could analyze if his rankings are significant, or just a result of random chance ?
To analyze his rankings requires quite a few assumptions; how many good relics does he get to roll on, what days does he do it on, how many good substats are upgrades per relic, etc. You can use a binomial distribution to calculate his probability, assuming various factors, but results significantly vary based on those assumptions. Probability aside, it is never good to use anecdotal evidence to justify a method. For example, there are always people that win the lottery, despite the low probability. Does that mean that there is some method to solve the lottery? Most likely, no. This isn't the best comparison, but the point is even if you were to prove that the probability based on normal odds is low that he gets these relics, it doesn't prove his method works.
Personal thought on this : you can see this is how the rng work but just add this detail of mine that they cant fully calculate the randomness which mean there are still some factors they missing to predict the roll cause in the end of the day this was made by group of individuals so there is a way to controll it or at least minimize the rng
It would be interesting if you could statistically say, it is more likely that the RNG is the same day-to-day and patch-to-patch, than that all of the other claims are true. They've got the data for old patches, I wonder if it's enough.
It would be doable if the sample sizes weren't so different. Because of their data collection method, certain days get far more samples than others. It would be like comparing apples to oranges at that point, I don't think you would be able to make any concrete conclusions.
Lmao, I usually dismiss such stuff immediately, but thanks for you effort, I really started laughing when you said that this guy invented"special UIDs" which ofc purely exists just to patch holes in his theory, to explain all the results that doesn't align.
i’ve had multiple teachers (including a UC berkeley professor) try to explain basic statistics to me and it’s legit never clicked until now- i guess all i needed was a video game example LOL
If i have a relic that alrwady has crir rate cdmg, a flar hp and no 4th stat it's automaricalyl goin wholly into that awkward stat that doesnt work and why is it always def% or def flat.
I'm 100% on board with doing voodoo to pray for better substats but their method is just too involved for something you do as often as upgrading artifacts. Overly elaborate rituals are reserved for banner pulls.
Whatever is said about that "technique", I used it several times, and in 90% cases I could improve the substats I was either aiming to improve or not against improving. It is up to you what you believe in, and there is nothing wrong in believing in the bias of RNG :)
My one piece of advice is to always remember that people are notoriously bad at being able to tell what is and isn't rng. That's what I said before analyzing it myself and getting similar results to you
i didnt know this subject was a thing til now, and tho the theory seems very weak in supporting data, i wouldn't be surprised if a certain bias existed. not the way its presented but depending on the substat, cause im oretty sure theres a bias in rolling in genshin artifacts, wouldnt be surprised if something similar was on star rail
My opinion from my understanding on how the rng in genshin/star rail is, is that it's not an rng client side but server side since the game requires you to be online to play all time, that it uses time (from the pulls we see hoyo record the time in which the wish are spent) and either uid or sid or a combination of both. Leaving us no chance to know how to abuse the rng.
Nah, I think it’s more like Bernard says, and that of RNG manipulation: Randomness isn’t a true value, and the fact that data acts like this says theirs definitely a pattern, since while I don’t follow Bernard’s manipulation guides as often as I used to (most b/c it’s expensive as fuck.) they have value: Cavern Corrosion manipulation theory states that are certain time intervals/times, certain probabilities are more often, and from my own testing, that’s pretty true. (A time around 1:30 on certain caverns give better substates, with times above that, like 0:49 or 0:30 being a lot worst, with 5/6 times, dmg sets giving 2 or more hp states, and at 1:30, 1/6) Also, it’s server, with certain strings being more common around certain times than others. (I saw a dude get a 4 state upgrade orb, and 2 hours later from that post, I got a 4 state chest.) Also, certain rolls are actually more frequent than others. (I get more high to low rolls on Fridays, Wednesdays just suck, and Saturday’s and Sundays are roughly average.) Theirs 100% a pattern, it’s just that Bernard and everyone’s attempts to find it kinda just suck, and since testing cost so much, imma just do CCTM, get 5 god pieces, and just hope for the best :/ (since I did CCTM, I’ve quadrupled my god relic stashes, since they’ve became easier, but not the easiest, ngl ;-; ) Also, sometimes placebo for some people is just better, even if it doesn’t seem to work.
Ngl I had already believed in "good days/times" to roll artifacts in Genshin, so even though I was skeptical I gave his sheet a try. After a while I basically adapted my own "house" rules on how to roll and only loosely followed the data sheet and felt I was getting tangible results from it. That said though, to be totally honest, I know it's confirmation bias copium to think the spreadsheet is useful in anyway. But it's legit just therapy for my monke brain to believe "UNLUCKYYY" when I don't hit my rolls so I still feel a net benefit in using it over not using it LMAO. Love the video btw even though it did more mental dmg to me than my actual relic rolls loool
I'm sure most of your analysis is correct and it's generally hard to believe in RNG manipulation until the evidence is pretty solid. One thing I'd mention though is that the RNG likely doesn't happen in C#. It's more likely that it happens on the server as doing it in the game would be way too insecure and easily exploitable. And it would also not be the first time a game used a completely artificial RNG system - e.g. some gacha games lowering their pull rates for the first day of a new patch, so making statistics and getting as accurate data as possible is definitely a good thing.
That's fair to say the game servers aren't necessarily coded in c#, though, c# is actually decent for backend applications. It isn't necessarily that far-fetched to assume the devs used the same language for both, though you could use a wide variety of languages for the backend, with users being none the wiser. Wonder if data mining has been able to reveal anything of the server calls.
@@toffiehacker3701 Gamblers in denial alert 🚨 The Chinese community debunked this time and time again, from “chest will respawn” myth to “RNG manipulation if you pull within this frame” to “manipulating relic stats” these are literally just survivorship bias + people only posting when they succeed + people preying on the insecurity of others. It’s just gambling.
@@toffiehacker3701 TLDR: It’s not popular because it’s horse sh*t, if you genuinely believe this then I strongly advice you to never step foot in any Casino.
@@StarsShatteredBeyond Unless they can count cards. But even if they can they still shouldn't because that's how you end up with broken knees in the alley out back.
I don’t really want to manipulate the stats of my relics at this point but the substats definitely is biased, getting a good substat is horrible compared to genshin impact and other gacha games and the reason idk but at least it makes me believe that there should be something behind
How convenient that special uid's exist. How conveninet that the pattern change every patch, so we cant use data from 1.0 to today. How convenient "its all server side bro" so we cant find anything about it from gamefiles or leaks. Its almost like there is a convenient way out to anything that could poke a hole in the theory.....
Well the server side is not a convenience is a security matter, long ago we would known the RNG formula so RNG manipulation would been a thing, the thing of special UID and changing the pattern sounds too exploitable to not be known publicly since before. Wouldn’t say convenient but a extremely elaborated hypothesis.
Answer: No. There is no such thing as RNG manipulation in Gacha games. This should be obvious to anyone who understands that RNG manipulation would cost the developers tons of money, and as such would never be allowed to exist. If it did somehow sneak into a game without the devs noticing, it would be patched literally the same day that someone posted a video showing it was possible.
It proves a bias, indeed. But one thing that differs this RNG from Pokemon Emerald is because the Emerald itself has been reverse engineered and it was found out to be a game which was ran by seed. I dont remember the exact number but if I'm not mistaken, it was running a seed from 0 to 65535, so you could entirely manipulate it and this is how speedrunners abuses it by manipulating the optimal time to start the game because the seed rolls are predetermined by the flow of time in the game. However, its different on HSR tho, all testers are doomed for confirmation bias because there's no way in hell that Hoyo would actually show you the substat rolling chances from their dev tools. All we could do is setup a table and have a sampling size large enough for it to be concluded and even then, we are still not sure if that was the actual chances of the rolls. The study would be filled with bias because I'm pretty sure the narrative of getting HP/DEF for most of their relic rolls weren't just minority, its the majority who weren't even part of this study and a large chunk of that data has been left out which could contribute for the study to be concluded. So I'd say that there is a predetermined weighted average stat rolls for each and every stat in the game, with HP and Def rolls being the highest. The 25% bs rolling chance should go away considering that the actual CV rolls proves it rarity in the game and numbers were tested in CN community where the each substat dont have an equal distribution chances for the relic to have 25% chance distribution for each substat. As for its actual numbers: docs(dot)google(dot)com/spreadsheets/u/0/d/1-MgfpwtN0PwR04eiqJp-PcaPB12oIkeaz-icEp-61b0/htmlview# Just replace (dot) with . because youtube obviously doesn't allow links to be posted. That is the sample size conducted in the CN community.
I did try the relic manipulation thing, and it kinda worked for me? For some reason resetting the string by upgrading a piece with 3 sub stats into a 4 stats, kept giving me a second upgrade into the 3rd slot. I got 9 straight upgrades into a 3rd slot by doing in it and it was in accordance to the sheet. It felt very impressive. I have no idea obviously if it works or not, however one thing that is a fact is that Bernard has multiple characters in the top 1% of better geared characters. That's either insane luck or something going on. I do believe that the theory is a Frankenstein of ideas and there is no clear model, but among the trash some piece of truth might be there.
I used to play Genshin before and it has been sort of quite obvious that the RNG isn't "fair" from there so I never expected them to come fair as well in HSR. You farm and enhance thousands of relics and yet you only manage to use only a few of them. You salvage/feed more than you actually use. You get more DEF stats than actual speed stats. You don't get ER link rope as often as you get DEF/HP. It's a really bad practice and I don't condone it but I guess at the end of the day, it prolongs the game from stagnation since if people were to finish up farming so quickly on a really fair RNG, they would have nothing to do while waiting for the next patch.
5 months later, did anything change? since there's now more data being collected and Pnoy has admitted that the Special UID thing has been debunked, and now claims that the seed changes around a 15min~60min cycle
I don't really know how they "debunked" special UID. There isn't a good reason based on the data I have seen as to why they just rejected this theory (no attempt to plot a bimodal distribution). As for the seed cycle changes, why would this be the case? Why wouldn't it change every second/millisecond, which is what would happen if timestamp is part of the seed? Also, by increasing the amount of categories they have, they also need more data to characterize everything as a result, so I don't know if you can really get anything meaningful out of their data as of now. A lot of their analysis is haphazardly suggesting theories and throwing out others without sufficient statistical evidence to back it up. To me, it doesn't seem like much has changed, though I haven't taken a great look of detail as to what they are doing now.
Remember guys, if its not 100%, its 50/50. Or 25/75 in this case... I do find myself troubled in this case, most people i know who actually (wouldn't say bought) tried these method before are people with worse relic quality. Most of the time this is due to them throwing away relics that could potentially be good or being too much of a hoarder and not rolling any relic unless its the perfect stats or they feel like rolling. And to be frank, i checked the "phropet/manipulator/even cringier, elio" and his relic quality are the same as people who are used to mmo/gacha with rng stats system. I know you shouldnt insult the guy but at a first glance, dude basically looked like tandrew ate who cant even fool people with half a brain cell nor is benefiting from any of it
Well, the community recently said Special UID was debunked and instead probabilities change every set interval of time. And also what do you think of clear time theory? (specific clear times leads to dropping more desirable substats)
I don't know how they debunked special uid given that they never presented a full distribution (which they don't have enough data to properly make afaik). As for clear time theory, I think the same issues presented in this video still apply. Way too little data to describe their categories and why 30 second intervals? If your relic rolls are seeded based on time, why wouldn't it be to the second/millisecond? I also really don't get how they judge "desirable", because these stats vary per character (some want break effect, others want crit, etc). I would also like to note that RNG algorithms aren't directly time related and depend on time if timestamp happens to be used as or part of the seed. There might be some kind of RNG algorithm out there that uses time outside of the seed that I am not aware of, but these are the only two ways for time to be utilized directly. In summary, way too many questions with not many answers. I also just dislike the idea of presenting a hypothesis as a legitimate method when there isn't enough statistical support.
I trust myself. Using the strats i made 36 cv relic for myself, some other similar relics and some break effect relics. Being a newer player i am focused on upgrading trails and only farmed relics at the start of TL60 and during events, so these are great for me.
Math lad here: Too lazy to do a publish study in a YT comments section. But based on a cursory fiddling with the sheet and it's data source. It does NOT provide enough significant OR reliable data to prove it's less than 5% variable range. It relies on publicly submitted data which means it's definitely tampered with either intentionally or unintentionally. Meaning if anything. Your wasting your time.
General rule of making something: Kiss. And for those aren't familiar it means "Keep It Simple, Stupid!". Doubt they would do something much more complex than something like "rand() % 100" and comparing the output with a probability table.
That's a RIDICULOUSLY small alpha. I can understand that you're trying to make a point here (the confirmation bias is the main issue here as well as the pruning) but the alpha chosen here definitely feels biased to try to further this argument An actual RCT needs to be made for this to prove any statistical significance, not this confirmation bias garbage, very true
From what I recall, standard alpha cutoffs are 0.05, 0.01, and 0.001. There are 2 reasons I chose 0.001 as the alpha: what I stated in the video about being able to find seeds that produce an alpha lower than 0.01 relatively easily and reducing type I error. Given the size of the sample sizes presented here, you could easily make hundreds of thousands of groups from the amount of relics people roll daily, in which you would expect results on the level of alpha = 0.001 simply due to how many groups you could get. Yes, it is strict for something like this, but if you want really strong evidence your theory works, you should get down to this level of certainty.
@@TarZBlammo I understand the reasoning, and I suppose this level of strictness is fitting for the small sample size of the data It's just that the basis of the data, not your analysis, is a huge mess. Like you can be as rigorous as you want and can but biased data, especially to the evident case here, is truly unworkable. Like even if you did find the p value to be smaller than 0.001 it still wouldn't give us a good conclusion on the overall system is what I'm trying to say Your video is informative and its effect is positive, I'm in no way trying to imply otherwise, but I feel like more stress was given to the hypothesis testing when, in my opinion, the result of it would at best raise speculation It's like trying to extract statistics of a fair die using a loaded die. It just won't work out
There is no way this is real that would require the Hoyo devs to literally be insane. Any deviation from uniform randomness is confirmation bias. I trust Hoyo to make sure they don't give anyone an edge they're not paying for way more than these guys are reporting the data correctly and fully. Also it'd be insane for Hoyo to do anything but just a straight source of randomness independent of UID or anything else server side. Just have a mersenne twister initialized with a system source of randomness in a singleton and use that for everything this is a solved problem why do something that could go wrong or be exploited with more effort than that?
The bit about “special UIDs” just really irks me in particular. Idk if there’s a particular term for it, but it gives the vibes of someone being like, “If you tell me I’m wrong and provide evidence, then I’m not actually wrong because your evidence doesn’t count.” Like, stuff that contradicts your argument doesn’t get to be invalid just because you don’t want to be proven wrong.
@@nehuenconde7139 legitimately, this feels like moving the goalposts. As far as I’m concerned, there was never such a thing as SUIDs, and the fact that the narrative has changed to “it needs to be at an exact time” just feels like even more bullcrap.
@@johnnyjoestar3318 That's what I just said to another comment. I have nothing against the creator but I do feel like he's bringing in the ol' Filipino "haka-haka" crap we have in our country since he is Filipino himself. Basically, it's similar to superstition, such as "don't cut your nails on Friday or you'll get bad luck". Can't help but feel it's this with all the BS I'm seeing everywhere. "If your rolls didn't align to my theory, then it's just bad timing/days/luck"
This theory of RNG manipulation is inherently flawed because it is based on a premise that it is the stat line position that can be manipulated, rather than the stat itself. I am pretty sure Star Rail uses weighted substat RNG - I remember seeing something about that a long time ago, but I don't remember where I saw it anymore. Also worth considering that the Relic system is a semi-copy paste of Genshin Impact's Artifact system, which DOES have weighted substat RNG (crit stats roll WAY less often than atk/def/hp), makes the case for weighted RNG in Star Rail very likely. This becomes a problem since in two relics with SPD stats, one SPD stat may be on the 2nd line while the other SPD stat may be on the 3rd line. It would be registered then as a 2 and 3 respectively.
The weighted substats you remember is when the artifact is created at +0 what stats it has. This is talking about upgrading substats which in theory should have 25% chance on each line
![[2.3 HSR] Aeons Come in Pairs - A Honkai Star Rail Theory](http://i.ytimg.com/vi/iMRtiNQHmQo/mqdefault.jpg)
I love how humanity went from pythagoras creating theorems for understanding triangles to young adults figuring out the statistical significance of an rng manipulation theory for a game.
To be fair, they probably did the same thing back then but we just don't hear about it. Just like how future generations won't hear about this, but instead will know all of the things we made that had lasting significance.
I kwow that this is a joke, but what do you think scientists and mathematicians are doing in our generation? Things haven’t really changed, it’s just that Science and Mathematics are more accessible to the general public and therefore are used for trivial tasks like this one.
Actually humanity hasn't changed at all.
In despair they pray to a higher being/system for salvation. And that's also what happens here. People are frustrated by RNG and then a prophet appeared and promised them salvation as long as they believe. Rest is confirmation bias.
There was a question on hypothesis testing of the rates of getting a five star in Genshin in my statistics final last year haha
bet you havent encountered people calculating breast size of gacha waifus based on number of pixels
or successfully identifying/tracking a criminal/assaulter based on cctv footage of his shoes
or tracking down someone with an arrest warrant based on pizza boxes seen in his Twitter post while arguing with a grown child
Application security expert here who has some experience exploiting PRNG mechanisms. The thing that bothers me about this theory is that it relies on some weird implementation choices on behalf of Mihoyo. It implies that the state of the RNG generator only changes by user action, which would mean that either:
A) Mihoyo is querying RNG client-side for progression.
or B) Each individual player has their own RNG state serverside.
RNG manips would be almost as bad for a Gacha game as it would be for an online casino, and this isn't the publisher's first game in the space. I have to imagine they at least know the basics of preventing RNG attacks. With that in mind, B would be a wild choice. You would have to do it specifically, and it only opens you up to RNG manipulation attacks.
A would be even more insane and would essentially mean that a skilled attacker could control the RNG themselves. The fact that there isn't (to my knowledge) a thriving underground market for RNG hacks suggests that this isn't the case. (Yes- there is anti-cheat, but there's no such thing as foolproof anti-cheat).
And also, just several of the techniques and data here just don't line up with how I would expect a likely RNG attack for this system to work. I can theoretically explain multiple parts of it individually, but holistically its difficult to explain what might be happening in code.
Of course, just like the statistical analysis, this doesn't conclusively prove anything. RNG manipulations are some of the most inscrutable and arcane attacks out there, and even when they work, usually aren't 100% consistent or reproducible. They're some of the most fascinating attacks in speed-running, and if you want an example of an amazing one, look up "How We Solved the Worst Minigame in Zelda's History" by Linkus7.
From a programming perspective, I definitely agree. Speedruns have some really interesting manipulation tech, I remember for pokemon emerald, you could initiate the first pokemon battle at a specific frame to get really high IVs on your starter due to the way PRNG reset in that game. I don't really get why the study decided that blocking into hours/days made it easier to analyze because even if the timestamp is used as part of the seed, you would use something like a UNIX timestamp. To be fair, there are some weird techniques out there for PRNG manipulation, including the Linkus video you mentioned.
I had a wild thought a while back of trying to get a machine learning algorithm try to guess what I would get from rolls from gacha. I'd be betting that they use PRNG and not some sort of physics based RNG mechanism, and the limited info available to the client would be enough to reverse engineer it. Do you think such a technique would be feasible?
@@davidebicmaybe you could depending on the model you want to apply, but getting enough accurate data to make said model would be fairly difficult
@@TarZBlammo yeah I also think it's not an easily doable approach, I'd need to get a few thousand people to help me with the dataset, at which point it would be hard not making HoYo aware of such an endeavour. Then they could just slightly change the seeding algorithm and it'd be all wasted effort.
@@davidebic technically speaking even physics based RNG isn't really a thing on a macroscopic level since on our level everything is deterministic, on scales that quantum effects are significant then you get true and unpredictable randomness with a known distribution
Not me using Pythagoras to make a little girl throw a mahjong piece harder 😂😂😂
Doing Ratio's work and stopping the spread of the Enigmata. Nice job
It's so nice to see real statistics applied to something like this, loved the video! I think it would actually be very interesting to run a study on substats taking into consideration all the issues you mentioned. I doubt it would be possible though, since you need very high sample sizes and there is little incentive for players to submit stats if they can't see the results until the study is complete. Have you heard about Star Rail's 50/50 being closer to a 56/44? Some people are saying that the sample size is too small to reject the null, and I'd love to hear your take on that.
Thanks for the compliment! As for the substat rolls themselves, I think there would be way too many issues, especially given the necessary data collection involved. As for the 56/44, I don't think the sample size is an issue given the 15 mil+ data points that the study claims. If anything, the bigger "issue" is that people that post their pulls may want to brag about their luck, biasing the sample group. Reading more into it, the creator did reduce this potential source of bias as much as possible and did multiple statistical tests. On top of this, I am more inclined to believe the rate is actually 56/44, given that there is a legitimate programming reason why these particular percentages can appear (banner character being included in the pool of standard characters when losing). Generally when significant bias happens from a RNG algorithm, it is mainly due to programming error rather than a fault of the algorithm itself (unless you pull a no-man sky and use a really crappy one).
@TarZBlammo Yeah, like he said. I heard that after you lose your 50/50, you still have a 1/8 chance of getting the limited 5 star, and a 7/8 chance of getting one of the other 7 standard 5 stars.
Do you know what, just IF, 56/44 ratio is true, I must be very unfortunate being in the 44% 5 times, IN A ROW 😂😢
Reading this comment during breaktime in the middle of my Psychological Statistics class is giving me whiplash.
I saw a study pointing out that although it’s closer to 56/44, you have a lower curve of getting pity in Star Rail, so the mathematical expectation of both games getting and winning the 50/50 is about the same.
I was really interested in learning how to do this method until I remembered that I'm a psychology student that was forced to study statistics
Once I realized that the statistics samples were too low to prove anything and that there's a good amount of confirmation bias here, it made it clear to me that this is too much work for a low chance of it actually being legit
Is it really confirmation bias if there's real proof that it works? The specific channel on his discord is full of S+ relic rolls every single day, i have no idea why people are so antagonistic towards him when dude actually share how he got his relics to other players, at least try it yourself before dismissing the method. I got my Acheron to 80/200 crit ratio thanks to this so i'll keep abusing this until hoyo make changes lol
I would love to post images of my rolls and other people's roll here using the theory, but youtube doesn't allow links in comments
@@Amalia19 ah yes, because all the people who got dogshit relics are gonna excitedly post screenshots of their dogshit relics to contrast the S+ relic rolls lmao
@@LilacShowers Yes we also put our double crit relics that got ruined into the sheet, to make sure no one rolls on that specific time in the future since the string pattern in that specific time is shit
I don't understand the agression, is it because you can't believe all that relic you got could be better? How many times do you even get a good relic, once a year? lol we're getting that in daily basis, it's your loss anyway so i'm not gonna bother trying to convince you anymore
@@Amalia19 that entire comment shows zero understanding of statistics lol even if the method entirely worked you'd still get counterexamples because it's still rng even if mitigated and also while i fully believe at least some of you do put your data that contradicts the trend into the sheet (albeit seemingly with the mistaken impression that it must mean those are "bad days" instead of just unlucky data lmao) we weren't even talking about the sheet, we were talking about the discord channel, where you were bragging about people posting lots of S+ relic rolls every day, because most people would not in fact go and post their bad or mediocre rolls in that channel lol (which was my point before)
i don't think i was even particularly aggressive in my other comment; i mean, i could accept "dismissive", but pretty sure you're just the one being like super defensive, which... i don't really understand, because if your claim that this method really worked out for you is true then you wouldn't have much reason to be huffing copium about it and i'm not sure why you'd lie about that; maybe you've actually just gotten a few lucky relics in an ocean of shit that you've convinced yourself that nah the method works it's just that "that specific time is shit" so you're doubling down to deal with the cognitive dissonance that you seemingly just so happen to keep rolling relics at "shit times" lol
anyway go off i guess
...come to think of it if you can just randomly get "shit times" how does this even help lmao like if the theory is true it sounds like it just adds another layer of uncontrollable rng
I saw one of those videos a couple months ago, and with how odd programming errors can appear in all kinds of places in games I was open to the possibility, then I saw the sample sizes for the substat combinations and the disclaimer that the patterns actually "weren't 100%" and noped out, these people are reading patterns into white noise.
Im on the dumb side of the bell curve that's also sure that there isn't a way to manipulate it cuz of one thing. There'd be a lot more perfect relics if it was a thing.
Either way, after seeing that video, I stopped dumping all my xp into a single relic at a time and instead upgrade 1 line for separate relics. I’ve definitely been getting better results.
@@ZeeZeeVee because doing that does make sense. If you have, say, 20 relics with 3 substats, choose the best one among them and roll it to +15, you could get lucky and get a really good piece, or roll on wasted substats. If you instead roll all pieces to +3, you'll have a smaller, better pool of new "promising artifacts". Since you now have 20 artifacts with 4 substats, you can discard the worst one, and then roll to +6 on the remaining ones. Doing this repeatedly basically culls the worst artifacts and leaves you with the best ones of each step, instead of going all in on a single piece (who could have been discarded when getting them all to +3 or +6 due to bad rolls).
Doing it this way is more tedious and labour intensive, and most of all: an exercise of patience, since you don't roll the first good piece you see, instead wait until you have a bunch of promising pieces and then do a rolling session. You could have to wait more than a month to have enough pieces to "guarantee" at least a good one
@@ArielGoV Damn, this sucks, the wait time only comes because of the limited amount u can get per day.
I saw a post from someone who's EXTREMELY unlucky grinding for Trace Materials, they asked the customer service about is not getting any 4 star rarity is a bug. They replied with something like "I feel sorry for you but that is not a bug, just RNG" 😭 RNJesus really just does that
How many 6 wave trace domains did they do?
At least a few days worth of stamina I think? I didn't save the post link but it's in Chinese
@@1438wei eeeeh, that's not *that* unlucky. I mean it's definitely unusual but I have gone 2-3 days with no purples multiple times.
Some of those poor 5% ppl out of the 95% confidence interval do feel this way, unfortunately…that’s just how distribution works.
Yeah it just happens. Went 600 resin in a row with no purple drops once, sometimes it do just be like that.
I was noticing some of the things you said in your video with what I was experiencing with my account. It's not like it's going to hurt to follow the relic manip guides cause there really is nothing to lose right - but I got some of my best artifacts just randomly upgrading to 15 more than I do with the guides. Like you said, I feel like the data amount is indeed low enough that you can never really tell.
So really good video my guy! Like you emphasized, it's not really bad that people are studying this because after all - everybody wants them sweet builds. Again there is really nothing to lose when following the relic manip guide or not.
Feel the same, I think the more data they get its going to look like everything with 25%, making it just random so guess the best way is just keep farming and trying, also don't need perfect crazy build just to flex with every single character
@@asoret facts man... We don't really need the perfect 80/200 "good" looking builds to clear in the first place. Just keep farming if you're a beginner and the same with endgame players and you will get results whether or not you follow the relic manip guides. I feel the same way that it just looks like to me that if they took a million player's data (hypothetically) then it would in deed look like it's distributed equally
@@fooo0p Man I saw an Acheron who had like 95/260 Crit ratio on just combat start so it was a permanent thing not a temporary Bronya Ult or something like that, though 12% Crit Rate did come from Fu Xuan but that's something with a 100% uptime by default. It's been eating up the inside of my head making me farm better stuff for my 82/200 Acheron (82/240 once Sigonia planar set stacks up). I am blinded by jealousy everyday
@@KuroiShiAnimu damn you on sigonia set? unless is PF then it's not really optimal but holy you got that great build.
I am on 46/220 but I do have e2 and izumo (that 12% atk is not to be underestimated - it's 4 substats worth and that 12% crit is permanent with at least one more nihility so it doesn't need stacking like sigonia). So without buffs it's going to look like 76/220. With a permanent fu xuan and sparkle (with s1). I basically have 100% crit rate then.
@@KuroiShiAnimu But I want to point out that it's really nothing to be bothered by - those great builds you see are probably refresh tier builds with whale investments. You could even clear MoC and other end game content with a three star lightcone with acheron. But if you want to vertically invest to get to those same type of builds then you could just farm forever... though without whale things (such as refreshing trailblaze power) it's not really recommended.
This was an excellent and interesting to watch analysis, and I appreciated the overall neutral tone of letting the numbers/statistics speak for themselves. After it was mentioned at 12:35 how useful a bimodal plot would be to show evidence for there actually being "special" UIDs, it occurred to me that even breaking the numbers down by weekday and patch number is possibly unsupported by the available data. I'm guessing that lumping all the data together like that shows values even closer to a true random distribution, and that perhaps that's why they decided to break it down by day/patch/region to try to (perhaps subconsciously) overfit what little data they had. Regardless, I'd love to see (or be a part of!) a much larger data gathering attempt without problematic self selection biases such as the "special" UID notice.
LETS GO my favourite genshin math guy doing star rail math??
frfr
Every once in a while, gamers have to be taught about statistics again
Thank you for proving and explaining the confirmation bias. I thought a similar thing when I saw that a “Special UID could cause what they call a shit day.”
That’s like doing an experiment with a rat in a maze and when the rat doesn’t find the cheese you just don’t record it because “The rat is having a bad day”.
Something i'd be interested to see is if there was a bias, not for the slot that gets upgraded, but the stats themselves. I would think, if there was a bias for specific substats upgrading, then it'd be based on the stat itself rather than the slot the substat is in
Precisely my doubt on this whole thing. Why would hyv make a particular line more likely to be upgraded when they can just increase the weight of a particular substat like say, flat Def? That was what I thought when I heard "RNG manipulation" too. This just seems too weird to be true.
Yes it's been confirmed substats like break effect, defense, hp, and attack are more likely to be upgraded than others. Just like how energy regen ropes have a significantly lower chance to be rolled on main stat
break effect huh? no wonder my firefly is eating good with 208 break effect just from relics@@moxielovesu
@@moxielovesu It is confirmed that some stats has more chance of being on an artifact than others, but being more upgraded there isnt any confirmation
Love it. Sorry I saw this late. Been busy. But yes everything you said here is correct. The shit uid thing was a made up word we used so it’ll be up to the player if they want to upgrade further or not. We even mention it more on the sheet how to be aware of it now.
Dividing it even more and getting everyone’s uid…. At that point yep. I don’t got time for it. It’ll have to be up to my discord. Which it does seem they are working on multiple things.
I try my best to not be biased actually. All data inputted is recorded if the players entering are trusted(has been there long enough etc) (yes we fear players messing with us and giving us purposely bad data)
You don’t really need all out high level relic stuff yes. But of course everyone hates getting bad flat rolls right? The sheet was originally meant for that. Dodging stats you don’t want. You can try to use it to try to snipe substats. But honestly even with data that’s hard to do. Our eyes aren’t looking directly at the server, all we can do is record what we see in game. The most legal way of looking at it.
I do understand your point of view and yes statistically what we do is not optimal. And it does become biased over time. Can’t really blame us for that though. Limited resources and all. Not all of us are gonna burn materials for what is needed. Players tend to get picky. Let alone even getting them to help with the data. It’s like accidental bias.
We overall let the player decide to use our data or not also. Nothing really to lose cause you are gonna be rnging your upgrades anyway.
Thanks for the video man. Hope I saw it sooner.
Thanks for watching the video. It is good to know that there is at least some kind of screening process regarding participants. However, I still think that anyone participating in the data study shouldn't be able to see the results in real-time as that does interfere with results. I would even go as far to say as this is a experimental flaw. I know you and your team try not to be biased, but things like confirmation bias happen on a subconscious level, without intent to do so.
Though I understand if this change can't be made since part of the reason you can convince people to give data is by providing them results that they can use to roll during a patch. It is what it is, I hope you and your team have luck in the future getting more people to participate!
@@TarZBlammo not gonna lie. I’m surprised how much we data with the support we recently got. It’s kind of biased to certain days still. But at least all of the days are above 1k+ tests. Huge improvement over 2.0 patch. Only time will tell how it goes from here.
I wouldnt say nothing to lose, what you lose is relic exp on your checker and junk relics to test roll and credits and time on leveling these checks, and time on waiting for specified slight improvement days
9:15 I have serious doubts Hoyo uses Unity RNG for pulls or artifact upgrades. First, it needs to happen on the server. And second, particularly pulling RNG must be foolproof or they could get into serious legal trouble. I definitely hope for their sake that they're using scientific quality RNG.
Also, why would Hoyo implement complex strategy to upgrades if just using RNG is much simpler. If there is a bias, it's due to programmer error.
c#'s rng algorithm has been scientifically verified and is more than good enough for applications like this. As for rng regarding pulls, there aren't exact legal parameters for it as generally, the community verifies rates through websites that store players' pull data and variation within 0.1% seems to be common for the current wish data recorded so far.
As for your complex strategy comment, I completely agree, I don't exactly know why anyone would force themselves to create a completely new rng algorithm when there are already existing ones.
This video showed up well before any of the videos touting it as THE reason relics roll poorly, and I'm glad I gave it a watch as it's potentially saved me a lot of wasted time, energy, and effort were I to have fallen for the manipulation hearsay that's out there
this remind me of the videos about chest respawn routes on genshin in its first year lol
I have a stats project I'm really procrastinating on but this video's lowkey giving me the energy to start it
We all know if a relic has atk%, crit rate, crit dmg, and flat def as substats that every upgrade will go into flat def.
magically it becomes 10%/5%/5%/80%
I respect the lack of anger towards the SUID stuff, I don't think I could've contained myself if I made a video on this
When I tried it really felt like I just wasted a few hours of my life trying to understand some bullshit theory with copout elements that go "yeah uh if it doesn't work you're just on a bad time lol"
But even if we discard all the statistics, it makes no sense for this to exist anyway. If it existed at a time, Hoyo would have patched it out already as this kind of exploit would be bad for them, and they have way more data than we could ever gather
idk man but my luck suddenly sky rocketed after trying the "manipulation". 5/6 relics rolled mostly on crit values.
i tried his manipulation, only 40% works and it's really wasting time to read his data
Spoiler: No, lmao.
This is like the most basic common sense people should had. Imagine if this was real, one month later people would have their perfect relics, nobody would play the game anymore. Implying, of course, that this wouldn't be treated as bug and fixed by Mihoyo asap.
1. The relics system in Genshin and HSR is universally hated by everyone, if everyone have free good relics, people would actually play it more often. Do you think people only play the game to farm relics or something?
2. The 5 stars 50/50 drop rate is literally bugged in HSR and is actually 57/43 in your favor and they have yet to fix it, so there is a precedent of them just not fixing this
@@sammysaito529The 50/50 thing is most likely intended. "Wow my luck in this game is so good! I lose 50/50s way less!" And more people play it when they hear people say that
@@Boarbatrice Not really, people hate 50/50 no matter what, this stuff only starts to matter in higher sample sizes, aka if you play long enough, otherwise it's still effectively 50/50 for you. Also it's not necessarily intended, as OP already said in the video, the numbers check out if you assume that the Rate Up character is part of the "permanent" cast on their banner. Wuthering Waves seems to have the same issue, I hope these never get fixed though
@@sammysaito529 The 56/44 is 100% intended, if it's a bug they would have fixed it (duh). When you lose 50/50, the rate-up character is included with the standard line-ups. It's basically just a weighted coin-flip to make the players feel better about their pulls, they don't want people quitting because they lose five 50/50 in a row.
@@sammysaito529 1. That's the point. To keep people playing.
2. That's 400% hearsay. It's some weird rumor "some guy in China analysed millions of data points". Ok, where is the data?
That guy does not take criticism very well. Many people have tried to point out to him the issues with his data such as the low sample sizes and poor haphazard methodology. He's just not interested in listening. He seems more interested in building a community of yes-men around him. Basically things like the special UID are things he tacks on to make his theory increasingly unfalsifiable. The way the rates supposedly change every day of the week, depending on how long your clear time is, resets each patch, and can jut be randomly different for some people. Every single one of these works to make the theory unfalsifiable. I don't think he intends this, but he has a very strong bias towards his theory and interprets all new data as fitting the theory.
I got that feeling too. Completely biased
Agreed. He loves to use big words to convert your average people into yes-men, typically those people who just do nothing but regurgitate whatever they find online from certain creators.
@@broccoli4487to a degree, do think it works, since the data founded is good, and it can and does match, but the idea that it changes per patch is kinda un-justified. (CCTM is pretty valid, since data matches:
7 days of testing, with each day doing a run of traces mats at 1:30 M a day, and estimating to be 80-90 4 star trace mats + another 7 days, half at 1:00 M and the other at 2-2:30, and it gives half as much, to close to 5%)
Theirs def validation to his claims:
CN of GI 3 years ago talked about a method like this, it just never went anywhere because people kinda gave up on the project. (Or it went underground, idk)
Also, he kinda kicks u out of the server after awhile, so I 100% understand the skepticism.
Such a good explanation for simple statistics, better than my classes at least. And a reaaly good critical analysis of simple prospective investigation. A nice way to get into statistics
Thx for making this video to save me the trouble of testing it. I figured it had to be BS, but I am not knowledgable enough about statistics to break things down well like you did.
Man this was a great refresh for my psychometry class. It's the first of your video that I watch but if not you should definitely make a series teaching statistics through gachagames. Homework made fun !
Dude please do more stats about hsr. This shit is super interesting holy moly
Thank you for making this video
bro rly went "if your data doesnt agree with my theory then its just a shit day for you" lol okay
literally lol or "random id" you just got randomly cursed
Not exactly. I guess we need to reword that. It was simply just a word we used to throw out there. The outcome is still the same if you think about it. Except the %s are flipped. That’s it like instead of the highest % being common for you then it would be the the bottem % that is common for you and the highest % on the sheet being least common for you. If you think it this way. All you would have to do is aim for the lower %s.
@@BigBoiPnoy I believe its a much safer assumption, that if your data is sound and a trend is established that "special UIDs" are just outliers statistically and nothing more as you have no way to prove they even exist, if the data collected isnt establishing a trend that paints a much worse picture that special UIDs are being used as a scapegoat to feel better about wasting your time trying to crack a complex algorithm with only 1 month of limited data
@@dankmeimes4274 on the recent updates to the sheet. The discord editors and moderators debunked SUID. It’s strictly something that happens during the time of day you are upgrading. Basically certain hours of the day have reverse outcomes that contradicts the overall data. But if a player waits 1-2 hours it matches data after. So basically with this knowledge, overall in a 24 hour period, there’s more cycles of one pattern over the other. And that’s why on data. No matter how much we record. It’s not a perfect 25% each line.
@@BigBoiPnoy this can just straight up come down to low sample size. Reroll 100000000 relics on 1 acc then we can start talking about statistical significance.
After going through college classes specifically for statistics in studies, I instantly saw all those flaws in the creators’ study lol
Hopefully a better made study and data collection will be done jn the future
It’s funny how I got a better understanding of the Goodness of fit concept from a gaming video than I did from my lecturer at University.
Same
So glad you mentioned the "special UID" thing.
It seems that every time the data doesn't distribute in the way that he wants it to he creates a new subset to explain it. All with no actual testing, of course.
I think its pretty clear that he's p-hacking.
This kind of reminds me of the Spotify shuffle thing. People were complaining that Spotify’s 100% randomized shuffle didn’t actually feel randomized as sometimes you would get the same two or more songs in a row. Spotify wound up making their shuffle not a true shuffle to make it *appear* more random. What if that is something like this?
It could be, but I don't think haphazard data test such as the one in the video would be able to prove that. I mean they don't even publicly place on their sheets what kind of significance they have, so the statistical rigor really isn't there
Standard alpha is .05, in class the lowest we’ve done is .01
I’ll consider doing this upon a bias sample. I did not expect to learn stats from a Stair Rail video lol
Way different from Tectone's foolish attitude. Thus the difference between those who are wise and those who are stupid is expressed.
Why is bro talking like he’s the main character
@@adogewithwifi1652they're not wrong though
proof?
@@adogewithwifi1652 Just finished an AP english class or smth
If you think that much about Tectone may be you should just admit that you love him
I actually was sort of willing to give this the benefit of the doubt... until I saw the bit about "special UID days". Creating a special "cheat" category that's specifically designed for data that doesn't fit your hypothesis and then asking your data collectors to self-classify is COMICALLY against every fundamental of good data gathering. There are ways to deal with outliers in statistical data, and this is aggressively not it.
Tbh, never knew about that from when I used to do Relic manipulation, but hearing about it now, makes it sound like a scam, ngl :/ (but tbh, kinda glad I did it, and got stuff from it:
My QQ is top 1% in the states, My firefly is top 5%, my Himeko is like top 20%, and my Fuxuan is like top 5%
It did help, and kinda glad for it.)
Hell yeah, we love to see real stats used out there in the wild
Good work, gotta imagine this took a decent chunk of time to process, lol
from my experience reseting string its really does work, i usualy got every substat that got up after using the reset things i didn't encounter that much anymore
I totally took 2 stats courses in university to max edging my HSR relics
I have little experience in statistic but this is insane... you went quite deep to actually analyze the bias theoretically... honestly too much variables to be able to "manipulate" and yet the probability is for each sub stat is mostly close for every sample... at the end this is "probability", tendency/bias could have been an accident...
result: just farm relics until you get a relic with atk crit rate crit damage and speed on it so you would roll everything a stat you would want
Your voice is soothing, very nice
Such thing was believed to be true in other gacha Fate Grand Order as well but not about relics, about gacha results that it's based on seed and the time when you roll
The fact that u need to take probability and statistics course to understand this is crazy. I hated those
Well done. Great explained!
So here's how the algorithm ACTUALLY works:
If you have three good substats, you get three to four upgrades on the one you don't want. You're welcome, problem solved.
I loved the video and agreed with pretty much the entirety of it ! Even with my limited statistics knowledge, I know how flawed the data collection methods used for the sheet are, among other issues.
However, one thing that I find compelling which keeps me from outright dismissing Bernard’s videos is the fact that on a lot of character artifact leaderboard websites, he does appear to be very highly ranked with many characters, competing with whales who constantly spend jades to refresh their stamina (which Bernard claims he doesn’t do). If he’s able to go toe to toe with such people by using his method, can you provide any insight as to how we could analyze if his rankings are significant, or just a result of random chance ?
To analyze his rankings requires quite a few assumptions; how many good relics does he get to roll on, what days does he do it on, how many good substats are upgrades per relic, etc. You can use a binomial distribution to calculate his probability, assuming various factors, but results significantly vary based on those assumptions.
Probability aside, it is never good to use anecdotal evidence to justify a method. For example, there are always people that win the lottery, despite the low probability. Does that mean that there is some method to solve the lottery? Most likely, no. This isn't the best comparison, but the point is even if you were to prove that the probability based on normal odds is low that he gets these relics, it doesn't prove his method works.
I see math, my brain turned off. All i remember is the word "bias"
Personal thought on this : you can see this is how the rng work but just add this detail of mine that they cant fully calculate the randomness which mean there are still some factors they missing to predict the roll cause in the end of the day this was made by group of individuals so there is a way to controll it or at least minimize the rng
It would be interesting if you could statistically say, it is more likely that the RNG is the same day-to-day and patch-to-patch, than that all of the other claims are true. They've got the data for old patches, I wonder if it's enough.
It would be doable if the sample sizes weren't so different. Because of their data collection method, certain days get far more samples than others. It would be like comparing apples to oranges at that point, I don't think you would be able to make any concrete conclusions.
Lmao, I usually dismiss such stuff immediately, but thanks for you effort, I really started laughing when you said that this guy invented"special UIDs" which ofc purely exists just to patch holes in his theory, to explain all the results that doesn't align.
Thanks for explaining. Where can we submit our Data? Sounds like an interesting study.
You can submit your data on Bernard's discord server, it can be found in the description of any of his videos
i’ve had multiple teachers (including a UC berkeley professor) try to explain basic statistics to me and it’s legit never clicked until now- i guess all i needed was a video game example LOL
If i have a relic that alrwady has crir rate cdmg, a flar hp and no 4th stat it's automaricalyl goin wholly into that awkward stat that doesnt work and why is it always def% or def flat.
I just finished a semester’s worth of homework in preparation for this video lul
I'm 100% on board with doing voodoo to pray for better substats but their method is just too involved for something you do as often as upgrading artifacts.
Overly elaborate rituals are reserved for banner pulls.
But are the sample normally distributed though?
This is a good video
Whatever is said about that "technique", I used it several times, and in 90% cases I could improve the substats I was either aiming to improve or not against improving. It is up to you what you believe in, and there is nothing wrong in believing in the bias of RNG :)
My one piece of advice is to always remember that people are notoriously bad at being able to tell what is and isn't rng. That's what I said before analyzing it myself and getting similar results to you
i didnt know this subject was a thing til now, and tho the theory seems very weak in supporting data, i wouldn't be surprised if a certain bias existed. not the way its presented but depending on the substat, cause im oretty sure theres a bias in rolling in genshin artifacts, wouldnt be surprised if something similar was on star rail
100% Agree the special uids are what irked me with this method
This is very impressive, appreciate the effort you put into this
My opinion from my understanding on how the rng in genshin/star rail is, is that it's not an rng client side but server side since the game requires you to be online to play all time, that it uses time (from the pulls we see hoyo record the time in which the wish are spent) and either uid or sid or a combination of both. Leaving us no chance to know how to abuse the rng.
Nah, I think it’s more like Bernard says, and that of RNG manipulation:
Randomness isn’t a true value, and the fact that data acts like this says theirs definitely a pattern, since while I don’t follow Bernard’s manipulation guides as often as I used to (most b/c it’s expensive as fuck.) they have value:
Cavern Corrosion manipulation theory states that are certain time intervals/times, certain probabilities are more often, and from my own testing, that’s pretty true. (A time around 1:30 on certain caverns give better substates, with times above that, like 0:49 or 0:30 being a lot worst, with 5/6 times, dmg sets giving 2 or more hp states, and at 1:30, 1/6)
Also, it’s server, with certain strings being more common around certain times than others. (I saw a dude get a 4 state upgrade orb, and 2 hours later from that post, I got a 4 state chest.)
Also, certain rolls are actually more frequent than others. (I get more high to low rolls on Fridays, Wednesdays just suck, and Saturday’s and Sundays are roughly average.)
Theirs 100% a pattern, it’s just that Bernard and everyone’s attempts to find it kinda just suck, and since testing cost so much, imma just do CCTM, get 5 god pieces, and just hope for the best :/ (since I did CCTM, I’ve quadrupled my god relic stashes, since they’ve became easier, but not the easiest, ngl ;-; )
Also, sometimes placebo for some people is just better, even if it doesn’t seem to work.
Ngl I had already believed in "good days/times" to roll artifacts in Genshin, so even though I was skeptical I gave his sheet a try. After a while I basically adapted my own "house" rules on how to roll and only loosely followed the data sheet and felt I was getting tangible results from it.
That said though, to be totally honest, I know it's confirmation bias copium to think the spreadsheet is useful in anyway. But it's legit just therapy for my monke brain to believe "UNLUCKYYY" when I don't hit my rolls so I still feel a net benefit in using it over not using it LMAO.
Love the video btw even though it did more mental dmg to me than my actual relic rolls loool
i tried it and it did the opposite almost everytime
special UID bro just try the opposite I guess
@@dankmeimes4274 when i did opposite it always worked normally 😭
I'm sure most of your analysis is correct and it's generally hard to believe in RNG manipulation until the evidence is pretty solid. One thing I'd mention though is that the RNG likely doesn't happen in C#. It's more likely that it happens on the server as doing it in the game would be way too insecure and easily exploitable. And it would also not be the first time a game used a completely artificial RNG system - e.g. some gacha games lowering their pull rates for the first day of a new patch, so making statistics and getting as accurate data as possible is definitely a good thing.
That's fair to say the game servers aren't necessarily coded in c#, though, c# is actually decent for backend applications. It isn't necessarily that far-fetched to assume the devs used the same language for both, though you could use a wide variety of languages for the backend, with users being none the wiser. Wonder if data mining has been able to reveal anything of the server calls.
There is no Santa Clause, there is no Easter bunny, and there is no RNG manipulation lmao, it’s been debunked by the Chinese fandom.
It's basically just pulling ritual. Except they actually believe it's real lol.
This is why relic manipulation is not popular. You guys judge it first not even trying it. Thats fine for us
@@toffiehacker3701 Gamblers in denial alert 🚨 The Chinese community debunked this time and time again, from “chest will respawn” myth to “RNG manipulation if you pull within this frame” to “manipulating relic stats” these are literally just survivorship bias + people only posting when they succeed + people preying on the insecurity of others. It’s just gambling.
@@toffiehacker3701 TLDR: It’s not popular because it’s horse sh*t, if you genuinely believe this then I strongly advice you to never step foot in any Casino.
@@StarsShatteredBeyond Unless they can count cards. But even if they can they still shouldn't because that's how you end up with broken knees in the alley out back.
When the manipulation of RNG also depends on RNG
I don’t really want to manipulate the stats of my relics at this point but the substats definitely is biased, getting a good substat is horrible compared to genshin impact and other gacha games and the reason idk but at least it makes me believe that there should be something behind
Nice video :)
How convenient that special uid's exist.
How conveninet that the pattern change every patch, so we cant use data from 1.0 to today.
How convenient "its all server side bro" so we cant find anything about it from gamefiles or leaks.
Its almost like there is a convenient way out to anything that could poke a hole in the theory.....
Well the server side is not a convenience is a security matter, long ago we would known the RNG formula so RNG manipulation would been a thing, the thing of special UID and changing the pattern sounds too exploitable to not be known publicly since before.
Wouldn’t say convenient but a extremely elaborated hypothesis.
Answer: No. There is no such thing as RNG manipulation in Gacha games. This should be obvious to anyone who understands that RNG manipulation would cost the developers tons of money, and as such would never be allowed to exist. If it did somehow sneak into a game without the devs noticing, it would be patched literally the same day that someone posted a video showing it was possible.
I expected a star rail video, instead got a statistical analysis course
It proves a bias, indeed. But one thing that differs this RNG from Pokemon Emerald is because the Emerald itself has been reverse engineered and it was found out to be a game which was ran by seed. I dont remember the exact number but if I'm not mistaken, it was running a seed from 0 to 65535, so you could entirely manipulate it and this is how speedrunners abuses it by manipulating the optimal time to start the game because the seed rolls are predetermined by the flow of time in the game.
However, its different on HSR tho, all testers are doomed for confirmation bias because there's no way in hell that Hoyo would actually show you the substat rolling chances from their dev tools. All we could do is setup a table and have a sampling size large enough for it to be concluded and even then, we are still not sure if that was the actual chances of the rolls. The study would be filled with bias because I'm pretty sure the narrative of getting HP/DEF for most of their relic rolls weren't just minority, its the majority who weren't even part of this study and a large chunk of that data has been left out which could contribute for the study to be concluded.
So I'd say that there is a predetermined weighted average stat rolls for each and every stat in the game, with HP and Def rolls being the highest. The 25% bs rolling chance should go away considering that the actual CV rolls proves it rarity in the game and numbers were tested in CN community where the each substat dont have an equal distribution chances for the relic to have 25% chance distribution for each substat.
As for its actual numbers:
docs(dot)google(dot)com/spreadsheets/u/0/d/1-MgfpwtN0PwR04eiqJp-PcaPB12oIkeaz-icEp-61b0/htmlview#
Just replace (dot) with . because youtube obviously doesn't allow links to be posted. That is the sample size conducted in the CN community.
Its all fun and games until a genuine math wizard shows up
I did try the relic manipulation thing, and it kinda worked for me? For some reason resetting the string by upgrading a piece with 3 sub stats into a 4 stats, kept giving me a second upgrade into the 3rd slot. I got 9 straight upgrades into a 3rd slot by doing in it and it was in accordance to the sheet.
It felt very impressive. I have no idea obviously if it works or not, however one thing that is a fact is that Bernard has multiple characters in the top 1% of better geared characters. That's either insane luck or something going on.
I do believe that the theory is a Frankenstein of ideas and there is no clear model, but among the trash some piece of truth might be there.
While I didn't fully understand the whole stuff but it was quite an interesting perspective
I remember when people found out about seed tracking in battle cats. That's one way to rlly kill a gacha game.
Whats funnier though is that his relic manipulation is based on days. Today is Monday? Its gonna roll to 1st and 3rd substats.
So, it reveals why speed is so rare for me to get
I used to play Genshin before and it has been sort of quite obvious that the RNG isn't "fair" from there so I never expected them to come fair as well in HSR. You farm and enhance thousands of relics and yet you only manage to use only a few of them. You salvage/feed more than you actually use. You get more DEF stats than actual speed stats. You don't get ER link rope as often as you get DEF/HP. It's a really bad practice and I don't condone it but I guess at the end of the day, it prolongs the game from stagnation since if people were to finish up farming so quickly on a really fair RNG, they would have nothing to do while waiting for the next patch.
5 months later, did anything change? since there's now more data being collected and Pnoy has admitted that the Special UID thing has been debunked, and now claims that the seed changes around a 15min~60min cycle
I don't really know how they "debunked" special UID. There isn't a good reason based on the data I have seen as to why they just rejected this theory (no attempt to plot a bimodal distribution).
As for the seed cycle changes, why would this be the case? Why wouldn't it change every second/millisecond, which is what would happen if timestamp is part of the seed?
Also, by increasing the amount of categories they have, they also need more data to characterize everything as a result, so I don't know if you can really get anything meaningful out of their data as of now. A lot of their analysis is haphazardly suggesting theories and throwing out others without sufficient statistical evidence to back it up. To me, it doesn't seem like much has changed, though I haven't taken a great look of detail as to what they are doing now.
Your voice and the way you talk is very similar to the elevator person in fnaf sister location!
2:47 I would expect that be the case for my 50/50, right now is 4 out of 13 😢
Remember guys, if its not 100%, its 50/50. Or 25/75 in this case...
I do find myself troubled in this case, most people i know who actually (wouldn't say bought) tried these method before are people with worse relic quality. Most of the time this is due to them throwing away relics that could potentially be good or being too much of a hoarder and not rolling any relic unless its the perfect stats or they feel like rolling. And to be frank, i checked the "phropet/manipulator/even cringier, elio" and his relic quality are the same as people who are used to mmo/gacha with rng stats system. I know you shouldnt insult the guy but at a first glance, dude basically looked like tandrew ate who cant even fool people with half a brain cell nor is benefiting from any of it
Well, the community recently said Special UID was debunked and instead probabilities change every set interval of time. And also what do you think of clear time theory? (specific clear times leads to dropping more desirable substats)
I don't know how they debunked special uid given that they never presented a full distribution (which they don't have enough data to properly make afaik). As for clear time theory, I think the same issues presented in this video still apply. Way too little data to describe their categories and why 30 second intervals? If your relic rolls are seeded based on time, why wouldn't it be to the second/millisecond? I also really don't get how they judge "desirable", because these stats vary per character (some want break effect, others want crit, etc).
I would also like to note that RNG algorithms aren't directly time related and depend on time if timestamp happens to be used as or part of the seed. There might be some kind of RNG algorithm out there that uses time outside of the seed that I am not aware of, but these are the only two ways for time to be utilized directly.
In summary, way too many questions with not many answers. I also just dislike the idea of presenting a hypothesis as a legitimate method when there isn't enough statistical support.
I wonder if people could join together too see stuff it would be fun
I trust myself. Using the strats i made 36 cv relic for myself, some other similar relics and some break effect relics.
Being a newer player i am focused on upgrading trails and only farmed relics at the start of TL60 and during events, so these are great for me.
hard cap
Math lad here:
Too lazy to do a publish study in a YT comments section. But based on a cursory fiddling with the sheet and it's data source. It does NOT provide enough significant OR reliable data to prove it's less than 5% variable range. It relies on publicly submitted data which means it's definitely tampered with either intentionally or unintentionally.
Meaning if anything. Your wasting your time.
Ok ok… I need to prepare for my economics exam… no need to remind me of that in an HSR video☹️
General rule of making something: Kiss. And for those aren't familiar it means "Keep It Simple, Stupid!".
Doubt they would do something much more complex than something like "rand() % 100" and comparing the output with a probability table.
Farming those pieces more than thinking about these manipulations, might be better for you to get your piece faster
Tried it 10 ten times 4/10 worked so i don't think this really works, this is not pseudo RNG, this is based on pure Arithmatic RNG
That's a RIDICULOUSLY small alpha. I can understand that you're trying to make a point here (the confirmation bias is the main issue here as well as the pruning) but the alpha chosen here definitely feels biased to try to further this argument
An actual RCT needs to be made for this to prove any statistical significance, not this confirmation bias garbage, very true
From what I recall, standard alpha cutoffs are 0.05, 0.01, and 0.001. There are 2 reasons I chose 0.001 as the alpha: what I stated in the video about being able to find seeds that produce an alpha lower than 0.01 relatively easily and reducing type I error.
Given the size of the sample sizes presented here, you could easily make hundreds of thousands of groups from the amount of relics people roll daily, in which you would expect results on the level of alpha = 0.001 simply due to how many groups you could get. Yes, it is strict for something like this, but if you want really strong evidence your theory works, you should get down to this level of certainty.
@@TarZBlammo I understand the reasoning, and I suppose this level of strictness is fitting for the small sample size of the data
It's just that the basis of the data, not your analysis, is a huge mess. Like you can be as rigorous as you want and can but biased data, especially to the evident case here, is truly unworkable. Like even if you did find the p value to be smaller than 0.001 it still wouldn't give us a good conclusion on the overall system is what I'm trying to say
Your video is informative and its effect is positive, I'm in no way trying to imply otherwise, but I feel like more stress was given to the hypothesis testing when, in my opinion, the result of it would at best raise speculation
It's like trying to extract statistics of a fair die using a loaded die. It just won't work out
There is no way this is real that would require the Hoyo devs to literally be insane. Any deviation from uniform randomness is confirmation bias. I trust Hoyo to make sure they don't give anyone an edge they're not paying for way more than these guys are reporting the data correctly and fully. Also it'd be insane for Hoyo to do anything but just a straight source of randomness independent of UID or anything else server side. Just have a mersenne twister initialized with a system source of randomness in a singleton and use that for everything this is a solved problem why do something that could go wrong or be exploited with more effort than that?
The bit about “special UIDs” just really irks me in particular. Idk if there’s a particular term for it, but it gives the vibes of someone being like, “If you tell me I’m wrong and provide evidence, then I’m not actually wrong because your evidence doesn’t count.” Like, stuff that contradicts your argument doesn’t get to be invalid just because you don’t want to be proven wrong.
SUID doesnt longer exist. Now the sheets say that you might try again in some minutes. Its based on time, not in some SUID
@@nehuenconde7139 legitimately, this feels like moving the goalposts. As far as I’m concerned, there was never such a thing as SUIDs, and the fact that the narrative has changed to “it needs to be at an exact time” just feels like even more bullcrap.
Legit feels like some faith healer bs pulling the ol' "if it didn't work, it's because you didn't believe enough"
@@johnnyjoestar3318 That's what I just said to another comment. I have nothing against the creator but I do feel like he's bringing in the ol' Filipino "haka-haka" crap we have in our country since he is Filipino himself. Basically, it's similar to superstition, such as "don't cut your nails on Friday or you'll get bad luck". Can't help but feel it's this with all the BS I'm seeing everywhere. "If your rolls didn't align to my theory, then it's just bad timing/days/luck"
This theory of RNG manipulation is inherently flawed because it is based on a premise that it is the stat line position that can be manipulated, rather than the stat itself. I am pretty sure Star Rail uses weighted substat RNG - I remember seeing something about that a long time ago, but I don't remember where I saw it anymore. Also worth considering that the Relic system is a semi-copy paste of Genshin Impact's Artifact system, which DOES have weighted substat RNG (crit stats roll WAY less often than atk/def/hp), makes the case for weighted RNG in Star Rail very likely. This becomes a problem since in two relics with SPD stats, one SPD stat may be on the 2nd line while the other SPD stat may be on the 3rd line. It would be registered then as a 2 and 3 respectively.
The weighted substats you remember is when the artifact is created at +0 what stats it has. This is talking about upgrading substats which in theory should have 25% chance on each line
Random video from youtube Alghorithm... but i loved it.