RNG Taken Too Far!

Skeezy · January 31, 2019, 2:55pm

Touche. Although you realize thats all a big assumption. I know you prefer data but just bear with me.

The trick itslef I actually attempted to record. However updates to the game and increases to rates made it more difficult to spot and every attempt costs about $100-150. As 2-3 multi had to be done to spot the algorythem. Its main benefit was only for cashers who spend $ 250 or more. However even without data the trick allowed me to actively predict success. Meaning after it was spotted yes, I could actively predict my good pulls. Since I could accurately predict the cracked screen animation, that was proof enough for me. Not for you, I get it.

The farming benefit was so easy, obvious and beneficial for a player to do themselves, that I never felt a need to proof it. I figured those that tried it would see or they would quite frankly be like you and never try for themseleves because they are already set.

Iunderstand that you want data and there is some spread throughout my posts on gamefaqs. I did actually log some of my success. However the methods success was so highly obvious, that I felt if a player didnt try it, it was on them. I can’t fault your disbelief but all they literally had to do was try it. If not shrugs.

Garanwyn · January 31, 2019, 2:58pm

In a weird way, it’s actually more work. Java doesn’t have a built-in function that directly returns anthing other than ms. My guess would be that someone was using a library of convenience functions and just called the wrong one…

That said, the question of consequences is an interesting one. The real downside is it massively inflates the effective correlation in the aggregate results of the PRNG. Suppose you generate a seed that initiates at a 1-in-10M result. If you have 100,000 people logged in and active during that second, and mix on ms, then we might randomly expect 100 of them to get that result. Bad, but not awful. If we mix on seconds, 100,000 people simultaneously get a 1-in-10M result, and that skews the aggregate statistics badly. On average, during testing, you’d never notice it, because any single client follows the correlation statistics of a single instance of the PRNG.

Skeezy · January 31, 2019, 3:14pm

My main point to the board and fellow players is when it comes to rng don’t be close minded. Never think you know for sure, because people have found exploits in this similar rng.

Will a usable exploit/tendency be found in E&P not sure. Should we decide that there isnt based off face value? Most do. However there is always the possibility that they are wrong. Once you’ve seen more than one rng not be random, you know there is always the possibility. Do I know if we will find one…no.

Kerridoc · January 31, 2019, 3:37pm

Fair point. In statistics, we refer to a “null hypothesis”—that is, the neutral proposition that we will study to see if the data leads us to believe something else.

For example, many players believe that starting boards for raids are influenced by their choice of heroes. Bring lots of purple attackers, this theory goes, and you’ll get a deficit of purple tiles. This is a testable hypothesis. The null hypothesis is simple: all colors are equally likely. We then collect lots of data of consecutive boards and can apply standard statistics to see whether the data are inconsistent with that hypothesis. (Spoiler: actual boards are indistinguishable from random regardless of what heroes you bring; see Color Stacking Fairness Project)

Your proposition is similar: Summons results are not random but have some hot streaks and cold streaks that can somehow be expoloited. Where @Brobb, me and others have been struggling is to set that up into a testable form. The closest we’ve come is your suggestion that summons are better between :35 and :55 of each hour. That’s testable: do 1000 pulls within that 20-minute window and compare to the published odds. (Actually, we would first have figure out what power we need for the test, which will then determine the required sample size.)

You also suggest a “three-pull test”. I’m not quite sure what this is; do you do three summons and, if you get a 4* or 5*, quickly do ten more? Perhaps you could lay out a clear testable hypothesis here so we can try it out.

Also let me beg you to adopt the dictionary spelling of algorithm.

Brobb · January 31, 2019, 4:14pm

If it was “easy and obvious”, surely it would have been very simple to collect a bit of data to test it, no? But you didn’t do that? Why not? Because it was so “easy and obvious”? Hmm.

Usually the easier and more obvious something is, the easier it is to collect data on it, and the more obvious it is that the data shows that it works. But not in this case, huh?

Yup, this . Because if you can’t test it, then what are you basing your conclusions on? How you feel? What is “obvious”? Many things that people feel are obvious turn out to be false: that’s why we test.

Brobb · January 31, 2019, 4:26pm

That’s interesting. Would that effect be something that should concern us, I wonder? Individual players’ results would still be effectively random, but players’ experiences would be relatively highly correlated. I can’t think of any reason that would necessarily be a problem, but there might be something I’m not seeing.

If a large enough group of players logged their data then I suppose that might reveal the correlation. Maybe they’d feel sad about it?

Garanwyn · January 31, 2019, 5:00pm

That one made my day! You’ve got the best sense of humor.

In general, it’s a problem for the MMO manufacturer’s pocketbook, not a problem for individual players. One might care in the sense that it likely inflates the number of very rare xxx in the game, and if one didn’t happen to get that xxx, one’s disadvantage is magnified.

But we can feel fairly sure SG doesn’t do this. Titan loot is the easy counter instance. It’s all distributed at the same time. Even accounting for some people being on different servers, 10 people in alliance chat all going: “Wow! I just got a Damascus blade!” might, just possibly, get remarked upon in the forum.

Skeezy · January 31, 2019, 6:31pm

Well 3 is just a chosen number.

Most of my efforts back then and now, are ultimately to find the least costly and most effective way to summon. That is why I think of my “methods” as not exploiting but more about avoidance. Some games I dont believe singles are a good way to go, because the rotation of a good seed doesnt come around often enough to ever be counted on. Many games(especially older) that use loot boxes are like this where you could do singles for multiple days or even the length of an event and never hit the seed for desired item. Bulk opening is usually the best way to try to get whats desired. However its a complete gamble and will get costly no matter how you go about it.

I’ve had much success with sparratic groupings of singles on this kind of rng summon. Spread out singles will actually get you a super super rare(dokkan) or 5* somewhat frequently (not months apart) . In comparison to multis and how often they can be dead (though the bonus chance on this game is a nice touch) . We all know pseudo random can easily give you multiple dead multi’s in a row which is an expensive and for most players frustrating loss. In pacing yourself your more likely to catch a decent seed for less money or at least thats the theory and it seems to be pretty effective.

3 just happened to be the number I went with, for one because of costs, I feel like most players would be able to do a couple sets of 3. And in the previous game 3-5 seemed to be effective enough to be beneficial. For example the Event characters are usually lower percentage to get than normal 5*. However its all based on seeding so if your ever seeded to even get an event character early on, your more likely to hit it rather than drain all your gems in mere minutes. To me bulk really is this rng’s highest form of a gamble unless overall rates are high, like well above 5%… Single groupings does still leave room for failure but less so in that the spending is controlled and you have more opportunities to catch different seeds if you really spread it out. In theory/practice it does well in comparison to 1 sitting multi’s which do tend to yield results but easily get very expensive if you dont get a good result.

For me this has worked well but is a less solid than what is mentioned in previous posts because of the lack of tracked data. Unlike those this isn’t obvious and needs testing from someone besides myself.

People have tested singles vs multis but usually not well. They do a test of however many and calculate an average. To me that kind of averaging misses any ques, if any . They lack 2 main ingredients, multiple sittings and a decent amount of time to make a complete observation. Streak tendencies are usually also not observed.

Edit:after brobbs post

@brobb but does it make sense?

In the games I did play that invloved loot boxes success were announced in world chat so I could actually knew when any player would get the desired item and I could inquire how many boxes it took wether they got it 1st try etc. Your assumption is that I dont do research or track anything. I just dont do it like you do it. Mine is actually effective because Im not just averaging everything.

In your anti-bias your telling me you’ve never pulled or witnessed event cards being pulled in 1st 3 singles? I havent even been playing that long and I have. Also have seen other players exclaim and make videos of it. However any even small shred of evidence will likely be chalked up to randomness even though you know its about seeding.

Your also acting like what Im saying is untrue. Is it? Can you not do multiple multis in a row and get no 5*? Is that even uncommon? Are mutli’s not expensive? Do testers usually consistently track streak tendencies across multiple sittings? Or do they just do a number of pulls and look at the average like yourself? In looking at averages of numbers pulled how would you spot any tendencies if any? Sorry that was me being a smartass.

Our debate has reached its platue. Can we agree to disagree without snide remarks? Cause I’ve kind of been there done that and I know there is no getting through to you and vice versa.

However I will leave you with one last jab. You would be correct if the game tracked your pulls in order to decide when you are successful but you dont think that and neither do I. We both know seed values determine your luck. And in all common sense knowledge what is better? one chance or multiple chances?

.

Again needs proper testing though.

Brobb · January 31, 2019, 6:47pm

Evidence? None? Okay.

We agree about this completely.

Brobb · January 31, 2019, 7:48pm

I try not to be a fascist about spelling and grammar, but I’m sorry: I can’t cope with much more of this.

Sporadic, not “sparratic”.
Cues, not “ques”.
Plateau, not “platue”.

If you mean summoning an HOTM from one of your first three single pulls, then yes, I have heard of it. I’ve even done it myself - two or three times, I think. Is this because I have a magical system? No. It’s just dumb luck, and we’d expect to see some dumb luck.

Do you find this surprising? Why? We know that there is a 1.3% chance of pulling an HOTM, so if you summon 3 times, you’ve got about a 4% chance of drawing one or more HOTM. So if all the players in your alliance summoned three times, we’d expect that one would probably pull an HOTM (on average). Would that tell us that they have struck a favourable algorithm and should summon more? We have no reason to suppose so.

Of course you can, and of course it’s not uncommon. With a 1.3% chance of drawing an HOTM from a summons, if you summon 30 times then the chance of drawing one or more HOTM is less than one in three - so if everyone in your alliance summoned 30 times, the chances are that most people (about two thirds, on average) would draw no HOTM. Would this tell us that they had struck an unfavourable algorithm and should stop summoning? We have no reason to suppose so.

No. They’re considerably cheaper than singles, on a per summons basis.

Testers look at averages, because no basis has ever been suggested for segregating results by day or time or moon phase or relative humidity or months with and without an ‘r’ in them. Without such a basis, we’d just be data mining if we were to chop our data around clusters of success - and because we’re humans, we’ll always imagine that patterns exist when we’re really just seeing the streaks that typify randomness.

Your suggestion that summons executed between *.35 and *.55 might be more fruitful than at other times is the first specific time-based proposal I’ve read on this forum . We can collect some data on it and test for significance. I find it quite absurd (you’re just doing the human thing and imagining that a pattern undergirds your own clusters of success) but I’m always happy to be involved in testing the absurd - my 5-10 summons over the next month will be executed at *.45. I hope that the paying players among us can contribute more substantially, so that we can collect five hundred or a thousand observations, build a nice little confidence interval and see whether 1.3% falls within or without it.

This is such utter gibberish that I can’t even tell the point you are trying to make. Do you think that ten single pulls somehow generate a better chance of summoning an HOTM than one 10x pull? Have you never been taught any basic statistics?

Gryphonknight · January 31, 2019, 7:59pm

Hi guys.

This is fascinating but maybe slow down and cool off before the topic starts discussing Nazis.

Brobb · January 31, 2019, 8:03pm

Godwin is watching, waiting…

Edit: Point of order - haven’t you technically lost by being the first to bring jackboots to the conversation?

Skeezy · January 31, 2019, 8:12pm

And you say all of this knowing seed values determine luck.

Thanks for the spellcheck
You say all of that knowing seed value would be what determines if you got desired card at all 1st, last, middle etc.
And actually yes, that was exactly how it worked on dokkan. As stated I could go cash again and have next to no dead multi’s. Vs the low end when most of your multis are dead. What dont you get? Never said any of that about E&P only dokkan.
Multis force you to use 3k+ gems at once. 1 multi all 10 cards are pulled under the same seed value. Same 10 can be pulled with different seed values using singles. Youd be better off using your powers of averaging to determine how often players get event card 1st try. I mean if you really wanted to use your averages to help determine seeding luck.
See #3 again. And basic? Yes.

Brobb · January 31, 2019, 8:17pm

You’re welcome.
You seem to think that seed value alone determines outcomes, which it does not. I don’t know what “dokken” is and have no interest in knowing. I like E&P - that’s why I’m here at the E&P forum.

Evidence? None? Okay.

1.3% of the time. Didn’t even need my calculator.

Skeezy · January 31, 2019, 8:22pm

Interesting.

So even not knowing or seeing for yourself, your convinced I’m wrong.

3.touche. Correct. There is no way to confirm this although many players believe in summons such as this cards are determined as soon as you hit the button. Ive even debated that myself because I have found flaws in that logic as well.

genuwinely interested in how you got that average. And I will tell you why its flawed

Brobb · January 31, 2019, 8:29pm

Nope. But I asked repeatedly and persistently for the evidence you collected to test your theory, cheekily suggested that perhaps you didn’t collect such evidence, and was eventually gifted your admission that, indeed, you didn’t bother to collect any evidence at all.

So it’s not that I’m convinced you’re wrong about anything, it’s that I’m convinced you have no way at all of knowing whether you were right or wrong. Hence, me neither.

Skeezy · January 31, 2019, 8:34pm

Evidence is on gamefaqs. Thats all I can say the game now gaurantee’s a high card every multi so…yeah. The only evidence is in the board itself and must be searched. Im good…you do it.

3.in the brashness of averages you have solved nothing. Im looking at an average that could be achieved in a mutitude of ways with no clue as to how its actually achieved or data tracking players attampts across multiple sittings and seeds. Seeds are not the only thing that determined luck but they are a large part of it.

Lastly fully research rng manipulation please. At your liesure that way you can find a legit programmer site past all the pokemon mess. If I stumble across a link to one of the ones I know is good, I will post it for you. It will help understand way more than you do now.

Garanwyn · January 31, 2019, 8:38pm

Without being able to comment on Dokkan, since I’ve never played it, let me try to help explain the difficulty we’re all having with some of your points:

Only the worst PRNG in the world are so highly correlated that successive draws will produce similar results with any regularity. Even using a standard bad PRNG like a Linear Congruential Generator (think: the rand() function in C/C++) won’t do that. It exhibits spectral correlation, but not instantaneous correlation in the way you’re describing.

So we’re all having trouble understanding why getting 3 good pulls would imply the 4th, 5th, etc would also be good. The PRNG design would have to be critically bad to permit this.

If you don’t keep very good records of attempts vs. outcomes, it’s really easy to fool yourself on how well or poorly you’ve actually done.

You’ve said a couple of times that you spent tens of thousands of dollars on Dokkan. I don’t know the odds in that game, but I do know that’s a LOT of money. Players in this game who have spent that much generally have all the HOTM, as well as the best event and seasonal heroes. No special luck required.

It would be very easy to combine large expenditures with even very average luck and come away with the belief that there were a genuine pattern underlying your results. Gamblers do this all the time, and that’s what @Brobb is talking about with people who try to time roulette wheels, etc.

Statistics gives us some very good tools for testing whether a dataset matches claimed probabilities, whether it contains correlations, whether hidden causal factors exist, and so on. But without a data set, without applying those tools, all you’re really doing is guessing.

And unless you kept counts of attempts vs. successes, you don’t even know if your true results were better than, worse than, or the same as published probabiliites. Humans are very good at convincing themselves that they are lucky (or unlucky) despite evidence to the contrary.

Note: I’m not saying Dokkan didn’t have exploitable patterns. How could I? I have no data from which to even venture a guess. But I am saying that proving any game has exploitable patterns requires data.

Brobb · January 31, 2019, 8:56pm

If you want to discuss your past experiences, claiming they have some significance, then I’m happy to do so if you can back yourself up. If you expect us to do your research for you then it’s going to be hard to take you seriously.

Let me share what the devs have told us, and which no analysis to date has been able to disprove. Each summons is an independent event with a 1.3% chance of drawing an HOTM.

So our unadjusted Bayesian prior is 1.3%, and although we can’t be certain it is correct, quite a lot of time has now passed without anyone being able to disprove it. It would be pretty hard for us to differentiate between a true probability of 0.5% and 3%, of course, but we certainly have no evidence that individual summons are not independent.

If you want to discuss this stuff, then by all means come here prepared to discuss it. Asking forum participants to do your research for you is not really cricket. (And what I have previously read about rng manipulation suggests that its significance is utterly overblown by people who just don’t understand randomness. Urban myths about people exploiting PRNG generators are usually not verifiable.)

Garanwyn · January 31, 2019, 8:59pm

There’s very strong evidence that it’s doable (and has been done) on console games. Although “manipulation” is a silly word to use for what people have actually done. “Exploitation” would be more accurate. Having total physical control of the platform is a beautiful thing

I’ve never seen any data for PRNG manipulation on MMOs that wasn’t very, very equivocal