RNG is not random.

[kiztent]

One way or another, I don't think we (the player base) are going to be able to gather enough data to either prove or disprove streaks in the original number stream. Nor do I think that it really matters in the long run whether there ARE streaks in the original stream, because it's highly unlikely that any one person's sampling of that stream is even consistently the same distance apart in the stream (i.e. every 50th number), let alone having them actually get several calls in a row from that stream. The original number stream could be completely free of streaks (doubtful, but possible), and any given player could still see them due to their sampling of that stream.

So, how exactly does one test the randomness of a random number stream? Never mind that we are sampling a select portion of the stream, how do we know what we get is a random stream or not. What's the "measure" of randomness of a set of numbers?

[Jarak_akn]

Even in a zone with no players in there are still going to be some randoms going on in the background. (Things like weather changes etc.)

What I was trying to say about being solo in a zone was that while it wont actually make a stuff of difference, I may make you feal better because at least you know that no one is stealing your skill ups :-)

Jarak.

[Lothay]

So, how exactly does one test the randomness of a random number stream? Never mind that we are sampling a select portion of the stream, how do we know what we get is a random stream or not. What's the "measure" of randomness of a set of numbers?

Chaos Theroy anyone?

The strange oddites of math and science are truly more amazing than fiction. I'm not a mathematician. I don't even play one on TV. I've a passing education in physics and the sciences. I'm not an expert. But it seems to me that measuring randomness is pretty darn tricky.

As for the RNG in EQ; in order to get a "clean" version, you would need to be the only active set of events (no PC, NPC or scripted events) in a zone when it first came online, shortly after the number was seeded. After it has been up for a while, the measurement would be clouded by the presence of other events making your determination of causality and/or correlation difficult.

As for the tone of this discussion; hyperbole and idiomatic language aside, the tone and emotion level of this conversation is getting a wee-tad heated. Self-Moderate, or the big, bad closing-stick will be applied.

And as ever, thank you to Tanker for dropping by to add information to the discussion.

[Shjinn]

One thing people tend to ignore (and the first poster seems to miss this) is that all this is probabilty. That means that indeed everything in the probable range can actually happen. I sense an attitude that if something with a very small probability occurs, then something is broken because it is virtually impossible that the event can happen.

And to streakiness. What do you think is the chance of a "perfectly random" series? How about a 10101010101010101010101010101010 series? The chances of that exact distribution occuring is 1 on 4,294,967,296 (or 1 in 2,147,483,648 if you didn't care if the series started with a 0 or a 1). That is the same exact chance of having 32 0's or 32 1's in a row. If you want no "streaks", then your chances are exactly the same compared to having a perfect bad (or good) streak.

In other words, a "perfect distribution" of randomness is extremely improbable due to the nature of statistics and randomness. It's an oxymoron.

Shjinn

[Shjinn]

My last post in relation to tradeskills and 1 in 20 chance of getting skillup:

Using the same logic as my previous post on perfect streaks and randomness, the chances that you don't have a streak somewhere along the line (and in the case of skillups a BAD one) is very very slim. A perfect series of skillups (1 in 20 for say 190 to 250 or 60 skillups in a row) would be highly improbable. I don't think I've ever seen a report of anyone having a skillup exactly once every 20 attempts for more than 2 skillups. That being the case, the likelihood of having a bad streak of 100+ is high.

In the perfect world in which 190 to 250 takes exactly 1200 attempts and we can assume that the skillups will not be distributed evenly, the only conclusion we can make is that bad streaks will be almost impossible to avoid.

Shjinn

[Kieroth_whiteleaf]

Itek -

1) "one we found after a bit of research into fast/good RNGs."
Someone said that they NEEDED a very fast and very reliable RNG and probably had one of the best in the industry.
Oh, that was me.

Duh. And did anyone argue? I don't think anyone was going to say they intentionally got a slow obsolete one.


2) "Without going into the details of the algorithm (which I won't, for obvious exploitable reasons), I will say that there is only one stream from the RNG per Zone, i.e. all random numbers calls use the same generator that was seeded once, when the Zone was started."
Someone said the RNG had to be server (machine specific) and was only half right. It's machine specific with instance seeds.
Oops. My bad. But close enough for discussion purposes.

Again, common sense. If there was only one RNG it would get overloaded FAST, and clientside can't be done for exploit reasons. Duh.

3) "Random numbers are a large part of EQ, and lots of different messages require them in their handlers."
Who was it that said that EQ was totally dependant on random numbers?
Itek, again.

What RPG isn't? Going back to the first pen and paper games they're all random. Hell, monopoly and life are almost wholly based on random numbers (either dice or spins). The only ones that AREN'T are things like sports and arcade games.

4) "I'm not going to comment on streaks except to say that we don't cause them on purpose at the RNG level. I'm not intimately familiar with every system in the game, but I don't know of any that artificially creates streaks by "changing the odds" either."
Ok, there we go. It's official. Nothing to see here move along. (Really not being sarcastic for once. Thanks Tanker for finally debunking the "they are out to -fasten- us and I can prove it" myth.)

To the best of my knowledge, no one has ever seriously thought that the streakiness (IF it does indeed occur) is intentional. A few people had a loose theory (it's possible that...) that the streakiness was intentional to make things more interesting (ONE hard fight here, a good run there, etc.) but because other dev posts have implied that there isn't supposed to be streakiness it's filed under the "possible but doubtful" theory.

With all due respect to Tanker and co., even most of them say things like tanker's "I'm not intimately familiar with every system in the game", and I have yet to see a dev post saying "there is no streakiness (intentional or otherwise)". The closest things I've seen are Tanker's post which says that no known system does it on purpose, and Hartsman's post that the RNG works fine over the long term - neither fact which is in dispute.

Having also been following a similar thread on the eqforums page, I'm seeing a lot of other arguements which would lead me to strongly believe in favor of streakiness.

And Itek, keyword based encryption (of which the german code is a derivative) is very easily broken by hand given enough time (takes me about 2 hours) and if a computer is used for the frequency analysis instead of doing it by hand it only takes a few minutes of work to analyze the data the computer spits out.

Humans have better pattern recognition than computers - if they DIDN'T, then OCR and voice recognition technology would be flawless. Computers can't account for subtle variance in patterns the way a human can, because it isn't capable of interpreting data. Computers also have extreme difficulty in areas such as language translation and AI has a long way to go. *IF* computers were so good with patterns, then we wouldn't have things like pathing bugs in EQ because the computer would already KNOW "that's a wall, I can't walk through it" so paths wouldn't have to be hard-coded in the more complex zones. Collision wouldn't be such a problem either.

Try telling a computer "if this touches this, do this". Accounting for 3D space, the equation necessary for that function is over a screen line of text long JUST to establish if touching is occuring.

Tell your computer to block all incoming porn on your computer. Bet you can't. The best filtering software on the internet can't block all the porn without blocking something that you SHOULD have access to with only that filter in place. But a PERSON can look at ANY website and say "yup, that's porn" or "Nope, not porn" within a few seconds of seeing it.

Sounds like humans 3, computers 0.

Also, computer generated codes *AREN'T* random, if you want to get down to semantics. They're usually complex algorithms based on clock cycles, although other types occur. Due to the binary nature of a computer, it's not too good at anything other than "on" or "off". They're designed to give the appearance of randomness, and the best ones do the best jobs of appearing random.

>>First most people have under 1000 people they recognize by face.

You have got to be kidding me.

I've taught almost 1000 students and I'd recognize any of them. That's not counting the members of my family, my 300 co-workers (of whom I probably know 80% or more) and my friends and clients.

Anyone who went to a large high school and knew the bulk of the students there would recognize well over a thousand people. Mine had 1400 students, and while I didn't know them all, I knew the majority, and that's not counting the teaching and support staff. Or again family.

And again with the casino arguement:

If you're not going to read my whole post, don't attack it. I've quoted the relevant part:

Humans aren't bad at pattern recognition. A small percentage of them are so bad at it that they suck at gambling and pay for all those casinos. MOST people that gamble drop some money, occasionally come out ahead, usually lose, understand that the odds are in the houses' favor and gamble more for fun than any thinking that they're going to become rich off it. The fact that the next pull could just as easily be the one that wins (which is true, by the way) is enough for some of them to throw a few more quarters into the slots. EVENTUALLY, given enough time, SOMEONE is going to win a pull. That "the next one could be it" is the random possibility that keeps a lot of people going, we do it all the time in everything from gambling (this pull could be it) dating (this girl could be "the one") to camping mobs (next spawn could be the drop I need) to job hunting (maybe this place could be the one that takes me) to test driving cars (this could be the one I like) to sex (this time we might get pregnant) - even when the "odds" aren't truly random (like job interviews, first dates and test drives) the "next one could be it" tendency doesn't have anything to do with lack of pattern recognition, it's all about the fact that the next "thing" could be the "thing" we're waiting for.

Statistically, I know the probabilities and patterns, and I know I have virtually no chance of winning the lottery. But I still toss out a buck or two here and there to play once in a while, because *IF* I *DO* win, it's worth the buck. If not, I didn't lose anything I couldn't afford to lose.

On a roullette table, odds say that if you sit through a large number of spins (like say 50ish) and bet the same amount on each, you'll likely end up roughly where you started. But almost no one walks up the the roulette table with $50 and walks away with $50. Why? Because most people either stop when they get ahead OR they think they'll make even MORE money if they stick with it longer. And if they end up losing it all, no big deal because they brought the money in knowing they were risking it in the first place. The casino setup (like putting the cash out spots in the middle of the floor instead of the doors, or hiring hot wait staff) just helps them from a psychology standpoint, not inhibiting your pattern recognition.

If anything, it shows a high level of pattern recognition, because the casinos know exactly how to place things so the maximum number of people will spend the maximum amounts of cash at their establishment.

I know that the hot waitress at my table who is winking at me is PROBABLY doing it so she gets a better tip. She *MIGHT* be interested but I know she's probably not (because I recognize the pattern of cute girl + winking = higher tips). I'm STILL going to give her a slightly better tip and maybe try to get her number because despite my pattern recognition ability I can afford the extra buck (or few) and the potential gain is more than an offset (in my opinion) than the known loss.

Don't confuse pattern perception with acceptable loss.

And the number of people that actually suck at pattern recognition is large enough to be noticed, but small enough to fit well within the concept of standard deviation.

MOST people who gamble KNOW they're going to lose. But they don't care, because IF they win, they win enough to make it worth it, and IF they lose, it's an acceptable loss.

And I'd be willing to bet that far fewer than 20% of people in MOST areas (outside of places like vegas/atlantic city/etc) actually spend any appreciable amount of money gambling anyways (outside of the occasional binge vacation trip) - I'd venture to guess that it's closer to 5%. And anything under 5% generally falls under standard deviation.

Psychology is more a factor in casinos than pattern recognition.

Lastly:

To test your theory in actuality you would need to provide a reasonable sample size of /random 0 1 rolls with non-likely streaks. (Since they have spent the time, skull-sweat and money to make a good RNG I wish you luck in your quest and long life.)

*OR* I would need a large sample set (of roughly a billion) numbers broken into smaller chunks (like a thousand or so) that could be independently weighted and evaluated against the whole to check for non-likely streaks. Which I am reasonably sure I proposed in my first post.


Summing up my arguements for the skimmers:

1) People don't suck at pattern recognition
2) No one thinks (well, probably SOMEONE thinks, but they're in the minority) that the RNG is intentionally broken. Nor is it intentionally streaky (unless POSSIBLY - though doubtfully - in a subsystem used for some specific single issue that no current devs are aware of)
3) There is a significant amount of evidence to support an arguement that there is possibly a bug somewhere within the code which leads it to be streaky. Main points of debate in favor of this hypothesis seem to be:
a) Fizzle rates
b) Abnormally long skillup runs
c) Combat where individual mobs seem to parse for significantly higher dps over long durations than other (same) mobs of the same level/type/zone.
4) There is minimal arguement that OVERALL the RNG "evens out" over long term.

So far, from my perspective, I have seen a lot of good arguements to support streaking, almost no arguement to support non-streaking (ie, evidence or parsing, anecdotal or otherwise), and the Dev's have clarified certain points which don't seem to make a case for one side or the other. Which again I don't want to make it look like I'm attacking the Devs OR trying to make them sound bad - I'm very aware that there is a high volume of code and almost none (if any) of the original coders are there to interpret some of the older/more complex bits and trying to read someone else's code is like trying to translate a foreign language.

Is it broken? *I* *THINK* so. Based on *MY* interpretation of the *available* data, it would *appear* to be streaky.

Honestly, the only way that it could be COMPLETELY resolved one way or the other is to have a large sample set run and publicly posted so that people could do the data analysis independently and come to conclusions OR if one of the devs spoke up and said that it was actually messed up (because it'd be doubtful given the evidence that they'd be believed if they said it was fine). I don't know whether SOE could or would post that kind of data - *I* don't see any way it could be used for exploitation but I can definitely understand that it is POSSIBLE.

SFG

[Lothay]

And closed.

I said play nicely. It is quite clear at this point that most posters to this thread are not going to accept the position contray to their own. Rather than watch this debate get any more personal and accusatory, I'm shutting it down now.