Salvage is BAD for skillups

[KyrosKrane]

At last year's Fan Faire, Tanker implied that all combines with trivial 15 or less are automatically no fail.

[Boleslav]

Hijack alert, sorry…

Bolas wrote:
"Have several people either flip a coin 100 times or write down what they think might happen if they flip a coin 100 times.
You'll be able to tell who actually flipped the coin and who made up the list of results. Why is that?
Somewhere in that 100 you'll have either 6 tails in a row or 6 heads in a row if it was an actual list generated by flipping a coin."

I'm a sucker for math truisms like this one.

I figure the odds of not getting at least one stretch of 6 heads or 6 tails in a row in a run of 100 flips to be about 22%. Here's how I figured it.

The probability of getting the same result six times in a row for your first 6 flips is .03125, or .5 to the 5th power. It is to the 5th power because it doesn't matter what the first result is. Starting on the 7th flip, the chance to finish a stretch of 6 in a row is .015625 (.5 to the 6th power). This has 94 opportunities to happen, as each flip starting on the 6th flip could be the last in a stretch of 6 in a row. The probability of it not happening for all of those 95 opportunities is (1-.03125) * ((1-.015625) to the 95th power), or about .2204, yielding a 22% chance of not getting a stretch of 6 in a row (either heads or tails).

I created a repeatable setup on Excel to generate 100 flips of a coin (0 or 1) and ran it 84 times. 69 times there was at least one run of 6, and 15 times there was not (18%), not too far from what I figured.

Not as good a test as you thought. Streaks of 5 would be better, but still not telling as you would expect to not get a streak of 5 about 5% of the time, but you would expect at least one streak of 4 over 99% of the time.

You can now return to your normal discussion.

Boleslav Forgehammer
Paladin of Brell in his 67th Campaign
E'ci – Sacred Destiny

[Neebat]

Hijack alert, sorry…

No... I don't think it's a hijack. This thread was started by someone wanting to talk about the impact of streakiness in the random-generator. Your analysis and testing is fairly relevant, because it shows that streakiness happens, maybe not as often as someone would think, but more often than most.

We still of course don't know if the EQ RNG is streakier than pure randomness would dictate.

[Phantron]

I'd be really surprised if EQ didn't use the same random number generator algorithm that virtually every game and probably 95% of all programs use. There is simply no need for anything more complicated than something like the C rand() call for most application of randomness.

[Palarran]

Well, it depends how quickly random numbers need to be generated. I wouldn't be surprised if two random number generators were used, rand() and something that is quicker but less random when the normal RNG can't keep up. EQ servers do need to generate an unusually high number of random numbers...

[Itek]

ARRRGGGHHHHH!!!!

I simply can not NOT post... I tried I really REALLY tired not to....

Everyone takes ONE step back and a deep breath now....

Ok I've said this before and I guess no one noticed or they weren't here at the time.

EVERY SINGLE THING IN EVERQUEST DEPENDS ON THE RANDOM NUMBER GENERATOR!

Click on "create a new character" button and it randomly determines your race, class and even gender. The name generator uses random. Combat uses random. Drops are random. Spawns are random. If it wouldn't take too much resources and make players whine mob pathing for roamers would be random. RAIN, for pity's sake, is stinking RANDOM!!

Their ENTIRE BUSINESS, and it IS a business, depends UTTERLY on their RNG being the best it can be.

If I hear ONE MORE comment on "streaky-ness" I swear to the Bright Lord Agnostic I'll pop a blood vessel and stroke out.

Again....

HUMANS ARE INCREDIBLY __BAD__ AT DETECTING PATTERNS!! THEY _SEE_ PATTERNS WHERE _NONE_ EXIST AND _FAIL_UTTERLY_ TO SEE OTHERS!! ........ LAS VEGAS IS BUILT ON THE FALSE BELIEF THAT "STREAKS" EXIST.

It's called "the Law of Independant Trials" and it ALWAYS works. I nearly commented when the first flaunting of this MATHEMATICAL LAW was posted. But now I have to.

"if the roll is above 50 you expect the next roll to be below 50"

I call BULL-HONKEY!! No, you do NOT expect ANYTHING because the second roll is COMPLETELY independant of the previous roll. (And the NEXT roll for that matter.)

PROOF: Reduce the size from 1 to 100 to 1 to 2. Flipping a coin you CANNOT say "the last flip was heads, the next one should be tails"

"You expect, in a truely generated rather than made up list, 6 in a row in a sample"

YES. Expectation -here- is based on the odds being VERY HIGH (nearly 4 in 5 trials) of such a thing happening. If you did NOT get a "streak" of 6 you would have an UNUSUAL sample set.

Neebat stated the two methods for determineing true randomness.

1) Even distribution.

There is a Tanker (I believe) quote where they ran a BILLION sample size trial. And got PERFECT distribution. Which is actually UN-expected. The quote (which I think the search engine could find) stated the developer thought the algorithm used was TOO perfect.

2) Unpredictable.

If someone, outside SOE, can predict the next random call result, without hacking, my hat's off to them.

It's a computer-generated-pseudo-Random-Number-Generator. It's not perfect, it cannot BE perfect.

BUT, and I've said -this- before too....

IT'S THE ONLY p-RNG WE HAVE AND WE ARE STUCK WITH IT!!

Unless, and until, one of us gets a job at SOE to make a better one.

Now to accusations of deliberate malfeasance. (for the non-legal that means "people abusing their jobs")

a) Without proof such alegations are 'per se' defamation, and legally actionable. (meaning: if you can't back up what you said, you can get sued)

b) Programmer = God-HimSelf-Almighty. If the 'hairy-badge-of-Berrtox' combine would normally have a trival of 108 and a min-failure rate of 5% but people with skill 300 would expect 5-[(300-108)/40] failure rate BUT the PROGRAMMER sets the failure rate at 12% .... TOUGH. That's the way it IS. Back on the old MUDs if a senior-enough Wizard toggled your gender... you're a "neuter" now.

c) I read the posts implied in the "a mod flamed me" thread. No, the mod did NOT flame you, they told you to follow the RULES and post support for your statement. Rather than do so you edited your previously SLANDEROUS comments to be mearly baselessly accusatory and tried to make yourself look like the victim.

"Salvage .... blah blah blah ... I see your lips moving but .... blah blah blah...."

Remind me exactly WHAT your alternative is?

"Salvage took my good card out of the shoe of the RNG." That's what the original poster was saying. That because Salvage pulls a random number it prevented him from getting good numbers on the combine in the first place.

I - The 'combine' success number is called FIRST. It's independant of the Salvage call.

II - There is sufficient time between each click of 'combine' that the RNG has cycled a few million times, therefore making any 'streak' in the numbers output by the algorithm itself MEANINGLESS.

III - In America in Blackjack the dealer finishes dealing himself the second card before going around the table for hit/stand player decisions. In Europe the dealer WAITS until after all player actions and THEN deals his second card to himself. DEEP and PENETRATING analysis of both forms have PROVEN that there is NO SIGNIFICANT DIFFERENCE between the two methods.

Games theory ladies and gentlemen. The people at SOE do it for a living. I've been doing it for over 20 years. (Not for a living, yet. Need to finish that piece of paper.)

"It depends on how quickly random numbers need to be generated..."
"when the normal RNG can't keep up..."

Palarran the computer generates RNGs faster than most non-technical people can comprehend. A giga-hertz processor can preform, literally, billions of calculations a second. Presume that a server needs to preform a thousand operations for a single RNG. (which would make it just about the most HORRIBLE p-RNG ever) Now let's presume that 99% of the processor is loaded before we can get to RNG operations. (At which point one would PRESUME that professional IT people would be BEGGING for an upgrade. 99% load is potentially disasterous.)

A billion operations a second. We only get 1 percent of that so 10 million operations for RNG. But each RNG takes 1000 operations. That's 10,000 RNGs a second. .... At least....

Don't think we are gonna run out ANY time soon...

[Zanaspus]

Listen to Itek. He knoweth what he speaketh.

[Boleslav]

Nope Neebat, this time it was a hijack. I was examining Bolas's assertion that it would be a tip off that a series of 100 coin flips was not sufficiently random if it did not contain at least one streak of 6 in a row. I was suspicious of the comment when I read it, and had to investigate it to satisfy my own curiosity.

Then I posted it here for Bolas's reference, mostly because I didn't like the condescending tone of his post about most people not having enough math background to understand an RNG. That was unnecessary, and mean spirited of me. L Had nothing to do about the RNG, Salvage, or anything.

Itek: I got stressed out just reading your post. I can imagine how irritated you must have been when you wrote it. I hope posting it had a bit of a calming effect for you. It is frustrating to have the same discussion again, and again, and again though. Unfortunately, nothing any of us write today is going to stop the next iteration of this topic in a month or so.

Anyway, I agree with the probabilistic and processing ideas behind Itek's post even if I don't share his level of angst about it. I don't get so irritated with folks asserting streaky-ness, which is why I occasionally suggest some tests and what not, hoping that people would look at things from a different perspective when some elementary analysis was provided.

Boleslav Forgehammer
Paladin of Brell in his 67th Campaign
Tunare (E'ci) – Sacred Destiny

[Phantron]

I remember reading an experiment where a professor had his students roll a 20 sided dice 100 times and record it, and other half of the class just make up 100 rolls. In all but one of the cases the fake rolls have no repeat rolls (e.g. 2 followed by a 2) and those are all detected because you expect a 0.6% chance statistically to not get a repeat roll after 100 tries. This I think illustrates the fact that people think 'no pattern at all' must mean there can't be streaks. Probabilities predicts the existence of streaks given large enough number. You can find your birthday in the digits of pi even though there whatever pattern (if any) responsible for the digits of pi certainly isn't remotely comprehensible to humans.

It's also a good way to tell the difference between an ununusal event and finding something fishy. If I rolled a 20-sided dice 100 times and get no repeat rolls, it's unlikely but there's still a 0.6% chance of happening. Now when you've half of the class (whatever that is, let's say 10) claim to all get a 0.6% chance event happening, then you should question if it's truly random since the chance of that happening is 0.6%^10 = something a lot lower than the chance of winning the lottery.

[Neebat]

ARRRGGGHHHHH!!!!

Chill. I for one don't really care if the RNG is particularly streaky or not. I'm curious, but not worried. It can't be good for your blood pressure to get this excited about it.

Click on "create a new character" button and it randomly determines your race, class and even gender. The name generator uses random.

Careful there. Those are almost certainly randomized client-side and therefore using a completely different RNG. The client-side random operations aren't going to impact tradeskill out-comes. Especially when we've been told there's one RNG per zone and these things happen outside of any zone where anyone can tradeskill.

If it wouldn't take too much resources and make players whine mob pathing for roamers would be random.

Actually, I believe mob pathing IS random in many cases. I'm not totally sure, but I believe some roaming mobs have a chance to turn back, turn aside on another path, etc. Of everything you listed, this is the only thing that would use the RNG outside of player control in an otherwise unused city zone.

*EDIT* I forgot about NPC spell casters. Pretty sure they'll draw a random number to decide what to do when standing still. *EDIT*

Their ENTIRE BUSINESS, and it IS a business, depends UTTERLY on their RNG being the best it can be.

Nah. It depends primarilly on the quality of the box art. As much as it hurts an engineer to say this, the psychology of selling games has little to do with the quality of engineering. But we're not here to discuss the business model.

I could present a good case for players receiving a more interesting game world if in fact the RNG were sometimes extremely, extra streaky. As an example, consider a stream of melee hits. If you hit something long enough, a *good* random generator makes the total damage extremely predictable and the combat system dull. A streakier generator would provide more variety in the outcome for long encounters. A good RNG gives you failed hits and exceptional hits, but tedious encounters. An RNG with more serial correlation gives you exceptional encounters.

If I hear ONE MORE comment on "streaky-ness" I swear to the Bright Lord Agnostic I'll pop a blood vessel and stroke out.

From what I can see, this has been (until now) a fairly level-headed consideration of the possibility that there might be unusually high serial correlation within the EQ server RNG. I'd love to know what troubles you so much about the idea.

HUMANS ARE INCREDIBLY __BAD__ AT DETECTING PATTERNS!! THEY _SEE_ PATTERNS WHERE _NONE_ EXIST AND _FAIL_UTTERLY_ TO SEE OTHERS!! ........ LAS VEGAS IS BUILT ON THE FALSE BELIEF THAT "STREAKS" EXIST.

This has been said several times and needs to be repeated. The reality is that "streaks" do exist in many RNG's. I'm don't know the status of streakiness in EQ, but I know it gets exagerated. Just because it has less of a pattern than most people think doesn't mean there's no pattern at all.

It's called "the Law of Independant Trials" and it ALWAYS works.

In a true random system, that's absolutely true. One of the requirements underlying that law is that you start from the same state every time, and with computer RNG's, that's never true. (Because they are rarely reseeded.) Some of us are curious how close we are to ideal behavior that would follow that law.

"if the roll is above 50 you expect the next roll to be below 50"

I call BULL-HONKEY!! No, you do NOT expect ANYTHING because the second roll is COMPLETELY independant of the previous roll. (And the NEXT roll for that matter.)

Careful with pronouns. The author of your quote intended to talk about what "The Average Joe" expects. In fact, the average joe has some very odd notions about randomness. He was actually making the same point you are.

Neebat stated the two methods for determineing true randomness.

I'd say they're methods for detecting non-randomness. I linked in a tool above for testing a random stream and it mentions them both.

There is a Tanker (I believe) quote where they ran a BILLION sample size trial. And got PERFECT distribution. Which is actually UN-expected. The quote (which I think the search engine could find) stated the developer thought the algorithm used was TOO perfect.

Yep. I believe EQ's RNG has been well tested on this and I said so. I gave it merely as an example of what we know has been tested to compare that with what we're not sure about.

If someone, outside SOE, can predict the next random call result, without hacking, my hat's off to them.

My hat's off too if someone can pull this off. The original poster seemed to assume that he could make predictions, but I don't really know if that's true or not. He also assumed it only worked to his disadvantage. That is what started this whole thread. If you COULD see that you're on a streak, then by all means, skip it.

Regardless of whether or not someone could make statistical predictions based on the past behavior of the RNG, it is still fun, (for me) to talk about both the impact this would have and the ways to rule it out.

I disagree with the original poster on both the assumption of predictibility / serial correlation / "streakiness" AND the impact it would have on the game.

IT'S THE ONLY p-RNG WE HAVE AND WE ARE STUCK WITH IT!!

Unless, and until, one of us gets a job at SOE to make a better one.

Yep. That's why there are folks interested in understanding it better.

Now to accusations of deliberate malfeasance. (for the non-legal that means "people abusing their jobs")

Where? I haven't seen this anywhere. The guys at Sony really seem genuinely concerned with enhancing our experience, not making us miserable.

b) Programmer = God-HimSelf-Almighty. If the 'hairy-badge-of-Berrtox' combine would normally have a trival of 108 and a min-failure rate of 5% but people with skill 300 would expect 5-[(300-108)/40] failure rate BUT the PROGRAMMER sets the failure rate at 12% .... TOUGH. That's the way it IS. Back on the old MUDs if a senior-enough Wizard toggled your gender... you're a "neuter" now.

Yep. We're in their world. I like it. I seek to understand it.

c) I read the posts implied in the "a mod flamed me" thread. No, the mod did NOT flame you, they told you to follow the RULES and post support for your statement. Rather than do so you edited your previously SLANDEROUS comments to be mearly baselessly accusatory and tried to make yourself look like the victim.

Before you call it slander, consider the fact that my statements were true. I was assuming the reader was familiar with the posts already made here on the board. In my revision merely suggested that the mechanism might be necessary eventually. (Which is pretty clear now, since it has actually been implemented.) I wasn't accusing anyone of anything.

II - There is sufficient time between each click of 'combine' that the RNG has cycled a few million times, therefore making any 'streak' in the numbers output by the algorithm itself MEANINGLESS.

I've seen this stated several times. I think it's hyperbole, but I tend to agree with the sentiment. The RNG normally has a heavy workout in the 2.5 seconds between combines. In an otherwise unused zone, there is no combat to draw on the RNG, there are no mobs dying and respawning. There's a few hundred NPC's, generally not doing much at all. I would tend to agree that even that much activity is enough to make tradeskill combines less than 50% of the RNG activity.

Don't think we are gonna run out ANY time soon...

Regardless, in CS there are FAST RNG's and slow RNG's. I'm willing to assume Sony's using a fast one and leave it at that.

[Phantron]

One melee round (dual wield) has at least 20+ RNG checks. Just this activity alone is easily 10 times the number of RNG usage than tradeskills. I'd assume the RNGs are done on server-side otherwise there would be hacks that lets you tell your cilent that you always rolled a success.

[Neebat]

I'd be really surprised if EQ didn't use the same random number generator algorithm that virtually every game and probably 95% of all programs use. There is simply no need for anything more complicated than something like the C rand() call for most application of randomness.

I think you're right. It would be nice to see a test that told us just how good the EQ RNG really is, but I don't believe it's important. The patterns inherit in the C/Java/VB version of rand() seem to be perfectly acceptable / imperceptible in this context. It's not a great random generator, but it's good enough.

One melee round (dual wield) has at least 20+ RNG checks. Just this activity alone is easily 10 times the number of RNG usage than tradeskills. I'd assume the RNGs are done on server-side otherwise there would be hacks that lets you tell your cilent that you always rolled a success.

Absolutely. There's other things the RNG does. Most of those, including all combat, we can discount, because we're usually skilling up in a low-lag, population-free zone. That still leaves enough that we aren't getting an exclusive RNG.

I think its safe to conclude the original point of this thread is thoroughly disproven here.

[Palarran]

Their ENTIRE BUSINESS, and it IS a business, depends UTTERLY on their RNG being the best it can be.

Based on this, I'd suggest that there's a reasonable chance EQ servers are using something like /dev/urandom:
http://www.gsp.com/cgi-bin/man.cgi?s...4&topic=random

The /dev/random device is suitable for use when very high quality randomness is desired (e.g. for key generation), as it will only return a maximum of the number of bits of randomness (as estimated by the random number generator) contained in the entropy pool.

The /dev/urandom device does not have this limit, and will return as many bytes as are requested. As more and more random bytes are requested without giving time for the entropy pool to recharge, this will result in lower quality random numbers. For many applications, however, this is acceptable.

Note that the slow part here is not the number generation itself but filling the entropy pool, which depends on external sources of random data such as a microphone.

I'm sure just using an appropriate deterministic function would be acceptable too, but that doesn't necessarily mean that's what the developers chose to do.

[YurienStonebow]

ok this thread is offically dull, irrating, annoying, and boring. i'm not going to bother reading it any more <click>

[Voltai Ausra]

besides the RNG

The simple fact of the matter is this. What does salvage do? it saves components for you on a failed combine. What does this mean? It means more combines. What does more combines equal? More chances for skillups. Press combine enough times and you will get a skillup. You may get back to back skillups, you may go 100+ combines without one and want to bash your head in. In the end salvage is your friend if you are a tradeskiller.

His assertion that salvage is bad for skillups is strange. Given that we all know more combines for less cost is a good thing? I'll lay a challenge to anyone that is interested and isn't a Master smith already.

Do do mistletoe cutting sickles for 8 hours and see how you feel about salvage and skillups. That ought to be a large enough base set. Count the number of additional combines you get from salvage. As well as the material cost savings. Then look at the number of skillups over the total number of combines. Does that number fall within the projected number of combines required for skill up? And if you haven't seen a projection on combine numbers, you spend to little time here imo

Man, I'm sorry to sort of rant on this. But yeesh.
I don't even need evidence of a mathmatical variety to know that this thread's title is incorrect.