failure rate too high

[kekenewah]

please ignore

[kekenewah]

Ok I recalculated my failure rate on the level X Divine healing potions with 300 skill and Alchemy mastery 3 and a 15% trophy. My failure rate on 1000 combines was 6.8 % which is above the 5% predicted by the calculator on the main page.

[melkorr]

1.8% as far as the RNG is concerned is insignificant. What you've seen is mildly not-quite-average luck. When you hit 15% failure rate, on a run about two or three times as large as that one, then it'll be statistically significant.

[Shismar]

I find the difference of 36% quite significant and the sample size is plenty.
Of course knowing that the EQ RNG is a little streaky, it might be a good idea to run something similar and look again. I checked my logs and for 221 combines I had only 9 failures, thats 4.2%. But my parameters are different, 12% trophy and only 2 AA.

[melkorr]

Where are you pulling 36% from?

The calculator is an estimated guess, based on the average run. Good and bad streaks happen, and the RNG throws of results from the average the calc gave you.

[Shismar]

5% is what it should be, 6.8% was observed, thats +36%.

[Mokara]

No, it's 1.8%

[melkorr]

I'm too lazy to do the math, but I'll take you're word that that's a 36 percent increase in the existing failure percent.

Either way you say it, the difference is still insignificant. When you log over 5000 combines minimum which show a deviation of 10% either way from the calculator, then we'll talk. And that's 10% like it's 1.8% above. Not your way.

When you have data that shows those results, post about it. I'll tell you that it's bad luck, and the RNG hates you. When you log 10k combines that show that deviation, I'll admit some statistical validity, and I'll consider a possibility of some bug with the game.

Note: If anyone here is logging 10k plus runs of combines, consider therapy. You're allowed to combine data from multiple people.

[Shismar]

The chance to get that result is 0.23 % using binomial probability distribution (exel has a function BINOMDEST, just plug in the numbers yourself). That is actually higher than I would have thought, still rather unlikely. My 9 failures in 221 is a 10% probable result, thats reasonable. In statistics us economists (math stuff not my strong point though) get suspicious when probabilities fall below 5%. That assumes a high degree of randomness and a large sample size (30+ observations at one degree freedom).

There have been errors in these formulas before, I'd be happy if that was just a freak outcome myself but would like to see more data.

[Roanne]

It falls to how you decide to define "random."

In the sense you cannot predict the next result, the RNG is random but in the sense of generating a matrix of equally distributed results across the range, its not very random. This seems especially true over small time period samples where clustering of results is obvious (although still not predictable). This is evidenced in combat when most people score critical hits within a few seconds of each other during raids where many people are swinging or in tradeskills when you get those "lucky" streaks of 2-3 skillups at the upper skill levels within a few combines.

My best guess is the RNG uses some sort of seed based on a sine wave with standard deviations off that seed to "randomize" the result (which makes extremely high or low results less common and most results falling in the middle somewhere), but how the RNG actually works has been a guarded secret by the people who know.

We can debate how large a sample would be required to determine a certain confidence level in the results for pages, but without knowing what kind of distribution we should expect, we can't say for sure. Its safe to say you need some arbitrarily large sample size to know if things are messed up or not, although not very practical.

Most of us assume an equal distribution over time of all possible results (which makes sense if your model is based on a % chance to fail or succeed), but I hypothesize the unexpected results are based more in flaws of how we think the RNG works than flaws in code formulas.

[Neebat]

What you're describing is an RNG behaving in a "predictable" way, and it would be massively exploitable if it happened.

There are some pretty good tests out there to detect it, and from what I understand, the developers have checked. The EQ random number generator produces pretty good random numbers. The streakiness you see is a necessary attribute of real randomness.

[Roanne]

Im not sure which model you are criticizing.

The ideal I'm describing is an RNG with uniform distribution of possible results (that is every value in the range has an equal chance of occurring every time a value is generated) with mutual independence (the chance of a single value reoccurring doesn't change after the value occurs). This is how most players assume the RNG works

But what I describe I think is actually happening is an RNG that tends to generate clustered results across a range that changes with time. The values are still mutually independent but not uniformly distributed at any given moment. The argument of uniform distribution over time can't be ruled out but I suspect the effect is a bell curve of results where the high and low end extreme results are rarer than the middle results, but this is just my guess based on in game observations over time. Its perfectly possible for the RNG to behave this way and still be unpredictable.

If the RNG behaved according the ideal assumption model, there would be no reason to keep its process a secret. The fact it is a secret suggests it does not behave this way and it's possible to exploit and therefore not "random." ...unless you define random simply as "unpredictable" which is where I started.

[MareeTP]

With a modified skill of 336 - doing Skinspikes XIII combines at trival 466 - I did 97 combines and obtained 38 skinspike potions.

Ive done a lot of clicking in my time and have done plenty of over 400 trivial skill combines in other trades - and Im hoping that either the RNG was just being evil or when I max the trophy it will magically put me over top of the step to more success... but that seemed an excessively low success rate to me.

So... I waited a while. Then I did a run of Celestial Healing XIII. Out of 100 combines, I succeeded 35 times. This toon is only level 25 - so I have no AM3 on her. Is that the difference here from the experience I have on my other toons? They all have mastery 3 aas.

I think its time to level the shaman

[KyrosKrane]

From the main site:
Trivial 466, skill 300, 12% mod, 0 mastery = 38% chance of success.

Change it to master 3: 69% chance of success.

[solithan]

If the RNG behaved according the ideal assumption model, there would be no reason to keep its process a secret. The fact it is a secret suggests it does not behave this way and it's possible to exploit and therefore not "random." ...unless you define random simply as "unpredictable" which is where I started.

I was under the impression random numbers were based on the concept of Seeds and if I knew the seed I might be able to manipulate the "randomness" thus the need for secrecy. Some of the common rumors are timestamp and loc. I've seen people that rolled exactly the same time got exactly the same number. But yes I dont think its random per say but more very hard to predict.

Random events in real life often run in streaks, going "cold" or "hot" is often seen in Casinos where the odds are very well known but luck still plays a factor.