So I’m not sure if something got wonky with that latest patch a few months ago, but since then I can’t seem to get a point added to a gem to save my life. Working on leveling gems for augmentation and at the 60% rate of success I’m not kidding my success rate is closer to 20% over like 100 rolls. What gives?
Bad luck is not a bug.
3 Likes
This.
It all evens out with a big enough sample size. I’ve gotten several 30’s in a row just as much as failing 80%'s.
Keep in mind, it’s NOT 60% or 20% or “any”% over XXX rolls.
It’s %Chance for each roll, whether you get 4 rolls, 5 rolls (for an Empowered GRift) or 100 rolls within a game session.
As Dragonmaster pointed out, I’ve had times when I got 4 successes at 60%. And there are times when I had 2 fails at 90%.
This is not a bug.
Wrong understanding of statistics and rng at all. 
-
yes, of course, the probability of fail on each try with the probability = 60% is always = 40% at any number of tries, but: a) the probability of 100 fails in a row is much more less than the probability of 10 fails in a row (though the probability of each try is always the same); b) the difference from the mean (sum of probabilities = 60 and 6 accordingly = the total number of successes for the big enough number of tries) = the same sense as the standard deviation has for these two rows are also different (significantly);
-
on the long rung you must obtain the number of successes = mean +/- desired number of standard deviations; for example, for the processes in the real life which are described by the normal distribution, the rule of 3 standard deviations is used usually, wiki:
For an approximately normal data set, the values within one standard deviation of the mean account for about 68% of the set; while within two standard deviations account for about 95%; and within three standard deviations account for about 99.7%.
- further, machines do not use TRUE random at all, they use PSEUDO Random Number Generators which are bugged (in the comparison with the true random) originally; as the evident outcome one can easily obtain the outcomes like we have in classics when there were players that never obtained legendaries or obtained them too seldom and these facts were confirmed by developers and fixed by them (hm, several years after, shortly before RoS). During all this time all such players obtained frustration and leaved the game just because it is not they are so unlucky, it so just very bad (and bugged) rng-realisation in this game.
And now we have again very rare items (primals) which - theoretically - can’t be obtained by some players in any reasonable amount of time (say, several years) just because it is random.
ok, all of these are just theoretical things (which are evident for anyone who is working with statistics). Is it possible to adjust rng in this current game with the minimal efforts for that to have random but to avoid its extremes like described TS? Yes, of course.
Ok, our situation with leveling gems for augmentation at the 60% rate of success. What do we know about true random and final outcome of many-many such tries? As I said early, we know that 99.7% of our possible outcomes will lie in such range:
mu +/- 3 x sigma = n x p +/- 3 x sqrt(n x p x (1 - p))
condition on n:
np and n(1−p) must be greater than or equal to 5.
Let us choose p = 0.6 (our probability) and n = 100.
Hence we will obtain such range:
100 x 0.6 +/- 3 x sqrt(100 x 0.6 x 0.4) = 60 +/- 14.7 = (45, 65)
I.e. in 100 rolls at the 60% rate of success we should (must - if we will have adjusted true random) have number of successes in the range from 45 to 65.
You can collect all these data and compare them at each step with theoretical ones. And if one obtained less than 45 successes in 100 tries you should to adjust the probability of success for that he/she obtained the appropriate number of tries he/she did not take.
More easy realization / improvement for rng:
- at each try we add to the additional variable the difference between the current probability of success and the obtained outcome (i.e. 0 - fail, 1 - success);
- at the next try we first add the current probability to the additional variable (+ store such value in temporary variable) and compare such value with the 0 and 1: if we obtained value that is <= 0, the outcome of this try will be fail with 100% probability; if we obtained value that is >= 1, the outcome of this try will be success with 100% probability;
- overall we will still have random with the same parameters as “true random” on the long run, but no more situations of 2 fails in a row on 90% probability (just because 0.9 + 0.9 = 1.8 >= 1) or many fails in a row on 60% probability (or many successes in a row on 30% probability).
For example, on 5 tries of leveling gems for augmentation at the 60% rate of success we will ALWAYS have exactly 3 successes and 2 fails just because the mean = 5 x 0.6 = exactly 3. The same situation we will have on the long run with the real true random (of course, +/- some deviation).
If one want to have more variability (wider range of possible outcomes), he/she can compare the mean not with the 0 and 1, but, say, with the -1 and 2 (i.e. one need to add desired shift for these boundary values).
In such case, in our example with 5 tries of leveling gems for augmentation at the 60% rate of success we will ALWAYS have exactly 2-4 successes and 1-3 fails just because the mean = 5 x 0.6 = exactly 3.
Considering there have been at least 3 mayor test of 1000’s of tries at 60% with and end result of a 59~61% I would say the RNG works as well as it possibly can so there is nothing to fix.
1 Like
Don’t forget: This is the Bug Report Forum. It is not a discussion or help forum. If you would like to discuss the percent chances of upgrading Legendary Gems, you should start a thread in the appropriate forum.
Please read:
1 Like
I want to note that the bug “no drop of legendaries at all” in classics was also the feature of some particular players, not all of them. So your 3 tests of 1000’s of tries at 60% can find nothing. You need to look at the individual statistics of each particular player (there are millions of them). And yes: 100 is also good enough number of tries and we see the big violation of the result (mean = 20%) from the theoretical range of values = (45, 65)%.
This was not a thing, there was a thing we called the pity timer that increased your likehood of getting legendary items the more you played until one dropped, for most of us this meant 1 or 2hours (can’t remember the exact time) per legendary, I honestly never saw a single post of not getting legendaries but rather useful ones which is an entirely different beast on itself.
I remember the text of original blue post with the explanation of the purpose of adding such fix. They said about existing players with no drop of legendaries at all and ones with rare frequency of such drop. I spent several hours for that to find it now (in archives) and can’t to do this.
Moreover, some old posts (PTR of RoS) with discussions about the system Loot 2.0 and this feature are also unaccessible now (you can view the text of the start of the post, but you can’t view the whole post).
Interesting situation: there is the feature in the game, but its explanation does not exist now. But also I found the posts with suggestions for adding this “pity timer” for primaries (red legendaries). Here logic is exactly the same as this feature had originally - to give guaranteed legendary (or primary) for ones who do not take them during a long enough time. And I want to note that this suggestion is really adequate just because we have the same small drop-rate of primaries now as one of usual legendaries in classic, and the thing that prevents players from taking usual legendaries in adequate time in classic was not fixed and it is still there.
P.S.: I do not see the reason for adding fix with the “pity timer” in the game with the correctly working random when all players took legendaries as intended. 
P.P.S.: we found original post and explanation of adding “the pity timer”. “Impossible situation” is possible even not due to bugged pRNG, just due to “some bugs floating around”. And I want to note that this comment was placed during testing of RoS which has much greater drop-rates of legendaries than in classic and much greater drop-rates than now primaries have.
www_diablowiki_net/Legendary_Pity_Timer (replace _ by .)