Guild Wars 2

Let me start by saying that the Fractal reward chests are not definitely bugged, they are only over 96% likely to be bugged. And not necessarily for everyone, just for some players at least.

I have been a regular completer of top level fractals for some time now, receiving what was probably an average return rate on the chests, those implemented at the time. In early March I noticed the start of a dry spell, which I duly attributed to RNG, in which I received no ascended armour chest drops. This was particularly noticeable to me since I was gearing up two characters to expand my raiding options. I kept track of how long and this spell extended to over 50 days, until April 19th, when the reward system was overhauled.

Following that, I have still not received a single ascended armour chest. I stopped opening the chests after 24th May and I am now accumulating them, but I did open them [T1-T4] each day until then, some 36 days in all.

This means a run of at least 86 days without a single ascended armour drop.

Now, if the stats reported by KING* are correct, or thereabouts, and if my maths is correct**, the chances of this happening to any specific individual are less than 1 in 2.5 million. This, to me, goes beyond simple RNG and me just being the unluckiest person in the game over this period.

I don’t know how many people do top tier fractals every day, but assuming 100,000 this would make the likelihood that even a single person receives no armour over this period less than 1 in 25, i.e. a greater than 96% probability that everyone receives at least one piece. And this is on very conservative maths, and how much less likely that it would be someone who was actually paying attention.

Unless of course it is somehow bugged for some players, incl me. A hangover from a prior DR implementation, a player flag, a general bug, whatever.

Support will not look into it, because according to their script it is simply RNG. Which brings me to the second part of the topic title. At what point will support concede that the numbers indicate that something more than RNG may be at play?

• 13% per day prior to 19/4; 6.9% per grp of [T1-T4] since 19/4
  • Product of straightforward binomial calculation of each of the two periods

For there to be a bug no one would have gotten an item. As soon as 1 person gets that said item, it proves there is no bug. There is no account based RNG, everyone who opens a chest has the same chance of getting whats on the loot table for that chest.

The data from KING is just that data, you cant expect to get anything at any questimated rate. You could open a million of them and still not get anything.

1 in 1,000 is the same as 1,000 in 1,000,000. You could open 998,000 and get nothing.

RNG rolls every time you open a single box. You have exactly the same chance each time.

Gator.5729:

For there to be a bug no one would have gotten an item.

Not remotely true of course. Bugs come in all shapes and sizes and can affect different items and different players in different ways depending on a number of factors.

Also, of course it is possible that you could roll a million times and get zero return, but there is a point that the statistical probability becomes so low that on balance of probability something else is in play.

36 days without any ascended drop is unlucky but unsurprising. Using [KING]‘s data from their runs, I’ve run simulations of 10,000 T4 runs (and separately, of 10,000 days of three T4 runs) and included all the chests (masters, expert, adept, initiate). Longer streaks aren’t unusual.

If everyone was going 36 days without ascended, that would suggest a bug. Currently, though, I see a tiny fraction of people reported “predictable” bad streaks. Some of us are just going to suffer bad luck.

Come back if you get to 90 days without an ascended drop and I’ll run the appropriate simulations again and help you present the evidence to ANet. (Alternatively, contact me via mail here or in game and I can help you set up your own simulation.)

(I used simulations instead of formulas, because in every discussion of probability we have on the forums, there’s always an argument about whether people used the appropriate formulas and variables. The simulation has the advantage of not caring about theory and is therefore a commonly-used technique for scientists and investment banks. It, of course, has the disadvantage of being harder to vet, which is why I offer to help anyone wanting to set up their own.)

Na, please read posts more carefully, that is 86 days without ascended armour. Are you seriously suggesting the 4 extra days is your make or break?

The simulation would be interesting, but it is simply a question of statistical probability, which it will be interesting to see if your simulation matches.

I was going strictly by the new reward system, which started on 19 April. For that, we’re talking about 49 days max, of which you earned 36 days of rewards.

While one could modify the simulation to cover 50 days of the old + 36 of the new, there’s no point in my strong opinion. Any potential bug would have been reset with the new mechanic. Moreover: there’s no way ANet could troubleshoot an issue with the old rewards, as the code isn’t available to their testers.

I’m really sorry you’ve gone so long without any ascended drop. That’s definitely frustrating and demoralizing. And it’s no comfort to say that your luck is bound to change, since we both know that luck remains the same, no matter how good|bad it was the last 5 weeks.

Still, streaks always end. And I hope you’re enjoying the gameplay (since, personally, I would have stopped playing in 2012 if rewards were my primary motivator).

I’m keeping an eye on [KING]‘s data and periodically re-running the simulations to see if a lot changes. Keep us posted about what’s going on for you and I’ll compare it to the simulations. ANet likes data, so if it turns out things don’t change, it will be useful ‘evidence’ to get the issue past the first tier of support and to the ANet folks who can usefully research potential issues with RNG.

I don’t agree on the usefulness or not of running the simulation across the two rewards systems. There are a number of scenarios [flagged account, residual DR issue] that would still hold, and it is worth assessing whether the return falls into a range of feasible or not. And of course the code is still available to them, perhaps at a little more trouble, it is true. Although it is more likely that if there is a problem it stems from an associated common piece of code rather than one that was replaced. There are many factors that could and would be common.

I guess my key point is that there comes a time when the statistical unlikeliness indicates an issue, what is that point for Anet – or will they continue to just point to RNG and do nothing.

First, let me say that I’m not trying to add to your frustration — I’d be similarly discombobulated by the lack of ascended you are seeing. Rather, I’m trying to point out the enormity of the task of (a) convincing ANet that there’s something to investigate and (b) what it would likely take (on their part) to follow up. While they could look closely at any time, they sheer volume of data they already have is going to convince them any single issue is a matter of perspective, rather than a problem with a mechanic, with underlying code, or with the actual thing that generates random numbers for them.

In short, extraordinary claims require extraordinary evidence. And as of yet, we don’t have that.

Some specifics…

Rhea Ravenseeker.7105:

I don’t agree on the usefulness or not of running the simulation across the two rewards systems. There are a number of scenarios [flagged account, residual DR issue] that would still hold, and it is worth assessing whether the return falls into a range of feasible or not.

In theory, I agree with you. In practice, I don’t see any scenarios in which the prior results are going to help convince folks at ANet.

For example: ANet has said repeatedly that there is no such thing as “flagged accounts” (for RNG) and there’s lots of evidence that there’s no such thing as DR for chests|bags|etc.

Rhea Ravenseeker.7105:

And of course the code is still available to them, perhaps at a little more trouble, it is true. Although it is more likely that if there is a problem it stems from an associated common piece of code rather than one that was replaced. There are many factors that could and would be common.

For argument’s sake, let’s assume that they do have ready-access. And let’s say we could show that your situation was, indeed, one in a million. How would they even begin to start troubleshooting without a test environment? Where would they look? The scenarios we can both imagine are odd, edge cases — if they were common, ANet would have seen them already when they’ve looked at RNG (either in their regular spot-checking or during some of the specific reviews they’ve done — I don’t know how often they do either, just it’s been mentioned that it happens).

The usual way to troubleshoot stuff like this is to generate a bunch of chests (in the test server) and open them and collate results, swapping around accounts and aiming to force an issue. With just the old code, out of context of its usage, they won’t be able to do that.

Plus, I wouldn’t be so sure that the code is available to them, or at least, not in a form that allows them to easily investigate it. (That’s an entirely separate issue about how well they document and manage their code base — I’ve seen worse and I’ve seen better.)

Rhea Ravenseeker.7105:

I guess my key point is that there comes a time when the statistical unlikeliness indicates an issue, what is that point for Anet – or will they continue to just point to RNG and do nothing.

I agree with you: there is a level of unlikeliness at which ANet would need to reevaluate whether things are working as intended. I just don’t agree that the current situation qualifies.

I recall back in the days when I used to find several dyes a single day!
For several months, (not playing the whole time) I kept finding dyes while my guild mates were finding almost none and a selected few were finding ABSOLUTELY NONE.
Then, ANet changed the way dyes drop in game.
So, from my experience, the global RNG is pure B.S.
I am not a gambling person, but I would accept a fair RNG in GW2.
The problem is, RNG is more RNG to some then others for some elusive reason.

Deepcuts.9740:

The problem is, RNG is more RNG to some then others for some elusive reason.

The entire point is that it’s random; some people are going to be lucky and others will be unlucky, over short periods of time. Random results only even out when the numbers are high enough.

So, mathematically, RNG is exactly the same for everyone.

What’s different is how we feel about it. Generally, people who get good drops are fine with RNG and people who get less-than-average are not. As humans, we tend to expect average or better, which is unrealistic, since just as many of us are going to get lower-than-average results as get better-than-average.

No one in the industry has been able to figure out a perfect balance of random + defined rewards — token-driven rewards lead people to feel a game is grindy and lacklucster; random-driven rewards lead to the feeling of unfairness. If the drop rates are too good, then people don’t value the rewards at all — common drops are considered boring and ‘worthless’ (for good reasons).

In short, RNG is the same for all of us, making it a fair tool to make drops more exciting than guaranteed rewards. The problem is: none of us want to be the unlucky player and (in order to have rare drops), only a few of us get to be lucky.

Illconceived Was Na.9781:

In short, RNG is the same for all of us

How can you be so sure? Unless I am the developer who made the code, I couldn’t state that.
As I told you: finding a certain item DAILY, for months, while others couldn’t find the same item AT ALL for the same amount of time (dyes in this case) is not RNG imho.
I am not talking about precursors that MIGHT drop once or twice in a lifetime or items that have such a lot drop rate that you can just invoke pure luck. I am talking about items that dropped like rain for some while others were just…living in the driest desert.

Again, since I did not code the thing, I might be totally wrong about all this.
As you said, it might just be my expectation.

Illconceived Was Na.9781:

In short, RNG is the same for all of us

Deepcuts.9740:

How can you be so sure? Unless I am the developer who made the code, I couldn’t state that.

I’m sure because there is no evidence of differences among people — just anecdotes (such as the one you relate), without any sort of actual data collection. I’m sure because ANet runs their own spot checks of how well their drop tables are set up. I’m sure because any time someone attempts to show there’s an exception, the evidence suggests otherwise.

Even if you & I worked on the code together, we couldn’t be sure that everything was working properly — the proof is in the results and those, as far as any of us have demonstrated, those results show RNG is the same.

And finally, my point was that the entire misunderstanding is the idea that any pattern suggests something non-random, whereas ‘true’ random behavior exhibits clear patterns. (See the attachment — the random distribution of stars on the left shows clumping, while the formula-produced results on the right are evenly spread out.)

Deepcuts.9740:

As I told you: finding a certain item DAILY, for months, while others couldn’t find the same item AT ALL for the same amount of time (dyes in this case) is not RNG imho.

Neither the original phrasing or what you’ve posted here allows anyone to compare what you experienced — as written, it’s inconclusive at best. How many days are we talking about? What was the specific item? What were the drops coming from? How many drops were you getting? How many were you buddies getting (each or cumulatively)?

Without the specific data, it’s hard to tell if something was predictable or if it was random with good|bad luck or if it was, as you say, “not RNG.”

For example, over a particular period, over several days of farming, I got far fewer rares than someone farming in my party. That sounds as if it might be evidence of different drop rates for the two of us. It wasn’t — it turned out, my build was just a lot less efficient at killing foes (I changed my build to match m buddy’s and voila, roughly the same amount of rares over similar periods).

Deepcuts.9740:

I am not talking about precursors that MIGHT drop once or twice in a lifetime or items that have such a lot drop rate that you can just invoke pure luck. I am talking about items that dropped like rain for some while others were just…living in the driest desert.

Again, without specific numbers, it’s hard to know what you mean by “like rain” versus “driest desert.” And it’s hard to know if that is unexpected for the drop rates or likely. Someone is going to get really long luck streaks — how long is “too long”?

Deepcuts.9740:

Again, since I did not code the thing, I might be totally wrong about all this.

Again, it’s not about the code — it’s about the data. I wouldn’t believe anyone from ANet who said they were sure of their code — unless that surety came from collating the vast amounts of data they collect and running all sorts of statistical analyses to show that, yeah, it’s random (and they’ve said they run such tests, although I’d definitely like to know what sorts of tests, how often, and what kinds of “slicing” they do — even statisticians sometimes lose the forest for the trees or vice versa).

Deepcuts.9740:

As you said, it might just be my expectation.

You’re not alone.

A while back, I had a really weird streak of forging masterwork daggers.

• Ran through ~120 and got 1 rare (expected: 6). <- Seems like a broken forge, since 120 inputs should be plenty, right?
• Ran through ~32 and got 6 rares (expected 1.5) <- oh, erm, about time?

If I had stopped counting at the first 40 forges, I would have thought there was a problem. Had I only counted the next 32 inputs, I would have dismissed “6 promotions in 8 tries” as “due me”, for bad luck in the past. Only by counting continuously through nearly 50 forges did my results start to look “average” (exactly as expected, as it turned out).

As humans, we expect “at least average” results, which we translate in our heads as “average is bad, since it’s not good and below average is really bad” — when “average” is just, you know, average and “below average” is common. We tend to forget to count “better than average”, because it fits our expectations.

tl;dr we humans are bad at estimating odds or understanding patterns by gut alone. Collating data is the only way to be sure we can tell which is luck and which is bad code/etc.