Public Service Announcement: The RNG in this game is not biased

[Codek]

Argument is invalid, no where did any of it mention making sacrifices towards the RNG gods.

Also Holding A+B down while looting to increase the success rate. (Or was it tapping B...?)

[CianaIezuborn]

Except when crafting, I hit a 10% chance failure 3 times in a row (that's 1:1000 chance) two days on the trot. Added, I crafted 46 items before getting a HQ when the HQ chance was always better than 6% and up to 15%.

World's unluckiest gamer? Doubt it. Spout all you want about stats when the engine works mathematically correctly. This game does not work correctly.

Failure bias, over hundreds of thousands of actions involved in crafting it would actually be less statistically likely to NEVER have 3 failures in a row. You don't notice the strings of ordinary successes you receive compared to the negative impact of strings of failure. Though in huge repetition, both will happen.

I ran Brayflox 23 times on my alt. When you consider the boss can drop the body or leg armor from 3 sets, that's 6 possible drops. That's about a 16.7% chance for any of the 6 pieces to drop, right? Assuming there's no bias towards a particular set, there's an 83.3% chance the drop will be something other than what I want.

1.4% chance of not getting the shirt over 23 runs is unlucky but not strange. It's harder to roll a 1 on an item roll and I do that all the time!

It sucks when you get probability based failures over and over again and it triggers a natural human response that the system is being unfair to you, though in reality it's being fairer than a human based system would have the capability of being.

When thinking about probability you have to consider your successes with your failure, and your failure with your successes. If you really believe that the system is broken then there needs to be proof provided that it's broken, which will involve thousands upon thousands of repetitions to observe how well or badly the data falls within the expected by 2-3 standard deviations.

As for my personal experience, I tend to do repetitive activity in bulk, and I've found success/failure/HQ rates to be about where I'd expect them to be over hundreds of repetitions.

[TetsuyaHikari]

Public Service Announcement: The OP in this thread is biased

[Nenin]

Ok so you've explained why a perfectly working RNG works perfectly.

What you haven't done is shown that SE's RNG works perfectly.

This.

OP if you want to prove a point actually do some tests in game and publish your numbers.

Most of us that have an issue with the RNG at present agree that the overall odds (with a big enough sample) match what the game tells us, but to get there it seems to roll long strings of good and bad luck more often than is believable (I say believable because software RNG is inherently flawed and only creates the illusion of being random).

The odds of getting 5 failures in a row on a 70% touch action are 0.00243, which is 0.243%. Now I probably craft a few dozen items in an evening and I see that string of bad luck multiple times in one evening, multiple evenings in a row. With those odds that simply shouldn't be happening.

So this. Not a single person here believes that a proper RNG doesn't work properly, just that the FFXIV RNG is broken or rigged.

Well what exactly would you be satisfied with? I'll gladly start recording all my meld/gather/crafting attempts and bump this with the results in a month if people want to see that. Until then everyone is just flinging around anecdotal evidence.

Most people underestimate how frequently RNGs are used in games like this. It's not only invoked when you swing a pickaxe or attempt a meld. It's also the backbone of all combat interactions.

Every time a mob swings at a tank it gets called half a dozen times at least. First it rolls to see if the swing is a miss entirely. Then it rolls again to see if the tank dodges it. Then it rolls to see if the tank parries it. Then it rolls to see if the tank blocks it. Let's say none of these happen and the swing goes through. It still has to roll again to see if the hit is a crit. Then it rolls yet again to determine the actual damage dealt within some range.

If the RNG created a patently nonuniform distribution every aspect of this game would be wonky.



I've got some real life stuff to attend to now but later tonight I can probably spend an hour or so spamming hasty touch with no steady hand, with steady hand I, and with steady hand II. I'll do the same for a few other abilities. Maybe I'll even record a bunch of gathering attempts. Since I know you won't take me at my word I'll screenshot the chat logs for posterity.

[CianaIezuborn]

Ok so you've explained why a perfectly working RNG works perfectly.

What you haven't done is shown that SE's RNG works perfectly.

It's a reasonable assumption to believe that a psuedoRNG runs properly in this day and age. A first(or second if the curriculum sucks) year computer science student can put one together with the .NET framework with very little work (the framework includes a Random class). The efforts required to create unit tests to verify it's functionality are slightly above trivial for most software professionals that participate in unit testing (which we all should!). Edit: On second thought it would be non trivial to test as you could never test all seeds, but I'm pretty confident you could develop a seed spread that would be sufficient.

Due to that reasonable assumption, it's on you to prove that there are flaws in the psuedoRNG and to identify what those flaws are. Everything I've heard from people can be explained by a lack of understanding of probability and failure/success bias.

Most of us that have an issue with the RNG at present agree that the overall odds (with a big enough sample) match what the game tells us, but to get there it seems to roll long strings of good and bad luck more often than is believable (I say believable because software RNG is inherently flawed and only creates the illusion of being random).

Runs of failure/success should happen quite often in large repetitions of data, it's less statistically likely for them not to happen. I don't know of any human that can observe runs of good/bad luck over a sufficiently large sample and make a gut call on whether there were to many or to little runs off the memory of that data. It requires intense calculation to make such a call, what you feel will always be subject to negative or positive bias.

How is software RNG inherently flawed? It's not random, but it is sufficiently removed from it's seed to more accurately model randomness than any thing a single human can physically provide.

Throwing a die is not a random act, however it is influenced by so many physical variables as to be indistinguishable from random to the normal human consciousness. This is the same principal that software driven psuedoRNG works off of.

[Sheer]

This was like a throwback to my first CS course at college and when we were learning how RNGs work - really nice to see you know what you're talking about, and that you're willing to take the time to explain it to others.

[StarMarmot]

During the early 1950s, the government set an elaborate scheme in motion, culminating in the present era when korean double agents used extra-terrestrial alpha wave stimulators to manipulate core game mechanic programmers at Square Enix during the development of Final Fantasy XIV. This covert act lead to the implementation of an advanced algorithm that operates through injected nano-machines in a battery of handpicked individuals, chosen by a council of powerful leaders from across the world.


Korean agent

The nano-machines use the product case as a vector, and once they have established working conditions within the human host, microcrystalline structures impregnate the soul of the subject with metaphysical ley-line disturbances and causes chakra impeding oscillations at a sub-atomic level. This, in turn, causes inevitable 4 chain missing streaks when they try to mine at 87%, 89%, 94% or 95% or when they try to use hasty touch during periods of irregular magnetic resonance in their immediate area of residence.

This is all based on FACTS. Take your "math" someplace else.

[Keoeson]

The Random Number Generator is working fine. The percentages listed for melding, gathering, and HQ chances are accurate to within 0.5%. The game is not shafting you, your memory is.

That's lots of conjecture and superficial knowledge. I'm really curious where you got that 0.5% from, do you have an insider source at SE?

Why does any of this matter?

I have no idea why you think it's important to explain binary representation of numbers, as that only affects the range of generated numbers. The real problem is how well the numbers are distributed over that range and you actually never get back to answer that question. Unless you work at SE it's even impossible to answer that question, as only SE developers know how their RNG is implemented.

It's true that there's no direct proof, that something is wrong with RNG used in FFXIV, but the reverse is equally true and there's enough anecdotal evidence to question the quality of the RNG.

If anyone wants to play with the standard pseudo RNG, here is a small test program, so you can see for yourself, how it compares to sequences generated by FFXIV:

#include <stdlib.h>
#include <stdio.h>
#include <time.h>

int main(int ac, char **av)
{
        double prob;
        int i, cnt, max;

        if (ac < 3) {
                printf("%s probability max_count [seed]\n", av[0]);
                return 1;
        }

        srand(ac > 3 ? strtol(av[3], NULL, 0) : time(NULL));

        prob = strtod(av[1], NULL) / 100;
        max = strtol(av[2], NULL, 0);
        cnt = 0;
        for (i = 0; i < max; i++) {
                double rnd = rand() / (double)RAND_MAX;
                if (rnd > 1)
                        printf("???");
                if (rnd < prob) {
                        printf("!");
                        cnt++;
                } else
                        printf(".");
        }

        printf("\nprob(%g): %d of %d (%g)\n", prob * 100, cnt, max, (double)cnt / max * 100);

        return 0;
}

[Astarica]

The only guys who think RNG can be flawed are the guys with a minimal understanding of randomness. They basically read a book somewhere saying RNG is pseudo-random, and that's about all they know. They don't even realize that the RNG is on the server that serves everyone, which already makes random even if the underlying system is not random. Let's imagine our system just generates the sequence 0 1 2 3 4 ... 99. For the purpose of say gathering we'll consider your gather a success if your % to succeed is greater than the number. However, this RNG also serves everything else that requires a random number, so it'd look like this:

A gathers a node with 95% success rate, gets the '0' (success)
B is fighting a lady wasp and asks for 50 random numbers (this is very conservative for a fight)
A gathers node again with 95% succes rate, gets '51' (success)

In reality, there are way more than one guy that's also hammering the server RNG at the same time. The server likely have given out thousands of random numbers before you can attempt to request a second random number again. Since the number of random number requests is, well, random, this means even if you knew the exact pattern of the numbers, it still does you no good because you have no idea how many numbers were requested in between your two actions. In our hypothetical example, as long as more than 100 requests for RNG occurs between any of your two actions, then you have absolutely no way to predict the number you're going to get, despite the sequence is completely known. You can't even get a rough idea of the range the number ought to fall under. And there's absolutely no way for any player to know the number of random number requests that has occurred between any of his 2 actions.