Looking for Bias in RNG... (Testing)

[Keeps]

It's an on going joke about how the RNG might like you or might not...   But after years of of playing I've noticed patterns in what should be random events.

Mutation special abilities come in groupings each game (That is if the ability devour shows up on first mutation, and steal shows up on 2nd mutation, more than a fair share of creatures will have devour and steal versus any other ability if you continue mutating creatures.)

TUing mutated creatures consistently mutates to a different ability in what appears to be an order and not random. (Like it's pulling from the same seed.)

Chaos Seed -> More often then any other ability I see a 3 point arrow strike as the first attack.

Card Draws -> When I've played a deck, I've noticed large groupings occur what appears far to often, enough that I can more consistently then what should be possible predict certain cards when playing some decks.

More often then not, I've noticed having one SoFr on the field can go upwards of 20 attacks before a special attack far too often, but 2 SoFr results in consistently 50% of attacks.  However the next game 1 SoFr and almost all attacks get the bonus.

We've had bugged RNG events several times in this game.

I've played many games like good old CiV2, worked on a pre-generated seed.  Because of this, if you immediately rolled back to the previous turn and did everything the same, the same events would occur.  These kind of semi-predictable elements, makes me think the EtG RNG works like it's pulling from a pre-generated seed and where it falls on the seed is generated by a random number at the very beginning.  Making it possible to reverse engineer the seed and given a few starting events, it's possible to predict the outcomes of all future random events for that game.

Now all these could be in my head, from simply playing too much.  But the question comes up far too often, to end my curiosity and any future speculation from fellow members, I propose an RNG Test to search for bias.

We can set the AI to have specific decks, and build specific decks to do testing of various 'random' abilities events to see if patterns emerge.

Anyone interested in participating in such a test?

[shinyarceus4]

I've always wondered if the AI cheated... this could turn out interesting.

Honestly though, my bigger problem is with the RNG putting all my pillars on the bottom. That's probably genuinely random, though.

[Calambar]

I have not done any thorough research (yet), but just from a raugh observation, I always have a feeling that unupped pandemonium does more harm to the opponent's creatures than to casting player... Maybe it's just me or maybe it is even intentional (?), I don't know. It is just what comes to my mind right now

[dragtom]

I have not done any thorough research (yet), but just from a raugh observation, I always have a feeling that unupped pandemonium does more harm to the opponent's creatures than to casting player... Maybe it's just me or maybe it is even intentional (?), I don't know. It is just what comes to my mind right now

because PU is (almost) always positive for the caster, wether it targets the opponent's creatures or the caster's.

[Keeps]

I didn't create this thread for more suspicion.  I am hoping for ideas for building the tests to make sure whatever method we use, covers all the possible variables and to find people will to do the tests.

[Leodip]

I don't think RNG is biased, except for the same bias most number generators have:
I happen to remember reading somewhere (I'll link it, if I find it back) that an RNG which extracts, for example, 1, will have a slightly less chance to extract 1 again compared to the other possibilities. This is usually abolished by decimal numbers, because, let's say we randomize from 1 to 7, or, to be more specific, from 1,000 to 7,999, and consider everything from 1,000 to 1,999 as 1, 2,000 to 2,999 as 2 and so on, even if we render 1,474 (still an example) SLIGHTLY less probable to be extracted, the other 999 numbers which make 1 will balance it. Still, not sure about this.

Spoiler for Hidden:

And this all was being said assuming Zanzarino didn't alter the RNG to make something particular.

[Jyiber]

It's not impossible to detect minor patterns in the RNG patterns, but the more factors that are in the generator, the more likely you are to mistake bad luck for a bias.

I'm Irish, so I think when I'm using things with RNG I'm pretty much cheating.

[eaglgenes101]

http://xkcd.com/882/

[rob77dp]

Here is something I'd like to put on the list "to be looked into"...

First turn quanta from Quantum Towers with Entropy Mark and a deck having x6 Supernova.  I am quite paranoid about over-stating and over-reacting to "this happens every dang game!".  However, I feel a bit validated over hundreds of PSN-bow matches that Entropy is less probable to get from the quanta on first-turn Quantum Towers than the rest.  Could a probability be worked up for it that I can use to compare with my real world results in the future?  Thanks! 

[Leodip]

Here is something I'd like to put on the list "to be looked into"...

First turn quanta from Quantum Towers with Entropy Mark and a deck having x6 Supernova.  I am quite paranoid about over-stating and over-reacting to "this happens every dang game!".  However, I feel a bit validated over hundreds of PSN-bow matches that Entropy is less probable to get from the quanta on first-turn Quantum Towers than the rest.  Could a probability be worked up for it that I can use to compare with my real world results in the future?  Thanks! 

I still don't think there's a bias there, but if you want to, go in the trainer, make a deck with only Quantum Towers and make your opponent play a deck with only Fire Towers with 6 SoBras.
Take account of two things:
FirstTurnEntropyQuanta*11/FirstTurnOverallQuantaAsideEntropy
AND
SecondTurnEntropyQuanta*11/SecondTurnOverallQuantaAsideEntropy

The number of both should be as close to 1 as possible, after averaging the tests you did. Working with at least 7 Quantum Towers means you have to do fewer tests, but still, it'll need plenty.

[Keeps]

What he is talking about is something that might happen from a mod bias which is another possible bias. 
I think what we really have is a case of seed bias too.  So to look for both.

I decided to run a test in a similiar vain here is the test looking for both, it will take tens of thousands of games to identify this though so I would like some help if anyone else does.

Play 1 quanta pillar, and play 5 turns with only the 1 quanta pillar out against AI3, replace the pillar if it's destroyed.
Record just the quanta produced each turn by the quanta pillar.
Repeat a lot. 

I have to gather hundreds of more games on this test alone.  However my first 100 games of testing it this way 500 random generation,  here is an interesting thing...  I'm already seeing an abnormal grouping.

Infact, when looking at the elements, top down as presented on the screen
This order:
Darkness   Water   Death   Life   Earth   Aether   Fire   Air   Gravity   Entropy   Time   Light

Over 98% of the stats I have gathered have a grouping in each row.  That is if I have a Fire, i have either a second fire or an air, if I have an air, I have a second air or a gravity, if I have a time, I either have a 2nd time or a darkness, etc..

Now it's only 100 games, so this might disappear, but it's an interesting phenomenon.

[Keeps]

So, based on the grouping problems I went to research if anyone else has this.

Turns out, it absolutely does:
The math.random() function in action script produces the same seed pattern so there is a seed pattern.
In addition, if Zanz did not build the function correctly, he can build a strong bias in effect doubling the odds for one particular result,  just like the old nymph bug.

Adobe has replaced the function starting with Flash version 11 see:
http://help.adobe.com/en_US/FlashPlatform/beta/reference/actionscript/3/flash/crypto/package.html

So what does this mean?
It means, random is far from random in this game, not even close to a decent pseudo random.  It means one person might have great 'luck' with a deck while another person does not.  It also means, you can force a deck to play better over short bursts than long bursts.  In other-words, a person playing 10 games a day might continuously win with a deck, while a person playing 100 games, might win the first 10, but not the next 20 or 30 because the more you play the more likely you will have bad draws, if you have a deck that continuously draws well the first 5 or 10 games you play it.

It means, that certain cards are more or less effective any given game, and depending on how zanz built this, the results are completely predictable.

Finally, there is a double odds bias if you built out several functions incorrectly, like he clearly did with the nymph bug.