Lorem ipsum dolor sit amet, elit eget consectetuer adipiscing aenean dolor

I've been tracking the Kraken since the "nerf"

I play PvP to kill time. Since Kraken was “changed” to 35%, I have seen the following results with regards to activation of devour:

Kraken in Team 2: 4/10 (40%)
Kraken in Team 3: 55/71 (77.5%)

Tell me it’s not broken.

1 Like

I find both Kraken and Fizzbang to be extremely good and efficient when CPU uses them.
When I must have the worst luck in the entire world using them myself to invade with, only massive frustration. So I never uses them much, beside having them on my defense teams.


Since the last posting it’s another 4/5. This isn’t a run of bad luck - when invading a Kraken team in the third spot, the devour rate is well above 35% and in fact looks to be twice that or even higher.

Is it too much to ask for an explanation of what the actual devour rate and/or what affects it? I know what the card says and I know that’s wrong.

It’s not broken. It’s bad luck.


At what point is it not? The binomial probability of this occurring given an actual 0.35 probability is exactly 3.18x10^-14. The probability of seeing 59 or more successes out of 76 trials is 3.74x10^-14.

This is absurdly improbable.

And yet every day there are billion:1 and trillion:1 and even bajillion:1 odds that happen.

Not gonna lie - it’s frustrating. But for every person observing your kind of bad luck, there is some number of others who are observing the opposite. I don’t track them but I know my opponents’ devour rates are nowhere near yours.

Do you have the actual data? I can try to run some simple test to determine if it is random and independent or not random and independent similar to what I’ve done with TDS resurrection, which was not random. Short testing of resurrection rates

Since it is not random, your probability calculation is not correct because, strictly speaking, the events are not random and independent.

There has been some additional discussion in this thread as well: The Dragon Soul resurrect trait triggers more often than it should

The format should be a simple sequence of 0 (cast but not devoured) or 1 (cast and devoured) in the exact sequence as it happened. There should be no omissions. If something is omitted, it should be noted as unknown. I need a sequence at least 75 and better 100 excluding omissions. Running test with omissions is quite tricky but still should be possible.

It will be interesting also to run tests with player sample and AI sample to see if there is something different about these. And it will also make the test more reliable with higher statistical power due to comparison.

1 Like

This is not a very mathematically correct statement, just your belief. :rofl:

TBH, a concept of “global” probability of the whole game implies that random seed for pRNG is generated each time for every random event on the server. And this is absurd. So, no, there is only local pRNG determined by local copy of software. Anything on such a global scale is simply not possible and would mean “global” conspiracy. :rofl:

To translate what I just said in plain language, only local events count. You cannot extrapolate concept of probability to the whole array of the game.

No, I don’t have any more detailed data than provided.

It’s presently 60 out of 78. The equivalent outcome on the other side is fewer than one activations, which is equally insane.

While I agree in large part with your assessment, A couple of pedantic clarifying points might be helpful here.

  1. Anything involving matchmaking, tribute, or chest payout is calculated server-side, which means not everything is local pRNG. Likely the server calculations also use pRNG, but undoubtedly with different tech and therefore a separate implementation of random().
  2. If it is not reseeded (which is a big no-no in general with a pRNG), there are local “events” that use random numbers and result in the step function being executed that may have no player involvement whatsoever; an example might be ambient particle systems or other visual effects.

Regardless, I’m not surprised if the pRNG isn’t randomly distributed as per the previous discussion (it isn’t), but I would be surprised if, over time, the average result doesn’t converge on the expected value, as that would indicate a serious failing in the pRNG algorithm itself. Personally I have enough faith in IP2 that I would expect they are honest when they say that they don’t mess with random results, but that doesn’t mean there couldn’t be extant bugs lingering.

It might be reasonable to save the server computing power, although probably not by much, if the client request already comes with some pRNG number(s) and is then resolved by the server according to the number(s) supplied by the client. That is how infamous EA loot boxes work in some cases. Some older EA titles indeed have pRNG generated server-side but this was a major slow down. Based on how fast chest opening produces results in GoW, it means that servers are really optimized for this or that request has some info to it.

While I suppose this is possible, it’s not done that way in Gems of War. The client doesn’t sent the server any pRNG material in the request. (This also represents a security risk that would need mitigation of some sort against packet replay or MitM, so isn’t really the best idea to begin with.)

1 Like

The hits keep coming. I have noticed that I am facing many fewer Kraken teams, but the results are the same. It’s 5 out of 6 this week, making it 65 out of 84 since the change.

The probability of this occurring is now 2.78x10^-15, which means this is less likely than devour activating once or fewer in 84 casts.

This is broken.

My current theory is that there are hidden coefficients on everything chance-based. Possibly dynamic. What they are or the specific mechanics are anybody’s guess. (Theory)


I’m against doing math on something I do for pleasure, but I’m using quite often Lady Anariel, which should be a 4-matches generator since she spams mightily. Only she doesn’t. I’m very lucky if she loops once in a battle.
All the looping stats are off, IMHO.

Also, the chances of Fizzbang and Nobend to explode, when in defense teams, are way different from what they are when those troops are used in attacking teams.

Yesss. Possibly not the same for every client.


I just can’t take it any longer!!! :scream:

PLEASE change the title of the Thread to read:



I feel so much better now…


My friend, take Webspinner and Goblin Shaman out for a go and you will observe some very, very interesting things that corroborate your Anariel observations. Namely that efficiently connecting spawned Gems with those two is all but impossible, when it should be one of the most devastating parings in the game.

This one aspect of game RNG has been confirmed to not be purely random. Thank the Justice League for this.

The Justice League still works excellently well. In defense, at least.
While I went greatly with Anariel (Kraken, Kraken, Anariel, Mab) for a day, and suddenly BAM! No looping.

I used Rock Troll, Kraken, Dark Troll, Keeper of Souls for a time. At the beginning they looped something absurd, I mean they could loop even with 4, 5 gems of the right color on the board. There was no way the algoritm wasn’t tweaked. Then suddenly BAM! No loop. At all. Then they quite reliably looped with more than 10 gems of the right color. Then they didn’t.

Please unmade them. Thank you.