Is the AI cheating on chances?

[byebytoad]

Once I reached stage 12 of a siege to take a settlement (early game) I had to wait additional twelve months with 50% chances of taking a fort every month.
Now the chances of being able to end a siege in 12 months with 50% chances are 1 on 1024 (it's like launching a coin 12 times and getting 12 times head or crosses).
And this without taking into account also the chances to end it in the previous stages.

And this is , at least for me,not an isolated case.
I'm under the impression that some sub-routine begin to work when things are going well for you.
I noticed also that while looking at a siege running, sometimes the 'time's wheel' has some sort of very brief freeze and very slightly (at least graphically) god back just by a tiny bit.
Did the developer ever said if the AI has some cheating?

 

[Dominion]

No.

 

[Turbo215]

Once I reached stage 12 of a siege to take a settlement (early game) I had to wait additional twelve months with 50% chances of taking a fort every month.
Now the chances of being able to end a siege in 12 months with 50% chances are 1 on 1024 (it's like launching a coin 12 times and getting 12 times head or crosses).
And this without taking into account also the chances to end it in the previous stages.

And this is , at least for me,not an isolated case.
I'm under the impression that some sub-routine begin to work when things are going well for you.
I noticed also that while looking at a siege running, sometimes the 'time's wheel' has some sort of very brief freeze and very slightly (at least graphically) god back just by a tiny bit.
Did the developer ever said if the AI has some cheating?

To my knowledge the AI doesn't cheat with dice rolls and if it makes you feel better i had 3 consecutive disease outbreaks for a 1 in 2744. This is the most up to date information i could find on AI cheating Link

 

[cuendillar]

The AI is definitely cheating, I got an event about my wife doing exactly that with one of my advisors.

 

[Sfan]

I was about to post the exact same answer as Dominion. The AI does not cheat with dice rolls. You just tend to notice it more when you're unlucky than when you're lucky, and to assume that being unlucky from time to time isn't normal. 2 common bias.

 

[PhoenixG]

It's just bias. I got lucky once that I was able to siege down a lvl 9 fort in one siege tick and I had only 7% chance on it.

 

[deathbypie]

https://en.wikipedia.org/wiki/Confirmation_bias

 

[FrogCrusher]

While I do agree that there is no cheat and long term probabilities are fine, I really wonder if the RNG are constantly uniform. I often saw series in dice roll, such as 3 times the same number in a raw, for the two side of the battle. Same thing with firing advisors. It happened to me several times to have the same advisors at least three or four months in a raw. And so on...

 

[macd21]

Once I reached stage 12 of a siege to take a settlement (early game) I had to wait additional twelve months with 50% chances of taking a fort every month.
Now the chances of being able to end a siege in 12 months with 50% chances are 1 on 1024 (it's like launching a coin 12 times and getting 12 times head or crosses).
And this without taking into account also the chances to end it in the previous stages.

And this is , at least for me,not an isolated case.
I'm under the impression that some sub-routine begin to work when things are going well for you.
I noticed also that while looking at a siege running, sometimes the 'time's wheel' has some sort of very brief freeze and very slightly (at least graphically) god back just by a tiny bit.
Did the developer ever said if the AI has some cheating?

1 in 1024 sounds like a lot, but it has to happen eventually. Just think about it: how many sieges do you go through in an average war? How many wars do you go through in an average campaign? And how many campaigns have you played?

Now multiply that by the number of regular players of the game. Someone is going to get that terrible series of roles, and notice it, and come to the forum and post about it. Today that someone is you.

If the AI really was cheating on dice rolls, there’d be no question about it, because the forum would be flooded with people talking about 1 in 1024 chance sieges.

 

[bly08]

Hard to know without a lot of testing. Ruler stats wasn't RNG and it took several patches to notice. For the rest we can only take PDX's word for it.

 

[rinehime]

While I do agree that there is no cheat and long term probabilities are fine, I really wonder if the RNG are constantly uniform. I often saw series in dice roll, such as 3 times the same number in a raw, for the two side of the battle. Same thing with firing advisors. It happened to me several times to have the same advisors at least three or four months in a raw. And so on...

Same number 3 times in a row is totally normal and expected - you're forgetting the times you didn't have the same number in a row (or didn't even see it).

The advisor situation is a bit different. The new adivsor that pops up seems to be tied to current available advisors. Also, with the dip advisors, 3 of them require special conditions (colonist or port) to appear, so if you don't have those the same ones show up often. Also, the total available advisor pool is quite small (only 7 each).

Because there's some limitations on which advisors show up, It's also possible there's a bug with how the available pool is being calculated - I've seen the same batches after firing all advisors a couple times a in row as well. If the pool isn't updating correctly, you might expect to see the same thing over and over again, even though the RNG is fine.

Hard to know without a lot of testing. Ruler stats wasn't RNG and it took several patches to notice. For the rest we can only take PDX's word for it.

If you're talking about the even-generated ruler stats (Dutch Republic/Talented daughters/Ottomans/etc) well, they were RNG, just not the correct distribution [they were d6-1 instead of 2d4-2]. That bug was solved because someone wrote some events to test it and ran it a crap-load of times. That's the sort of thing that has to be done to figure this stuff out.

There are a lot of people that complain about RNG's "only being pseudo-random" or maybe not being "really random." Unless there's a problem with how the seed gets reset, the problems with game's "randomness" are (almost) never due to the numbers produced by the RNG, but instead how the events or game code are scripted to interpret the RNG i.e. the dice rolls themselves are almost certainly fine, however how that gets converted to casualties, %-chance to end siege, etc. is where the problems usually lie.

 

[Badesumofu]

While I do agree that there is no cheat and long term probabilities are fine, I really wonder if the RNG are constantly uniform. I often saw series in dice roll, such as 3 times the same number in a raw, for the two side of the battle. Same thing with firing advisors. It happened to me several times to have the same advisors at least three or four months in a raw. And so on...

Random does not mean uniform distribution. It means random. There is no reason to think that there is any problem with the RNG in this game. Post above mine already went into more detail, so I'll leave it at that.

 

[rinehime]

Random does not mean uniform distribution. It means random.

To be fair to @FrogCrusher , each die roll should follow a random uniform distribution... Combine them together and you get something different, course...

 

[Dominion]

To be fair to @FrogCrusher , each die roll should follow a random uniform distribution... Combine them together and you get something different, course...

Speak for yourself. I get nothing but Quantum entangled Schrödinger's rolls. All numbers at the same time mimicking both all future and all past rolls until I look at them.
As soon as I look, they're all zeros, obviously.

SCIENCE!

 

[True_Heir_of_Rome]

It's just confirmation bias. Even the best and most level headed of us remember getting out sieged by the AI when we had a 71% chance and they had a 7%. I more clearly recall losing a battle where I had every single modifier possible stacked in my advantage and rolled 6 zeros out of 7 possible rolls before retreating in disgust, as opposed to the multiple stack wipes I had in that same easy war earlier today. The bad sticks around, and the good is simply expected.

 

[Tarkilth]

No.

 

[AVN]

2^12= 4096 # 1024

 

[Howl]

I once played a game of Risk, where one of the oponents opened with rolling 11 consecutive 1s. It was probably the shortest game of Risk in the history of Riskkind.
Odds should be somewhere around 3 in 1 billion.
So unless you show me some statistical significant data...

Btw, there has been a thread in the past which did investigate the anti-player siege code*.
Summary: Nah, all random.

*[if siege=player
and war=very important
dice_roll=(succesful_siege_roll -1)
else
dice_roll=random]

 

[Gratak]

While I do agree that there is no cheat and long term probabilities are fine, I really wonder if the RNG are constantly uniform. I often saw series in dice roll, such as 3 times the same number in a raw, for the two side of the battle. Same thing with firing advisors. It happened to me several times to have the same advisors at least three or four months in a raw. And so on...

Not having those "3 times the same number in a row" would make it pseudo-random. There are actually applications (e.g. spotify IIRC) where developer do pseudo-random generators such that they feel more "random" to the user. Pretty sure that is not the case here though.

On the other hand. There certainly can be cases where stuff that should be random is not in the end. Advisors could come from a pre-defined pool that is refreshed at certain times. No idea.

 

[byebytoad]

It's not about 'confirmation bias'.
I got the same impression with the rebels spam too many time. It's like something lock and unlock with the sub-routine that manage the chances.

With rebels is at times even more dramatic. I got very often increases for 3 months in a row with unrest of "just" 3-4%.
There's like a chance in 10000-30000 that this happens, and I see this happening several times during a campaign. It's too often the same similar pattern with rebels:
no increase for some months and then 2 to 4 increases in a row. It's not just an issue of chances, it's also the distribution that seems to follow a pattern.

And for some reason 50% of chances of ending a siege in 1445 is definitely not the same than 50% of ending a siege in 1750. The second is definitely way more likely.
That's why I wrote about the 'AI' cheating, because -particularly at the beginning of the campaign- the numbers shown to the player do not seem to represent the 'real chances'