Inno's version of probability

[Mallea Maleficarum]

It's been a couple decades since I studied Statistics and Probability, but Inno's apparent formula seems incongruous with what I remember. For instance, acquiring 97 blue prints without having a complete set doesn't fit the typical model, which I calculate at about 29. I do understand outliers, but when I create a scatter chart 97 is closer to center than should be the norm. Indeed, I show an average of 46.7 per complete set.

Okay, yeah, I enjoy recording data, and I maintain a worksheet for keeping historic data on which GB blue prints I receive. Discounting BPs received during aids of guildmates, neighbors and friends, and only counting BPs received during RQs and GE, 61.2% of BPs received are for the Colosseum and Notre Dame. My spreadsheet goes back for over a year and I've been a meticulous recorder. This does not seem like a random distribution, at least not in the way my textbook and professor explained it those many years ago.

I'm certainly not accusing Inno of being a greedy, underhanded predator, but I am most curious about their RNG and distribution methodologies. I'm quite alright with a company shading the odds a bit, but what I'm seeing is much more than a bit. I guess what I'm asking is the likelihood that Inno is using AI (or some data-crawling scheme) to discern what players need, then reducing probability that they will receive needed BPs.

Now, to demonstrate this is not a complete trashing of Inno, my historic receipt of FP packs while doing RQs is 31.8% higher than expected.

A response from someone more current in Statistics and Probability would be appreciated.

 

[Kranyar the Mysterious]

when I create a scatter chart 97 is closer to center than should be the norm.

Exactly.

People like to say that it's random, or the sample size, etc, but scatter charts don't lie. I've seen this in the game everywhere since day one.

It's the lousy random number generator they use. As you probably know, no computer rng is actually random, but this one is worse than many. I suspect they use it because it's free code, because if they are paying for it they are getting ripped off.

On the upside you do get more good runs than you should as well, which tends to even things out for the most part.

 

[Falconwing]

It is totally random, until you only need one BP. Then the Keep Looking Program kicks in and drops you to around a 0.286% of getting the BP you need. It goes back to random as long as you have at least 8 copies of each plan.

 

[UBERhelp1]

While your spreadsheet goes back a year, how many worlds and/or GB are you recording? I have a feeling that your sample size is too small, as the smaller the sample size the less and less accurate your data can be. At the very least, 100-1000 trials would be standard when talking about RNGs. (but of course more is better)

 

[DeletedUser37581]

While your spreadsheet goes back a year, how many worlds and/or GB are you recording? I have a feeling that your sample size is too small, as the smaller the sample size the less and less accurate your data can be. At the very least, 100-1000 trials would be standard when talking about RNGs. (but of course more is better)

Sample size depends on intended accuracy. If you want to get values that are accurate to within 1% (with 2 standard deviation confidence level) then about 10,000 samples are needed.

 

[DeletedUser29726]

While I have seen evidence before that inno's random number generator may have some defects, it's certainly "random enough" that when people try to complain about results their data is nothing out of the ordinary compared to expectations (bad luck of the magnitude that it's going to happen to everyone eventually at some point). That is: From the perspective of a single account, on a single world, it may as well be random.

The defects people have pointed out that may have merit have had to do with biases based on a coarse timer (i.e. a running theory that when you get diamonds from one world from a given GE encounter, that you should immediately do the same encounter on as many worlds as you can to gain diamonds there too. This might suggest them using something like the time only down to the minute as a seed for the RNG (certainly a bad practice if so - but not one you'll notice from just playing one world). I never had the patience to do a true test of this theory though.

 

[Vger]

...only counting BPs received during RQs and GE, 61.2% of BPs received are for the Colosseum and Notre Dame. My spreadsheet goes back for over a year and I've been a meticulous recorder. This does not seem like a random distribution,

Do we know that it is supposed to be a random distribution? Does INNO say it is? Have others collected a large enough sample size to suggest it is? I'm too lazy to try to answer my own questions.
The fact that something is chosen at random doesn't mean there will be an equal distribution of the results. Think wishing wells. Do you get diamonds as often as you get coins? Of course not. INNO sets the probability of getting coins much higher than getting diamonds. It's done using the RNG, but they skew the RNG value to favor coins or supplies over diamonds. Nothing nefarious about that. Do they do that with BPs? I don't know...Like I said, I'm too lazy to google it and see if others have collected enough samples to say. Is it supposed to be a random distribution? Where is that documented?

 

[Vger]

It's the lousy random number generator they use. As you probably know, no computer rng is actually random, but this one is worse than many. I suspect they use it because it's free code, because if they are paying for it they are getting ripped off.

I call BS. Every programming language released in the last 20 years comes with a built in RNG that is more than good enough for this game. I would guess that INNO uses a built in RNG, but maybe not. Maybe they rolled their own (bad idea). Or maybe they bought one (probably another bad idea). It just probably doesn't matter.
Any RNG that passes a basic sniff test is going to be good enough for this game. Off the shelf RNGs are only a problem when you are dealing with ultra high security sites, where predicting the next RNG value gives you a 1 up. Can you predict what the RNG will do next?

 

[Super Catanian]

Sample size depends on intended accuracy. If you want to get values that are accurate to within 1% (with 2 standard deviation confidence level) then about 10,000 samples are needed.

*(cough)* @Algona*(cough)* TheGreatDiamondExperiment *(cough)*