Originally Posted by chillinfar

I understand. Due to low quantity of objets to permute (6 on PPS), that random system can place answers in common positions. The only what is possible to calculate with current data is the frequency of some results on certain timelapse. And it isn't ensure that the next answer will be the correct.

We can't know how PPS is managing this (i saw your demo decompiling main program), if is pure random or has a pattern (or mix both to make the shape that has a pattern, fooling the player as you guess). Results are the only data available to work and may be different for each player.

I'll try to see pos of gachas later. I was seeing this many times, but i never wrote this to a text file.

I agree we can't accurately know how they do this. However, we can put more weight in the more logical or common solutions for a game such as this.
The main point to remember is that the server behind this isn't made to handle your game, but a couple million JP players, sending service requests at least every couple of seconds.

That means the server needs to be able to balance a load of possibly hundreds of thousands (or even millions) of requests per second.
In such kind of architecture, you tend to try and minimize any superfluous bit of logic that is pretty much useless to you...

As such, why add some weird rotating pattern to your code, when just randomizing crap works the same?
Applying a pattern requires keeping the pattern stored somewhere (either hardcoded or in database), fetching it, mapping the input data into something that can be put through the pattern, applying the pattern, mapping the results to the required response, and only then returning the value.

Just to make a comparisson:

The simple randomizer works like this:
[LIST = 1][*]Fetch six RND decimals between 0% and 100%.[*]Apply step-ups.[*]Map the 6 results to the "Rarity Table".[*]Fetch a new RND decimal for each mapped range.[*]Fetch data from the IDed cards.[*]Build and return JSON object.[/LIST]

Against a pattern:
[LIST = 1][*]Fetch the pattern list.[*]Fetch a RND decimal to choose pattern.[*]Fetch the chosen pattern.[*]Fetch six RND decimals based on the pattern.[*]Map the 6 results to the "Reward Patterns Table".[*]Fetch a new RND decimal for each mapped range.[*]Fetch data from the IDed cards.[*]Build and return JSON object.[/LIST]

As you can see, patterning is a more complex process.
Also, all fetches and mapings that aren't for RND values imply a call to a database, which can take up to 10k times more to process that a logical process. That's 2 for RND vs 4 for Patterns. So, not only is it more complex (meaning more processing times, more RAM needed, more caching calculated, etc) it actually has twice the delay times corresponding to DB accesses.

All that for something that just having a good randomizer will do equally as good? For no apparent reason besides we wanted a pattern in there?
That's just no good for implementation times, manteinance costs, server costs, etc.

Outside from that discussion i found something interesting: Trends ("tendencias"), and something in statistics named in spanish as "moda" (the most common elements from gathered data). I apologize to don't explain this correctly and confuse it with patterns.

In gacha as example, one R/SR card can't take the same position more than 3 times on consecutive turns. This happens even when other R cards are available, try it.

Extra stage from gambit event showed an opposing example, i never found a 2;6 entries as correct ones, but i saw 2;4, 2;5, 1;4, 3;6, 1;2, 1;6... and even 2;4 (mirror of the "missing" element). Post a screenshot if you see one.

This is not related with the random system, this time i'm discussing it's results. However, seeing that behavior can help on some situations.

Originally Posted by chillinfar

Outside from that discussion i found something interesting: Trends ("tendencias"), and something in statistics named in spanish as "moda" (the most common elements from gathered data). I apologize to don't explain this correctly and confuse it with patterns.

In gacha as example, one R/SR card can't take the same position more than 3 times on consecutive turns. This happens even when other R cards are available, try it.

Extra stage from gambit event showed an opposing example, i never found a 2;6 entries as correct ones, but i saw 2;4, 2;5, 1;4, 3;6, 1;2, 1;6... and even 2;4 (mirror of the "missing" element). Post a screenshot if you see one.

This is not related with the random system, this time i'm discussing it's results. However, seeing that behavior can help on some situations.

Ahh, good old math for probabilities
Be careful... when I said there are no patterns, I meant there are no programmed patterns.
Do something in a computer enough times, and a statistical pattern will emerge.
Sadly, such a thing won't be found by us checking a couple of accounts a couple of times. We're talking about compiling enough data that we could more or less find a pattern in the RNG generator

Those measures aren't detected on 1 day but weeks (except pero gambit, but thanks to stamina i was seeing a lot of results on last and extra stages). It still be a small part of data to detect a potential "pattern", so i prefer to use the other (and realistic) term, trends.

And trends can change over time. However, in Gacha if we simplify the elements to type of cards we got a nice scheme to find a good chance hunting R cards. Remember that N gacha always have (at least) 1 R card, we are lucky to see this condition, which is not present in GHQ or stages.

EDIT: Finally a 2;6 in event.

Except, N-Gacha don't always have one R-card. The base for N-Gacha is 6xN. It's just that the probability for having one of those be R is too high to not have an R in there most of the time.

->Open Spoiler<-

Got this one about 5 minutes ago

And trends don't differ from normal statistics. They are most definitely not used as a base for calculations.
Trends are used when the data you are meassuring is of a highly mutable type, and thus won't allow you to build a test sample big enough for what you need (or make building such sample a very fast paced and expensive process) and then invalidate it in too short a time.

When such a thing happens, you build the best sample you can in your "non-mutable" lapse of time (the lapse of time the data won't shift so much it invalidates your previous data).
Onle then, you can apply/study trends, by repeating that sample building all over again, and analyzing how the two samples change lapse after lapse (and thus, finding out the way the changes might happen over time).

The whole scientific set of calculations for trends give you the tools to do similar calculations as base statistics on said shifting data. But they won't give you anything accurate if you didn't have any starting data, Just starting from 0 with trends will fail as much as using common statistics on shifting data

In card list says 96% chance. That's why is very uncommon to see a 6N deck on N gacha (i found it ONCE in the whole time that i'm on this game).

that pic is from GHQ though xd