@Kerridoc, @Brobb,
Now that I’ve written and reviewed this… wow, it’s long. Grab your favorite drink and comfy chair before diving in.
No, “stacking” is not a classical term in statistics! LOL
I’ve run enterprise class UNIX systems professionally for 21 years and we have our own lingo. As example, an asterisk is commonly called a “splat” in my industry/country/part of the world because it resembles a bug that’s been smashed… i.e. gone splat. An exclamation point is a “bang”, a question mark is a “huh”… I’ll bore you no more with our terms!
My apologies for spawning confusion: I guess I expected “stack(ing)” was a term others would figure out… maybe my industry’s slang is less globally understandable than I thought.
The easiest way to describe stacking as I’m using it, though this example is actually probability rather than true random, is the classic panel with evenly spaced pegs and a base with cups to catch the balls that are fed in through the top. Most will have seen this and how it yields a bell shape of filled cups over time. The key there is “over time.” 10 balls falling down to the cups will not produce stacking, whereas 1000 will.
Another is the coin flipping @Kerridoc mentioned. If we accept all the inconsistencies with mechanical flipping done by a human (thumbs, muscles, strength, physical and mental fatigue, human err, etc.), the numbers @Kerridoc gave are correct. Here too you don’t get true random output with 10 flips, but you do with many, many more.
Note that a mechanical device and environment can be constructed which will almost completely obliterate random from a coin flip and thus get consistent results.
Computers do not use mechanical methods of RNG, they use time (most typically) and complex algorithms that can only simulate random. To be absolutely blunt: random, generated from a computer of any sort is not actually random, it is merely complex enough to create the appearance of random.
On a UNIX system, there are 2 virtual devices required for RNG. If something goes wonky with one or both, the results are similarly wonky or even produce nothing. If those devices are destroyed and recreated with each RNG you actually get much more predictable results. The code that drives the RNG is also a factor in achieving the appearance of random, and there are more and less complex versions out there that produce better or worse randomization.
Android and iOS devices both run a version of UNIX, and though I can’t guarantee it because I don’t know those operating systems, I strongly suspect they both have the random and urandom virtual devices. Just how good is the RNG code in your device? In mine? Does the RNG happen on your phone or on the servers? How good is the RNG on the servers if that is where the magic happens (which is the most likely scenario)? Plain and simple, the poorer the code, the less random the output. The more powerful the operating system and hardware, the better the RNG code can be, with the converse also being true.
Finally, and most important to this discussion is that since a computerized RNG is not truly random, if each iteration of the RNG is 100% independent of all others, i.e. no “knowledge” of previous results, it has a much higher chance of returning recurrent results than a mechanical system. The random and urandom devices approach true random only with sufficient input. (cryptographic generators flood the virtual devices in order to create strings highly indecipherable without a key). It is extremely unlikely E&P does a similar flooding because it is more processor intensive than a single feed. There are what, thousands of concurrent users, and so thousands of calls to the code if it is done on the servers rather than on your mobile device. The developers must keep a game moving without bogging it down and so will likely only initiate one call to the RNG to return a result. E&P is not secure cryptography, it’s a game, so it doesn’t require that level of randomization.
Stacking, as I’m using the term, is simply the pattern of results that occur over many, many iterations of a randomizing system. A single run produces no stacking. 10 single runs produce no stacking.
I can professionally state that what @tkrasky described is indeed possible. It has a relatively low probability, but is fully possible if I’m right about the independent nature of the RNG used by E&P. Argue if you want, but I have 2 decades of experience and many, many conversations with the insanely brilliant minds at IBM to back it up.
Where you quoted me @Brobb, I was responding to @kahree saying no one had brought the concept of stacking in. I was trying to say that in this topic I saw people referencing stacking - the pattern returned from multiple runs of the RNG.
My preference, though clearly not in sync with @Brobb (which is totally OK), is to have some level of stacking to prevent the wild swings of output as described by @tkrasky. @Brobb doesn’t agree, but hey, let’s have a beer and talk our opinions until we’re absolutely positive neither of us can convince the other to change.
And, @Brobb, with my background it is quite literally not possible to offend me. It simply can’t happen. I do, however, insist on civility for myself and I will not immerse myself in anything but. I’ve seen far, far too many places where it is absent and if left unchecked the result is always chaos and unmitigated anger. I’ll stick with civility.
Peace out y’all!