One other assumption about COH's use of the PRNG that a lot of people made was that random numbers were specific to them -- that if the RNG got seeded with a particular value, such as the clock, then their attacks all came from a specific sequence based on that seed. That couldn't be further from the truth. The same random number source was used not only for player attacks, but for dozens of other things as well. Other players, enemy critter to-hit rolls, pet to-hit rolls, some randomness used by the AI in its thinking, choosing which animation to play if there are multiple equivalent possibilities (some brawl variants had this, the dice roll emote used it, etc), reward rolls if anything on the same map was defeated.
By the time you queued up another attack, dozens or more random numbers had been generated in the interim. Indeed the unpredictable nature of when random numbers were needed and sampled from the PRNG introduced a lot of real entropy into the system, entropy fed mostly by player actions and amplified by complex interactions of systems. It wouldn't surprise me if that's part of why Arcana's statistical analysis showed a good distribution despite the underlying system being technically deterministic.
To be honest, even when this was pointed out on the game forums, it was suggested that this very fact could have made testing impossible to determine the flaw in the RNG, because a pattern in the RNG could be obscured by the fact that a test would not see it most of the time. It was difficult to respond to this objection in a satisfactory way although basic statistical mechanics says this is not really possible over long testing periods.
Here, however, what I would say is that so long as the RNG generated a sequence of numbers that showed no obvious correlations between n-consecutive generated numbers for sufficiently large n, and showed no signs of generating any particular number over any other, that the added complexity of sampling that generator at pseudorandom intervals cannot make the sequence look significantly more or less random. To put it more simply, combining two PRNGs does not tend to make the combined system more or less random. So I assume that if it looked random to me, it probably was and it if wasn't, it wouldn't look so to me, even with the added complexity of sampling.
Depending on the PRNG algorithm used and if it was cyclical (most decent ones are not), re-seeding the generator might have been completely unnecessary and was probably never done at all once the server was running.
Not just unnecessary, but also a potential source of non-randomness. Unless you reseed with a genuine source of good entropy, the act of reseeding itself could in some cases make the generated numbers less than random.
There is only one confirmed instance of an issue with random numbers in COH I can think of: purple recipe drop rates were observed by some players as being far lower than they should have been, and it was brought to developer attention thanks to the tireless efforts of players parsing logs and documenting tens of thousands of drops. I forget which red name posted the final breakdown, but it turned out to not be an issue with the RNG at all (which was functioning perfectly), but rather a bug involving an uninitialized variable in the rewards code. Said bug caused players to sometimes be incorrectly disqualified from the purple drop table, depending on what ended up in a specific memory address. I think that bug may have persisted for nearly a year before it was squashed.
There was one more issue like that I'm aware of, and one actual RNG error (the only one I'm aware of). In I9 beta, I confirmed that the dev statement that Pool B drops from mission complete 10% of the time was in error and actually occurred at about 7.3%. But that wasn't a problem with the RNG itself but rather a problem with the way the drop tables were constructed. The other was the random drop map distribution error: players discovered there were certain maps where the rewards generated were actually predictable with a high degree of certainty, if certain gameplay conditions were met. I analyzed the player posted data and took that one to pohsyb directly, and he determined that in that part of the code some shmuck wrote their own RAND that was extremely not random and often reseeded with a constant: that made those maps generate the same basic layout much of the time.
Separate from confirmation bias, there's also what I recently read referred to as the Wyatt Earp effect and I used to call the lottery winner effect. Players, even those with mathematical backgrounds (actually, especially those), would often miscalculate how rare an unlikely occurrence was by misunderstanding what an unlikely occurrence was in the first place. For example, if you take six six-sided dice and rolled them, it sounds extremely unlikely they would come up 1, 2, 3, 4, 5, and 6. But if you roll six six-sided dice and you are of a mind to over-recognize patterns you might have said the same thing about the dice coming up all ones, or all sixes, or all even numbers, or all odd numbers, or two each of 1, 2, and 3, etc. The odds of coming up all sixes is 46,656 to one. But the odds of the dice coming up *interesting-looking* is a lot higher, because people categorize a lot of different possibilities as interesting.
The Wyatt Earp effect is the effect of thinking something extremely unlikely - in this case Wyatt Earp surviving all of his gunfights - must be due to something other than chance. The problem in this case is that enough people got into gunfights back then that the odds of *one* of them coming out of many gunfights alive is a lot higher than the odds that Wyatt Earp specifically would have done so. We only talk about Wyatt Earp because he happened to be the one that did. The odds of winning the lottery are very slim, but someone always does. That doesn't mean Wyatt Earp wasn't a good gunfighter, it just means we have to be extremely careful about what we attribute to chance, and what is postscript selection that has nothing to do with chance.
To be statistically blunt, the odds of Wyatt Earp surviving all his gunfights is 100%, because we already witnessed it. We cannot calculate what it "would have been" we can only calculate what the odds of another random person surviving all of those gunfights might have been, and that is a subtle but significantly different thing.
Even when the player correctly notes that something they witnessed was unlikely, that doesn't mean it wasn't actually likely that *someone* saw it, and it just happened to be them. Ten misses in a row is unlikely. But so is ten alternating hits and misses. So is five hits followed by five misses. So is three hits followed by three misses ten times repeatedly. If you want to see something unlikely, the fact is every individual sequence is unlikely to occur *again*. But something has to happen.
