Just how random is the 'rng'?

[rcoutme]

I have just spent some time collecting the results of some combats. I would like one of the mathematicians who reads stuff in this forum, if you would please, to check out the results to see if they are skewed. I suspect that the rng is, in fact, skewed to unusual results (i.e. that more 'unusual' things happen than usual).

The following results occured in a game that I was watching (on debug mode where I was the gods and not participating). They are all AI-AI battles. The results are always posted as winner (hp lost) vs loser (hp lost). All terrain and fortification bonuses are factored in.

3.3(0) vs 3(3); 8.4(0) vs 1(3); 2.2(0) vs 4(3); 2(0) vs 4(3); 2(0) vs 2.2(2); 4(0) vs 2.2(2); 4(1) vs 2.25(3); 2.5(0) vs 4(4); 3(2) vs 3.3(4); 3.3(0) vs 2(3); 3.3(1) vs 3(3); 3(1) vs 3.3(2); 4(0) vs 4(4); 4(0) vs 4(4); 4(2) vs 3.3(3); 4(0) vs 1.35(3); 4(2) vs 3.3(4); 4(0) vs 1.1 (3); 4(1) vs 1.1(3); 2.2(0) vs 4(1); 4(1) vs 1.1(3); 2.2(0) vs 4(1); 4(1) vs 2.5(2); 4(1) vs 2.1(3); 1.1(1) vs 1(3); 2.2(0) vs 2(4); 2(0) vs 1.1(4); 4.4(1) vs 3(4); 4.4(1) vs 4(4); 4(0) vs 1.6(3); 4(0) vs 1.35(4); 4(1) vs 2.2(4); 4(0) vs 3.3(2); 1.35(1) vs 4(4); 4(2) vs 1.35(3); 4(0) vs 1.1(4); 4(1) vs 2.5(2); 2(2) vs 2.2(4); 4.05(1) vs 3(3); 4.05(1) vs 3(4); 1.1(3) vs 4(3); 2.2(2) vs 4(3); 3.3(2) 4(3); 3(0) vs 1.1(2); 3(3) vs 4.05(4); 4(1) vs 3.3(3); 4(0) vs 3.3(1); 4(0) vs 2.7(3); 1.1(1) vs 4(4); 3.3(3) vs 4(4); 3.3(2) vs 4(3); 4(2) vs 6.7(4); 2.7(2) vs 3(4); 3.3(0) vs 4(4); 4(3) vs 2.2(4); 4(0) vs 3.3(4); 4(3) vs 3.3(1); 4(0) vs 3.3(3); 1.1(2) vs 4(3); 4(1) vs 2.7(3); 4(1) vs 3.3(5); 4(2) vs 1.1(3); 4(1) vs 1.1(4); 3(2) vs 3.3(5); 3(1) vs 4.05(4); 4(1) vs 1.5(3); 4(3) vs 2.2(2); 4.05(1) vs 3(3); 4.05(0) vs 3(4).

To understand me, I realize that all of these combat results are possible, even likely. My question is: are they likely together? It seems that a lot of less likely results occur while the more likely ones are less represented. The reason that I chose the middle ages combats is because there are a nice, wide range of combat values.

All of these results came up in succession (i.e. I did not pick and choose the results, I just recorded them).

Edit: as I look at these results together, one of the striking things that I note is the huge number of times when one unit (the winner, obviously) never took any damage. It seems to me (from what I remember of statistics) that most of the results should be attrition, not one-sided victories.

 

[fret]

I imagine if you want to test the rng, then you should replay exactly the same battle a few thousand times and analize those results.

To analize the statstics when you are bringing in so many other variables with differant combat values etc will probably make statistical analasys a whole lot more difficult and unreliable.

I think people have allready proved that its random.

 

[Colonel]

AHH lots of numbers AHH everyone run attack of the numbers

have fun with anyalzein that

 

[warpstorm]

The RNG has been tested numerous times with very extensive tests (thousands of units). It works as well as any linear congruential one can be expected to. Streaks happen. If they didn't it wouldn't be very random now, would it?

 

[Longasc]

If I would not have been so bad in my latest statistics test, I would give this to Professor Olbricht...

FACT: Many people feel the RNG is not as unbiased and random as it should be. The percentage for a significant (5% error) result should be reached by a few 100 battles, no need to do a lot more of testing IMO.

 

[The Last Conformist]

Plenty of people fint it's too random, in that they want fewer streaks.

 

[warpstorm]

So, you are saying they want a smoothed RNG rather than actual random numbers. Given enough trials a true random number generator will have streaks of any length you can imagine.

 

[rcoutme]

Ok, I will accept that it is random. Please keep in mind, however, that I decided to compile the #'s because I have tended to see lots of times when one side or the other dominates. It could be streaks (I know enough math to know that the above is possible), however, lots of people in this forum have suggested seeing what might amount to abberrant results. I will bow to those who know better and assume that when I see strange results (as I said, I did not pick a group of battles, I just started recording them yesterday for the hell of it) that it is just that, strange results, although quite possible.

 

[The Last Conformist]

So, you are saying they want a smoothed RNG rather than actual random numbers. Given enough trials a true random number generator will have streaks of any length you can imagine.

That's want they appear to want, yes. Humans fairly consistently underestimate the streakiness of true random sequences.

Edit: For the record, I am perfectly happy with the present RNG, and would hate to see streakiness reduced.