The RNG and Dice Chess

[fgiannet]

I once tested max sizes of 50 and 80 (representing 50 tank and 800 man infantry battalions). One interesting observation, in between the frequent crashes, was how accurate the random number generator was over the long term. It was often within a few percentage points when calculating 50+ results. There were still statistically improbable results within those calculations (i.e. 5 consecutive results below <40, etc.) that would certainly give the impression the RNG was dishonest when playing with 10 strength units (“2 full strength units only get 4 hits with a 50% hit percentage?!?!”) but everything averages out accurately (with 50+ results) in the long term. Both perspectives, “the RNG is accurate/the RNG is statistically improbable”, seem to be correct.

I used to think dice chess was a crutch but now believe it might ultimately be more accurate. Timing the RNG can become a distraction (“25, 13, 16, 6......I might want to shoot some less critical units for the next two attacks until the results trend upward again”) and those anomalous results, where the a.i. consistently gets a hit (sometimes with only 1 or 2 shots) at 5% (tanks vs fighter aircraft...fighter aircraft vs. tanks....you know what I am talking about) can drain a lot of fun from the game (especially on Rommel). There is also the anomaly pointed out by Goose where aircraft always hit mine fields. There might be other such hidden mechanics built into the game but, in terms of hit/miss/suppression percentage, the RNG seems accurate in the long term. I was never convinced of this in the past. The company might see more sales, in future games, if they directly address RNG complaints with a demonstration of their accuracy (gamers can get turned off by real/perceived unfairness).

My old tinfoil hat conspiracy can be set aside now....there is no grand plan to lure players in only to change the results once we are hooked!

[rubyjuno]

Interesting. Thanks for explaining your tests and results. I switched to dice chess after playing Sea Lion and having a Spitfire surrounded by 6 of my fighters and attacked by each, all having a prediction of several hits, and the Spitfire escaping with 1 strength left. I was getting fed up of full random by then, although I suppose it is a more accurate representation of the randomness of war. To preserve my sanity, I prefer the (as you say) accuracy of dice chess, although sometimes the AI surprises me by handing out more damage than I would expect.

[fgiannet]

Interesting. Thanks for explaining your tests and results. I switched to dice chess after playing Sea Lion and having a Spitfire surrounded by 6 of my fighters and attacked by each, all having a prediction of several hits, and the Spitfire escaping with 1 strength left. I was getting fed up of full random by then, although I suppose it is a more accurate representation of the randomness of war. To preserve my sanity, I prefer the (as you say) accuracy of dice chess, although sometimes the AI surprises me by handing out more damage than I would expect.

That deficit of hits during your Sea Lion Spitfire engagement would have been made up in other attacks later on but might be wasted on something as simple as finishing off a 2 strength enemy unit (and receiving 8 out of 10 hits at 40%). Those missing hits would be, essentially, thrown away in that case. Dice Chess keeps those hits allocated to the units that statistically should be delivering them (rather than giving your Me-109 hits to a later unit such as a battleship firing on a destroyer).

I agree with you in terms of the randomness of war and can enjoy the randomness with certain setups (playing with historical OoB’s and raw troops for instance). Some setups are very limiting and a fickle RNG could be the deciding factor instead of tactics (even though those missing hits will be made up later with other units). It is not a chronic issue though.

It is funny that you switched to dice chess during Sea Lion. I put the game down for years after playing roughly 10 turns into Sea Lion and seeing some ridiculously improbable results. It reminded me all too well of absurd results given while playing Panzer General nearly 20 years earlier (experienced Wehrmacht HW being eliminated by Soviet conscripts, 8 strength AT gun in a city inflicting 8 hits on attacking engineers, etc.). I just assumed it would continue to get more ridiculous the more you progressed in the game (in order to keep you playing).

Seeing the statistically improbable results is what annoys me the most (I should probably stop looking ). It is too similar to being told that 2 + 2 = 3 (which also seems to be more and more common these days ).

[LevV]

Did i got it right - you are saying that for each bad result there would be a good result later on, that might appear on situation, were its absolutely unnecessary? The way you put it is seem to be beyond the basic statistical variation that evens out on it own (which it probably should be, essentially), to the point of good hits almost stored somewhere...
If i were to join this conspiracy, i would ask if save/load can break/restart the alleged chain of events? Or perhaps restarting entire game at all?
On a flip side from my own MP experience especially in big scenarios where over a single turn you can be making tens of attacks, sometimes it does seem like a bad day for entire group of veterans who can barely scratch an absolutely vulnerable target.
Overall i think its a perception issue. People tend to be excited about disturbing things, such as bad RNG results. I dont think good hits can be stored for you , its just a relatively good day after relatively bad one (or vise versa by the way, depending on how you look at it).
Also the big deal is that the predictions (and i think some calculation factors as well) are rounded up, so you would see for example 1 hit for 7 strength unit and 1 hit for 10 strength unit, which you should understand practically have a potential to differ. If i need one important hit im looking for most powerful unit to carry it out, not just anyone whos prediction shows that 1.
From my experience and understanding of dice chess it just has the variation limited to +1/-1 , thats why it dont feel so much disturbing.

[rubyjuno]

It is too similar to being told that 2 + 2 = 3 (which also seems to be more and more common these days ).

I know what you mean

As to LevV's post, in my experience save/load/restarting can indeed change the chain of events, although, like LevV, I'm not so sure that a bad result is balanced out later by a good result. This would lead to more questions - what if the bad result (or good result) is too late in the scenario for the balance to be applied? Just a thought. What a fascinating discussion - it's great that after all these years Panzer Corps is still promoting such interest.

[fgiannet]

Did i got it right - you are saying that for each bad result there would be a good result later on, that might appear on situation, were its absolutely unnecessary? The way you put it is seem to be beyond the basic statistical variation that evens out on it own (which it probably should be, essentially), to the point of good hits almost stored somewhere...

The hits/misses/suppression results are just categories representing a range of numbers. I do not believe there are hits stored for later use because the RNG is only dealing with numbers relating to it’s prior results (i.e. 10% of a total number of results must be above 90% for 50 rolls). I doubt the RNG even evaluates hits or misses (it might simply provide numbers to the main program).

I am using the terms hits/misses, probably inappropriately, because that is how we view the results but, ultimately, we are talking about probability (the probability of 20% of the results being above 80 as an example). The RNG can give statistically improbable results for 10, 20, or even 30 attacks in my experience. That might mean 80% of the numbers are below 50%. But it will return a higher portion of numbers above 50% in subsequent attacks in order to average all the results out (and I agree with RubyJuno that the results are not saved so restarted games erase prior results).

Your aircraft might have a lot of low rolls, using RubyJuno’s Spitfire example, but the RNG will make up the statistical anomalies during later unit attacks. 3 Me-109s attacks may return 80% of the numbers under 50% but the next attack might return 50% of the numbers above 80% as it begins to average the previous results upward (and it does appear to be about averages instead of modes, etc.). They will most likely be hits (due to their high values) but not definitively (conscripts vs Tiger Tanks in clear terrain as an example). The RNG is unconcerned with hits, misses, etc.

I agree with RubyJuno in regards to the randomness of war. My issue is with the statistical improbabilities that can happen when the RNG is trending downward. Seeing things like 5 consecutive results below 40% (about 1% chance of that happening), 4 consecutive results below 20%, 3 “9’s” in a row, etc. It is frustrating to see even knowing a statistically improbable high value set is in the future (and it may be just one high set, limiting it’s utility, making up for three low). There is no such thing as a real random number generator program (unless we are dealing with quantum computers). All these values are most likely derived from math operations using other numbers (probably primes). There are ways to ensure less anomalous results but those would demand more resources as well.

I did see, with 50 strength units, 10 results in the top 20%, 40 results at 80% and below. The distributions were nearly perfect. Within those 50 results, however, there might be 10 or 20 consecutive results containing statistical anomalies (only 2 results over 60% in a 10 result set, etc.).

The RNG seems to provide nearly perfect distributions after 50 results but can give statistically anomalous results in the short term. It seems accurate to describe dice chess as playing with the same numbers without those anomalous results. The RNG is very accurate in the long term and seems to produce the same distributions as dice chess over an extended period of time.

[rubyjuno]

Fascinating reading again. As you say, over time randomness will generally even itself out, but it's those times when you seem to get a bad run (back to the Spitfire again!) that it seems unfair. Not that war was ever fair...

[LevV]

the roots of this topic go far outside the usual gaming questions. i never gave a thought to what is real random or real random number generator, i thought it works everywhere just like Excel formula does. i think i now get a clearer understanding what was meant by fgiannet. in my current big MP map i had several bad turns where i think i experienced something like anomaly consecutive low results. on the other hand i recall time i played 1941 with somebody and first Axis turns for me were just fascinatingly successful (they are meant to be by definition in that case, but knowing how unlucky you can get its super exciting to see PzIV destroy entrenched conscripts without loosing ANY of those expensive over-strength points).
i dont think devs will ever react to this matter let alone make their RNG generator "real" or transparent for this game of the past, so i guess we are only here to find out what can we do to avoid frustration.
i see following questions and i think i'll try to find time for experimenting on my own:
1. can approx number of rolls be defined after which the anomalies tend to appear. for example is it possible to formulate say : "after 8-10 rolls you are suggested to quit and re-enter" ?
2. what other in-game moves interfere with the streak: buying /deploying / upgrading units? rugged defence rolls? unit movement? weather rolls?
3. is there personal streak for each player or is there continuous scenario streak?

on a flip side we all need some frustration in our life to motivate us try harder and upgrade our methods (i.e. buy more arty)

p.s. with a big map like 1939 or 1941 my other personal challenge is after how many attacks the game graphics crashes. as a rule of a thumb on a good weather day (air and navy in their prime) if you dont quit/reload after doing all most desired moves you have a pretty good chance to be eventually forced to re-do it all again (because of crashed game).

[fgiannet]

but knowing how unlucky you can get its super exciting to see PzIV destroy entrenched conscripts without loosing ANY of those expensive over-strength points.

I think you are accurately identifying an important but understated part of the fun: random results may simulate gambling (or produce a dopamine rush similar to gambling). You are rolling the dice and hoping for a good result every single time you press the button to attack. There is a dramatic difference when playing with dice chess or chess. I have started a game with dice chess and can already feel the lack of excitement when attacking (but am enjoying other aspects more). Who gets the initiative bonus is a large variable during random as well (further increasing the swings of fortune). There can be wild swings in fortune during war (Napoleon once famously said he valued lucky generals above all others). I do not mind the randomness itself. I just can not stand looking at the ridiculous numerical results that can happen far more often (statistically) than they should.

Speaking of gambling......"near misses" are psychologically far more addicting than either "misses" or "hits" so companies are incentivized to curve results lower (50% is actually 40% or even 25%) in order to keep the person playing. I have seen many people spend a lot of money in order to reach different bonuses in slot machines. They spend more money than they will receive in the bonus stage but can not help themselves because they were "so close to getting it earlier".

1. can approx number of rolls be defined after which the anomalies tend to appear. for example is it possible to formulate say : "after 8-10 rolls you are suggested to quit and re-enter" ?
2. what other in-game moves interfere with the streak: buying /deploying / upgrading units? rugged defence rolls? unit movement? weather rolls?
3. is there personal streak for each player or is there continuous scenario streak?

1. It seems to me that an abnormally good roll will always come after 3 consecutive bad rolls. Sometimes an abnormally good roll follows one bad roll. It also seems to give more abnormal results the longer I play (so occasionally quitting and re-entering is probably a good idea....hmm interesting....don't know why that has not occurred to me before ).
2. It does not seem to be affected by other in-game moves (unlike other games). Moving other units would change the results in Battle Academy but they probably used a different system because their results would persist if you quit and re-entered.
3. That is a good question. Sometimes, when playing with a historical OoB containing inferior equipment, I will simply stop attacking. The a.i. attacks me and seems to do worse than when I am attacking them. It seems like defense might get a different set of numbers from the attacker. An excellent question......

on a flip side we all need some frustration in our life to motivate us try harder and upgrade our methods (i.e. buy more arty)

I agree in principle but want the right kind of frustration. I want frustration caused by engaging challenges that require strategy rather than the frustration experienced by having to send the entire Luftwaffe to deal with one U.K. Squadron ("but they were Spitfires!" Yes, yes I understand ).