CMBN games seem to be more imbalanced...

[BletchleyGeek]

so 'unbalanced' really means 'balanced on a knife edge'. The slightest nudge will make the result fall drastically one way or the other.

Not that anything you've said about the nature of "game balance" is untrue, it's just not quite what GaJ was concerned about at the top of the thread.

There are many ways of 'balancing on a knife edge', one are scenarios like those late 1944 CMBB QB's, the other is "balancing" the scenario so winning/losing probabilities would be 50/50, even with "unbalanced" forces and asymmetric goals. Both kinds of scenarios would tend towards the nice CMAK Gaussian Distribution GaJ points us at. I just wanted to give concrete examples of scenarios - and the amount of work they require - which could lead to such a distribution.

EDIT: Indeed, assuming that both players will play "perfectly", whatever that means.

I would however disagree that we'll be able to "...develop scenarios with a quality near the best developed for CMx1." I'd say that it's possible with the broader set of tools available to develop scenarios even better than the best in x1

Hehe, yes you're right. But it's up to us to prove that true, right?

[womble]

There are many ways of 'balancing on a knife edge'...

Indeed. But it's the knife edge that's started this thread, not the 'balance' itself.

Both kinds of scenarios would tend towards the nice CMAK Gaussian Distribution...

Not necessarily. And those factors which drive the curve into other shapes are the ones in question.

I just wanted to give concrete examples of scenarios - and the amount of work they require - which could lead to such a distribution.

And of course, you're quite correct that it takes a lot of work to make a scenario balanced, especially if you have mildly divergent goals (can't be too divergent, or each force will just scuttle to its own objectives and wonder where the enemy's gotten to ). The difference GaJ is pointing up is that even if there's a 50/50 chance (whatever that may mean ) for either side to win, in the data from BN, that victory currentlt seems more likely to be a Total one than a Minor Tactical, compared to the data from AK. Even if the difference between the two is one AT Rocket hit or a slightly tighter mortar sheaf. In your example, the convoy might get half of its units away, most of the time, in x1, giving a draw or minor vic to one side or the other but in x2 it will either get the vast majority off, or virtually none, leading to Major or Total victory for the winning side. While you can design with the effects that cause this distinction in mind, you have to identify them first.

[Vanir Ausf B]

There have been several good explanations put forth here, all of which I think are contributing factors. If I had to pick which are the largest factors, I would say smaller battle size + longer battle length. The fewer forces you have, the more significant each event becomes. Losing a tank or having an arty barrage land on a concentration of infantry is more difficult to recover from with fewer reserves. The snowball effect builds quicker. Longer game length allows time for the advantaged side to pummel the other side into the dirt.

[BletchleyGeek]

And of course, you're quite correct that it takes a lot of work to make a scenario balanced, especially if you have mildly divergent goals (can't be too divergent, or each force will just scuttle to its own objectives and wonder where the enemy's gotten to ).

Indeed, in order to be some "conflict" there must be "mutually exclusive" goals for each side

The difference GaJ is pointing up is that even if there's a 50/50 chance (whatever that may mean ) for either side to win, in the data from BN, that victory currentlt seems more likely to be a Total one than a Minor Tactical, compared to the data from AK. Even if the difference between the two is one AT Rocket hit or a slightly tighter mortar sheaf. In your example, the convoy might get half of its units away, most of the time, in x1, giving a draw or minor vic to one side or the other but in x2 it will either get the vast majority off, or virtually none, leading to Major or Total victory for the winning side. While you can design with the effects that cause this distinction in mind, you have to identify them first.

I don't really see - yet, I'm still adjusting the map for that scenario - any reason intrinsic to x2 modeling which would make more likely the extreme result you mention. Do you have anything in particular in mind?

[GreenAsJade]

I had in mind the lack of cover, primarily.

The mention of scenario length is a good one too.

Neither of these are fundamental or "about the CMx2 engine". They seem to be more about the scenario design.

I think that this is a very very likely explanation for a large part of what we are seeing.

BUT then again ....

I made graphs of the first 80 results of CMAK, CMBN and CMBO, first by date. You would think that we were all learners at CMBO ...

http://gregories.net/bobster/bobster.cgi?function=scenario-spread-histogram80

... and yet when we were learning CMBO, we fought close battles. In the first 80 CMBO games played, there were negligible "Total Victories". Compare to CMBN, where there were a large proportion of the games with this wide margin.

Figure that one out!?

For some reason CMBN battles seem to be more precipitate...

GaJ

[Other Means]

Sorry – late to the discussion.

But I also have a suspicion that the current mechanics of CMBN are contributing. In my limited experience, the CMBN battlefied seems way more harsh. There is just hardly any cover anywhere. This appears to mean that the guy who spots the other guy's bulk of force first and manages to shoot at them gets rolling advantage. Once you're on the back foot, there's no-where to hide...

GaJ

Lack of cover yes – but also it feels to me like there’s a lack of concealment. Spotting seems too quick to go from sound to visual contact. And then, once you are spotted fire seems too deadly. Especially against buildings.

I don’t think they’re far off, but I do think they may still need tweaking. Maybe when more games have been completed while patched we’ll get a better feel for it.

[BletchleyGeek]

I had in mind the lack of cover, primarily.

The mention of scenario length is a good one too.

Neither of these are fundamental or "about the CMx2 engine". They seem to be more about the scenario design.

BUT then again ....

I made graphs of the first 80 results of CMAK, CMBN and CMBO, first by date. You would think that we were all learners at CMBO ...

http://gregories.net/bobster/bobster.cgi?function=scenario-spread-histogram80

... and yet when we were learning CMBO, we fought close battles. In the first 80 CMBO games played, there were negligible "Total Victories". Compare to CMBN, where there were a large proportion of the games with this wide margin.

Figure that one out!?

For some reason CMBN battles seem to be more precipitate...

So skill level doesn't fully account for the win/loss distribution. I find surprising that the influence of skill level is so moderate - by comparing the distributions for CMBO and CMAK.

Can you get separate graphs for CMBN scenarios played at least 4 or 5 times?

[Vanir Ausf B]

It may be instructive to separate QBs from scenarios. That would give you a better idea of scenario design vs. game mechanics.

[womble]

It may be instructive to separate QBs from scenarios. That would give you a better idea of scenario design vs. game mechanics.

Dunno. You're still having to account for 'first gen' CMBN maps which might exacerbate things as there's not much experience, and they're more complicated beasts to make than x1 maps.

[TexasToast]

My theory is that the difference is due almost entirely to the nature of the scoring systems. In CMx1, you always scored points for enemy losses and lost points for friendly losses. Since both sides generally took losses and those losses formed a significant part of the total score, the results tended towards the center. With CMx2 scoring, there is often not a penalty for losses at all, or perhaps just a bonus for keeping friendly casualties low. Therefore the final score is more extreme for the same battle action as in CMx1.

TT

[Erwin]

CM2 games seem more unbalanced or "unstable" than CM1 games in the same way that an old prop plane is "stable" but a modern jet fighter is "unstable" and has to have computers to assist in flying. We have to recognize that CM1 is a rather different game from CM2, so it's not reasonable to compare them too closely.

About the minimum game length I played in CM1 in the last years was about 30 min + up to 6 or 7 mins overtime. However, generally, CM1 games are much larger (map size as well as number of units) than CM2 games.

In CM2, because there are fewer units, the loss of one or two critical units can create an "avalanche of disaster" effect for a player. In a larger CM1 game where you often have companies of inf plus a company or more of armor/vehicles and other assets, the loss of one or two units is rarely critical. It is usually much easier to bounce back from a loss in CM1 than in CM2.

A typical CM2 scenario reminds me of a tiny 500 point CM1 game in which one usually had one or two really important units and if you lost them, you were toast.

Of course there are exceptions to all of the above, but I reckon that is why CM2 is seen as "unbalanced" compared to CM1.

[Calibration]

From GAJ's original graphs I estimate that

The probability that the CMBN distribution has the same shape as the CMAK distribution is <1%

The probability that the CMAK distribution is unbiased is <0.00001% !!

In the latter case I compared the observed distribution with a theoretical distribution having a mean result of 50% and the same width as the observed. The second result is far more significant than the first because of the sample sizes involved. As this discussion shows, how you interpret these results is open to question. For example, does the Axis bias in CMAK indicate the bias is in the game engine or that stronger players tend to favour the Axis when they get the chance to pick sides for whatever reason? But anyway, this result is not in the least surprising as it would be almost impossible to devise a truly unbiased CM type game.

[BletchleyGeek]

In CM2, because there are fewer units, the loss of one or two critical units can create an "avalanche of disaster" effect for a player. In a larger CM1 game where you often have companies of inf plus a company or more of armor/vehicles and other assets, the loss of one or two units is rarely critical. It is usually much easier to bounce back from a loss in CM1 than in CM2.

A typical CM2 scenario reminds me of a tiny 500 point CM1 game in which one usually had one or two really important units and if you lost them, you were toast.

I understand that you refer to a "typical CM2 QB". In that case, I agree that forces usually featured on QB's in CMBN are more "brittle" because of reduced numbers.

[womble]

My theory is that the difference is due almost entirely to the nature of the scoring systems. In CMx1, you always scored points for enemy losses and lost points for friendly losses. Since both sides generally took losses and those losses formed a significant part of the total score, the results tended towards the center. With CMx2 scoring, there is often not a penalty for losses at all, or perhaps just a bonus for keeping friendly casualties low. Therefore the final score is more extreme for the same battle action as in CMx1.

TT

There's always a proportion of casualty VP in QBs. Not always in scenarios, true, but QBs have a specified terrain/units VP split that varies with battle type.

[ClarkWGriswold]

Speaking anecdotally about the PBEM games I've played against a couple of friends, I've found that while the game results might show a decisive victory or total victory, the battles themselves are generally much more even and hard fought than the results would lead you to believe.

For example, I just finished a game where virtually my entire force was wiped out. I had one HQ unit that hadn't taken casualties, one Jeep crew, and one Sherman left at the end (along with a smattering of surviving tank crew members). All of my other armor (9 tanks and a halftrack) had been destroyed, and very few crew members survived. But I still won a total victory because I was able to destroy or disable enough of my opponent's forces that he could not drive my few men out of the victory location. He ended up having no choice but to surrender. A total victory for me, even though he had fewer losses in terms of vehicles and men.

An earlier game against the same opponent ended with me having three halftracks and a handful of engineers, while his entire force was wiped out. I estimate that I had 85% of my original forces KIA/WIA against his 99%. Even so, I won a total victory because I had a few guys on the VL and he had nothing left.

Two different games that I ended up "winning", even though my own forces were nearly wiped out and the battles themselves were very, very close.

[slysniper]

There have been several good explanations put forth here, all of which I think are contributing factors. If I had to pick which are the largest factors, I would say smaller battle size + longer battle length. The fewer forces you have, the more significant each event becomes. Losing a tank or having an arty barrage land on a concentration of infantry is more difficult to recover from with fewer reserves. The snowball effect builds quicker. Longer game length allows time for the advantaged side to pummel the other side into the dirt.

I AGREE WITH THIS COMMENT.

I think the game would give you similar results if the scenarios were designed with the same concepts and scoring as the older games.

If you set the scenarios with units losses counting as points and simple scores at lower values for objectives, the end scores would be closer, the time lengths seem long to me in many scenarios. If they are too long, almost always one side will clear the map of the enemy.

With the new tools, the games scoring reflects big wins with battles that would be close in the old system.

It does not take a chart to know the end results are more drastic now.

I am playing the same guys I have been plaing for years, old system - very rare to get a 70-30 win spread. new system. crushing victories with the loser scoring very low points in general.

The one thing impacting the game more now is the way morale works, once you truly start losing on the battlefield. It is hard to recover because units which are broken, do not ever fight well again. Leadership losses cannot be made up for. So,at that point in the game, not much to do except watch units die or run, and as pointed out, they cannot even run off the map like the real life units would.

So,, you want close games, set up the scoring like the old game, keep the game short enough that the battle does not become a route. Easy fix, but dont expect the designers to do that for you. it is not what many want.

[sburke]

The subject sounds interesting but frankly I haven't enough experience yet to contribute much. One thing I have noted is that in games where I felt the whole battle had been a close run thing, the VP points may not necessarily reflect that. One of the reasons why I avoid worrying too much about them and focus more on whether I think I achieved my overall objective with a reasonable casualty rate.

I am particularly interested in the suggestion from Normal dude of having an exit capability in order to create scenarios where a defender can be rewarded for inflicting casualties and preserving their force. That kind of creativity can result in much more realistic scenarios that could be more balanced as well.

+10 points for that suggestion Normal Dude, you are excused from KP duty for the rest of the afternoon.

[Sgt Joch]

In and of itself, I don't see CMBN or CMSF battles as more imbalanced than CMx1, at least not when you play them.

Tactically, they do play out differently in CMx2 and appear to favour the defense more.

The issue as others have pointed out probably has to do more with the way VPs are counted in CMx2.

[PanzerMiller]

I don't think disparate sample sizes is itself a problem. If there were 1000 CMBN games and 1,000,000 CMAK games, the comaprison would still be useful. The question is whether the "small size" in absolute terms of the CMBN data is "too small". A small size, in absolute terms, of the CMBN data set would certainly limits its usefulness. But there are 80 games in there: that's not ridiculously small. After 80 games, we can see that the "apparent trend" is for at best a flat distribution, maybe even favouring extreme results over even ones.

How many games would you want to see in the data-set before you started to thing that this result is statistically significant? I'm no mathemetician... but it seems to me that after 80 games, having 30+ of them with an 80% plus winning margin seems like a lot.

Is there any other relevant "characterisation" or statistic that would add to your understanding?

GaJ -- With all due respect, I think you're being a bit harsh on NormalDude here. I would agree with ND that you cannot make a truly meaningful comparison between 3,856 scenarios played in CMAK and only 80 scenarios played in CMBN -- especially when the individual actors that generated those two datasets have (no doubt) significantly different characteristics -- familiarity with the game engine being just one example.

And there is a BIG difference between 80 v. 3,856 datapoints and 1,000 v. 1,000,000 datapoints. A BIG difference. In this case, a dataset with 80 events is too small to draw meaningful conclusions about hypothesized flaws in game mechanics, scenario design, etc. Sample size IS an issue here.

Your later comparison of the first 80 scenarios from BO, AK and BN is starting to move down the right track, statistically. But the results from the full AK dataset indicate that the initial distribution is not fully representative of the "true" distribution and, therefore, difficult to interpret correctly on its own.

These are really interesting data and a good discussion. But the jury is still out.

From GAJ's original graphs I estimate that

The probability that the CMAK distribution is unbiased is <0.00001% !!

In the latter case I compared the observed distribution with a theoretical distribution having a mean result of 50% and the same width as the observed.

What are your assumptions concerning this "theoretical distribution"? The width is not the real issue...it's the variance...can you be a bit more specific?

[Calibration]

What are your assumptions concerning this "theoretical distribution"? The width is not the real issue...it's the variance...can you be a bit more specific?

Good question PM. By width I meant the full-width at half maximum which is of course just ~2.355 times the standard deviation for a normal distribution. My assumption is that the "theoretical" unbiased distribution would be normally distributed and have a FWHM close to the observed CMAK distribution as the latter is only mildly skewed. I generated a "theoretical" unbiased distribution with a mean of 50% and this FWHM. I then applied a Kolmogorov-Smirnov test to the cumulative distribution functions of both this "theoretical" unbiased distribution and the observed CMAK distribution to derive the quoted probability. Not a method a statistician would be entirely happy with no doubt, but the end result will qualitatively be correct however you do it.

[SlowMotion]

I could be wrong, but one feeling I have is that once you've lost some ground and your opponent knows where you are, it's hard to survive. There's so little real cover.

I think this is true in many QBs. Many maps have very little cover and when you have long playing time, the side with advantage can go and "clean the map". If there were more trees, infantry wouldn't be like sitting ducks that they often are now.

[GreenAsJade]

It may be instructive to separate QBs from scenarios. That would give you a better idea of scenario design vs. game mechanics.

Great idea.

I might do that if I have a chance.

Note that I for one never mentioned game mechanics. I don't think it's necessarily a game mechanics thing at all, except for the fact the buildings offer so little cover.

GaJ

[GreenAsJade]

Since both sides generally took losses and those losses formed a significant part of the total score, the results tended towards the center. With CMx2 scoring, there is often not a penalty for losses at all, or perhaps just a bonus for keeping friendly casualties low. Therefore the final score is more extreme for the same battle action as in CMx1.T

Another great point. There's a "binary" component to a CMBN result: either someone gets hundreds of points for a VL, or they don't!

OTOH, this factor was there with flags in CMx1, so I'm not sure: what makes it _more_ of a factor now? Is the relative weight of CMBN victory point things more?

GaJ

[GreenAsJade]

In this case, a dataset with 80 events is too small to draw meaningful conclusions about hypothesized flaws in game mechanics, scenario design, etc. Sample size IS an issue here.

Acknowledged ... and I didn't say sample size wasn't an issue. I do agree that the sample size is small. What I said was that 80 games is not _ridiculously_ small. It's enough, IMHO, to be able to say "gee, hmm... that's interesting!!".

I totally agree the jury is out. If the verdict was in, this would be a different discussion, eh?

GaJ

[sburke]

Good question PM. By width I meant the full-width at half maximum which is of course just ~2.355 times the standard deviation for a normal distribution. My assumption is that the "theoretical" unbiased distribution would be normally distributed and have a FWHM close to the observed CMAK distribution as the latter is only mildly skewed. I generated a "theoretical" unbiased distribution with a mean of 50% and this FWHM. I then applied a Kolmogorov-Smirnov test to the cumulative distribution functions of both this "theoretical" unbiased distribution and the observed CMAK distribution to derive the quoted probability. Not a method a statistician would be entirely happy with no doubt, but the end result will qualitatively be correct however you do it.

I think it should be a house rule that if you post something like this you need to send out a bottle of wine to every person who reads it or at least some excedrin. My idea of a Kolmogrov-Smirnov test is to see how many bottles of Smirnov some guy named Kolmogrov can drink while still standing upright.