Various and sundry
Mithel wrote:
You mention that we don't want firm 1% loss rates, that in actual combat loss rates would be high in one battle and low in another. That is still achieved with GD = 999. With 1:1 odds there is one loss rate (essentially determined by the attack value and the ground def efficiency) while 12:1 odds will generate twelve times the casualties.
Oh, definitely! The distribution I'm referring to isn't that easy to fool or suppress, in any case -- once you factor in terrain, leadership, weather, and so on, the loss rates will still be spread all over the place. What I'm concerned with is the law of unintended consequences -- we don't necessarily get what we expect when we make big changes. To wit:
Mithel wrote:
Now a new issue has come to my attention. The combat efficiency (from terrain, weather and night) has a huge impact on this too. I hadn't realized previously that a 1% efficiency drops your GD from 999 to essentially 10 and the dogpile problem comes back! So I need to either raise GD even higher or engineer the system so that combat efficiency can NOT drop below say about 5%.
Right. This is part of why I hesitated to go with really high GD, because of the wild swings in overall effectiveness. I think what you may find next is that really high GD makes defense of good terrain in bad weather phenomenally strong, because the defender usually has a higher net effectiveness. One division on a snowy mountain at night with effectiveness of 60 % and GD = 999 can resist being "dogpiled" by as many as 600 enemy divisions who have been reduced to a minimum of 1 SA. Under your system of 98 % GD, it would suffer 12 x 0.15 = 1.8 % losses per hour, right? Being outnumbered 600 to 1 and still only losing twice the historical average seems a little odd to me. EDIT: oops, 1.8 % per HOUR, 1 % per day. Right. Still, outnumbered 600 to 1 and only losing 43 % per day seems doubtful to me.
What matters here isn't how big you make GD, but how big your typical SA is. If it's in the 10 to 20 range, then it's irrelevant whether effectiveness is 1 % or 5 %, the rounding ensures the same result. Plus you would have to re-engineer the individual weather, night, leadership, river, national characteristics, unit-type, and terrain penalties to be very, VERY weak to avoid hitting 1 % in some situations where they are combined.
I suspect huge GD values will turn out to be exploitable in some weird ways, e.g. a human player could have 2 or 3 divisions take turns holding a snowy mountain province against stupendous numbers of the enemy, rotating them out as they become worn down. But as I'm a big believer in full-scale testing, I'm not rejecting the idea, I'm just telling you what worries me about it.
Mithel wrote:
Far more important is to eliminate losing a province that is hundreds of kilometers across in a matter of hours (which is a matter of seconds while playing).
Hmmm. I certainly agree with the annoyance this represents in gaming terms. But the math is tricky, because the full length of the engagement isn't really a few hours -- it's all the march time PLUS a few hours. What this effectively represents is provinces that fall as fast as armies can march across them. This happened fairly often historically, e.g. Rommel going from Sedan to the Channel coast, or the first few weeks of Barbarossa, or the Japanese in Malaya, or Rommel advancing from El Agheila to the Egyptian frontier. To cope with this kind of problem in single-player, I play fairly slowly, and that does limit my ability to complete long campaigns. However, given that HOI separates moving and fighting, I don't see an alternative to this happening at least some of the time.
In my original study last year, about 35 % of HOI battles were over in less than a day. I think the system needs somewhat longer battles, but given the shape of the distribution curve, I don't see any way to reduce the proportion below maybe 10-15 %. I suspect you'd have to decrease loss rates to much less than 1/10th their present level to accomplish even that much. The problem, as I see it, is the huge number of combinations of leadership, weather, night, terrain, and so on, that lead to tremendous swings in effectiveness, and thus to swift collapses in certain situations. So I can readily see the appeal that Nexus2012 is referring to, but I suspect that the difference made by GD = 999 will turn out to incremental rather than revolutionary, and it will introduce its own new problems.
saintsup wrote:
Why do assume that the historical result should be the center of distribution for an accurate model of warfare ?
It's just one sample.
Good question. My answer comes in two parts. If this game was JUST a simulation of the conquest of Poland, I'm sure you'd agree that the historical result should be in the center of the distribution, right?
However, because we have other campaigns we also want to model using the same rules, we aren't going to insist that Poland must work out exactly as it did historically. So the justification for letting Poland run fast or slow surely has to be that the same rules produce reasonably accurate results for other campaigns, and that the odd result for Poland is "good enough" when considered alongside all the others.
However, when we look at the other campaigns, we see contradictory complaints. On the one hand, the Germans take forever to conquer France; on the other hand, they roll over the USSR by fall 1940. The Japanese crush China and annex it rather than getting stalemated, but then the British and Americans are paralyzed by the task of launching amphibious invasions.
And the problem in each case isn't necessarily the game mechanics, e.g. movement and combat, but lame strategic planning by the AI.
I contend that at this point we don't really KNOW whether other campaigns are really running fast or slow overall, or whether movement and combat mechanics are to blame, because nobody has systematically gathered a useful set of repeated runs to show what the average campaign lengths are when AI problems are not a factor. We have plenty of player complaints about specific aspects of play, but not a lot of full-scale testing.
That's why I created the "Beat the AI" AAR last fall, to get some actual data about a historical campaign and compare it to what we get in HOI. And it was clear in that case that although the AI wasn't as good as a human player, when the game ran AI versus AI, it came out close to history -- and when it ran AI versus human, the result was still almost always defeat for Poland in about the same length of time regardless of whether Poland or Germany was the human.
So in the one case where we have enough data to really know what is going on, standard HOI came fairly close to the right length of time and the right outcome. It is only one data point, but it's based on several hundred runs under varying conditions and it properly accounts for the AI problem, so I tend to put some weight on that one data point. I hope that makes sense.
