Patch?

[Bayes]

Hi Strela,

Enjoying the game a lot so far, and fully understand the importance of keeping formula complexity and the number of parameters low (to help parameter calibration and avoid the so-called curse of dimensionality, etc.). The new design is in this respect much more concise compared to the older John Tiller Campaign Series (JTCS), with JTCS for instance having range effects specified uniquely for every weapon type for every range.

However, I am a bit puzzled about the fact that Direct Fire Range Effect in Panzer Battles affects all weapon types "equally". Should not weapons supporting long range combat lose their effect more gradually as range increases, relative to weapons with more limited range? Or maybe weapon type is more or less insignificant when compared to other factors such as limited visibility, etc., when modeling range effects?

I would for instance expect that a "8.8cm Flak 18" was more effective than a "7.5cm PAK 40" at range 10 since this is the maximum range of "7.5cm PAK 40", while a "8.8cm Flak 18" can fire up to range 16. Yet, right now, both weapons have the same effect at range 10. In the same manner, I would also expect that the kill ratio for Tiger I vs T-34 increased in favor of Tiger I as range increases from 1 to 6, and not be constant, like it is now.

I am thus currently in favor of tying Range Effect directly to weapon range (M in the formula below). For instance, what about reducing the effect of a weapon gradually to say 20% at maximum range. That is, instead of using the following formula to calculate Range Effect (with R = 1.5 and N referring to fire range):

1 + (N-1) * (R - 1)

why not use:

1 + 4 * (N - 1) / (M - 1)

which introduces the range M of a weapon.

Then "8.8cm Flak 18" would gain an increasing advantage over "7.5cm PAK 40" as range increases. Both starting with Fire Value 32, at Range 10 they would end up at:

9.4 (29% of 32) and 6.4 (20% of 32), respectively.

Still a somewhat simple formula, but with more emphasis on the deadliness of long range weapons at long range. "8.8cm Flak 18" would for instance have 20% effect at range 16 instead of just 12% as it has now.

Kill ratio for Tiger I (Quality A) vs T-34 (Quality C) would become:

Range 1 - 4.0:1
Range 2 - 5.7:1
Range 3 - 6.8:1
Range 4 - 7.5:1
Range 5 - 8.1:1
Range 6 - 8.5:1

In other words, a T-34 should try to close in to Range 1 as quickly as possible and not stay at range 6...

What do you think? Good idea or not? :-)

Bayes

I know, I was just wondering if maybe I've been seeing mostly uncommon results, which could mean the armor vs. armor might be working as intended.

Changing one thing without breaking something else can be very difficult, so I appreciate the team taking the time to get things to work the way they want after the changes.

And there is the rub - we are worried about the various consequences. John has built a modifier in for hard attack strengths that can be a variable so we can test increasing & reducing strengths without changing all the units values. We are also experimenting with various range attenuation effects. There has also been a spirited discussion about a 'reverse' density modifier for smaller units so that they are harder to hit.

As you can imagine every single one has pro's and con's and ultimately though they all sound impressive may have little impact on game play. For example in a PBEM game I am a member of an 88mm AA gun took out a T-34 HQ at 3,500 meters (14 hexes) and then on its next defensive fire shot disrupted a whole T-34 company at 3,250 meters. This was using the original code!!!

Some features to help fire values have been included in the code. For example the fire uphill negative modifier (simulating shooting at hull down vehicles) know has an inverse where a unit firing downhill get a positive modifier to represent hitting more vulnerable aspects of a vehicle. This helps to make hills even more important.

Personally, what I am seeing is that the game system is providing losses close to the historical without additional tweaks. That said, with all the additional multiplayer testing being done we are watching the statistics closely and trialling a few different changes to understand the impact....

David

[Strela]

Hi Bayes,

This is one of the areas we had a debate on when the question of armor results etc were mooted. Jeff Connor suggested a range based approach not dissimilar to yours. Of interest this is similar to the old board game Panzer Blitz.

We went back and looked at the numbers to decide what made sense. Below is a table that we used to compare possible solutions. The three lines for each unit is as follows; Current equation - range attenuation value 1.5, Current equation - range attenuation 1.2, Range based calculation.



Some examples from the table is that though the effects are applied consistently under the existing formulas there is clear differentiation. Using the current values, a Tiger at 7 hexes is as effective as a T-34 at 3.

Also of interest the change in the range attenuation value to 1.2 would essentially makes six hexes as effective as three hexes in the prior calculation. An example of the effect of a 'small' change.

Your option is a further approach.

That said, there is a key area that needs to be considered and that is the actual hard values. There is consideration for more than just the 'power of the shell'. Things such as optics etc are included when the value is determined. For example the difference between the value of 32 for a Tiger vs 17 for a T-34 takes in a lot of other considerations. Another example is that the Panthers 75mm is actually higher than the Tiger at 36. Why, because it was a custom built tank gun not a purloined AA gun.

So there are arguments to go multiple ways and with so many variables we are cautious not to throw too many more into the mix. We haven't released the patch yet so you can be confident that we're trying a range of different things and I'll mention your concept to John.

Thank for the input,

David

[Richie61]

Different game system, but goes into Tiger I vs T-34's in board game format.

http://battlefieldswarriors.blogspot.co....-1942.html

[Bayes]

Thanks for the thorough feedback and taking my input into consideration, Strela!

The problem I see with the existing range formula

1 + (N-1) * (R - 1)

is that when you change R from e.g. 1.5 to 1.2 you affect all weapons alike. E.g., if Tiger I becomes twice as effective at range 6, T-34 becomes twice as effective as well. As a result, changing R won´t change kill ratio because the range effect change is cancelled out when you calculate the ratio. You can only control the amount of kills overall by changing R. What I would like to do is to reduce the effectiveness of T-34 at range 6 and at the same time increase the effectiveness of Tiger I. The formula I propose achieves this effect.

Here is another formula that almost exactly replicates the Range Tables from JTCS, with similar properties (makes long rage weapons more effective):



Range Effect = 1/(1 - (3/4) * ((N-1)/(M-1)))

N - Range
M - Max range of weapon


Maybe using these well-tested values from JTCS could be an interesting option too.

BTW. I guess I should clarify why I bring up the need for long range weapons to be more effective at long range.

Here is a scenario where 20 T-34 attack 24 8.8cm Flak from a range of about 2.5 km. Because of the quickly degrading effect of ranged weapons in the current system, it turns out that most of the combat in this scenario take place at Range 1, after the T-34 have moved a full turn or so. I would have expected that the majority of T-34 would be lost at longer range, as they approach. Maybe also increasing the chance of Opportunity Fire would help here (I have spread out the 8.8cm Flak to maximize opportunity fire occurrences):



----
Also, thanks for the interesting link with the analysis of combat between T-34 and Tiger I, Richie61.

For Panzer Battles, using the current system, I guess a corresponding analysis would be:

Range 7+; Tiger I superior (T-34 cannot harm Tiger I).
Range 1-6; constant kill ratio of 4:1 in favor of Tiger, with more random variation at closer range.
Assault by T-34; kill ratio of 4:1 in favor of Tiger.

Bayes

[Riley D. Smith]

Very good analysis from all. I learn quite a bit just reading what you folks discussed here. Thanks for the depth.

RDS

[ComradeP]

Bayes and Strela: the new proposals still don't take penetration data into account, or how they could be amplified through the quality dynamic.

JTCS values would also allow a Panzer IV G/H to destroy a T-34 at 1.500 meters, which is not realistic if you define the defensive value as mostly being based on frontal armour and overall protection.

Bayes: in your first example, unless I'm mistaken you started substracting the modifier at range 1 instead of range 2 and the modifier I get for the PaK 40 is 1.22 whilst yours seems to be a different one.

1+(4*(1.5-1)/(10-1))->1+(4*0.5/9)->1+0.222=1.222.

[Richie61]

Can't speak for the game and how it does the factoring/ calculations, but the L/43 had to be introduced specifically because of the T-34. It had a 2,400 m "extreme range". The effective range of the L/43 was more like 1,000 to 1,500 m vs a T-34 on most days.

[Liquid_Sky]

Shouldn't time under fire somehow factor into this? Those flak guns are going to get an awful lot of shots off in the time it takes for those tanks to drive 2.5km

Just because the game says you are firing 3 times doesn't mean 3 shots.

[Richie61]

Shouldn't time under fire somehow factor into this? Those flak guns are going to get an awful lot of shots off in the time it takes for those tanks to drive 2.5km

Just because the game says you are firing 3 times doesn't mean 3 shots.

About 1 round every 3 seconds

[Bayes]

ComradeP,

the Range Effect formula in the first example uses fire range N and max weapon range M to calculate Range Effect:

1 + 4 * (N - 1) / (M - 1)

For fire ranges N = 1, …, 16, the Range Effect table becomes as follows:



Here are some more ideas that might address penetration and quality amplification, which you bring up in your post.

Right now, Armor Effectiveness is based on RAW Hard Attack Values. Not before after Armor Effectiveness has been resolved are Range Effect and Quality modifiers introduced. So, for Tiger firing at T-34 we get the factor 1.33, which then is modified by the Range Effect and Quality modifiers. Similarly, for T-34 we get the factor 0.5, again to be further modified by the Range Effect and Quality modifiers. This produces the following weapon performance stats currently:



If we introduce the Range Effect formula 1 + 4 * (N - 1) / (M - 1) we get the following weapon performance stats:



To get the additional effect you request, I believe maybe one idea would be to apply Range Effect and Quality modifiers directly on Raw Hard Attack Value first. Then, one applies the Armor Effectiveness rule on the range and quality modified Hard Attack Values. The resulting weapon performance stats would be as follows:



using Range Effect 1 + 4 * (N - 1) / (M - 1). Notice how expected kills for Tiger stays rather stable for a while, then quickly drops due to the Armor Effectiveness rule.

I guess the Tiger becomes too superior in the latter table, but this may be fixed by recalibrating the available game parameters. I have no idea how the above idea would work overall, but maybe something to investigate (if not already done).

Bayes

Bayes and Strela: the new proposals still don't take penetration data into account, or how they could be amplified through the quality dynamic.

JTCS values would also allow a Panzer IV G/H to destroy a T-34 at 1.500 meters, which is not realistic if you define the defensive value as mostly being based on frontal armour and overall protection.

Bayes: in your first example, unless I'm mistaken you started substracting the modifier at range 1 instead of range 2 and the modifier I get for the PaK 40 is 1.22 whilst yours seems to be a different one.

1+(4*(1.5-1)/(10-1))->1+(4*0.5/9)->1+0.222=1.222.