Question(s) re: Combat Bonus

[Redwolf]

Before the discussion gets more heated, let me say I posted before my first coffee, I didn't meant to be rude.

I will run some tests today or Thursday evening.

There is a dirty trick in CMBO which makes testing like this hard. People may think I got nuts altogether, but I am pretty sure that individual vehicles in one battle are preset with some random modifiers of how good they shoot, and they keep on doing this for the whole battle. I observed that quite carefully and am convinced that this is the case or else that I have an overdose of CMBO testing. It applies to some TacAI aspects as well, like the tendency of tanks to open main-gun fire on infantry on their own behalf.

So, this means if you compare 2 guns, one with and one without bonus, then this fixed "individual" factor will play in and you observe differences that are not caused by the the HQ.

[flamingknives]

Actually, I didn't say that ATGs got a bonus to hit, the tests I have run on this are inconclusive.

What I said was that DF arty (ATGs, Field guns etc) got tighter grouping when firing HE at ground targets. i.e. thespread of shell hits was smaller the higher the combat bonus was.

[Dschugaschwili]

Originally posted by redwolf:

Before the discussion gets more heated, let me say I posted before my first coffee, I didn't meant to be rude.

I will run some tests today or Thursday evening.

There is a dirty trick in CMBO which makes testing like this hard. People may think I got nuts altogether, but I am pretty sure that individual vehicles in one battle are preset with some random modifiers of how good they shoot, and they keep on doing this for the whole battle. I observed that quite carefully and am convinced that this is the case or else that I have an overdose of CMBO testing. It applies to some TacAI aspects as well, like the tendency of tanks to open main-gun fire on infantry on their own behalf.

So, this means if you compare 2 guns, one with and one without bonus, then this fixed "individual" factor will play in and you observe differences that are not caused by the the HQ.

If this observation is correct, there's still a way of testing this: Place one or more shooters and targets on the map, plus the HQ(s) you want to test. Now change the setup for the different tests: place the HQs to command all the shooters or place them in a remote corner of the map so they don't reach any of them. That way you have the same shooters in all tests, cancelling any possible variations within one unit category.

Dschugaschwili

[Michael Emrys]

Originally posted by Dschugaschwili:

If this observation is correct, there's still a way of testing this: Place one or more shooters and targets on the map, plus the HQ(s) you want to test. Now change the setup for the different tests: place the HQs to command all the shooters or place them in a remote corner of the map so they don't reach any of them. That way you have the same shooters in all tests, cancelling any possible variations within one unit category.

That might work, but I have my doubts. If the extra hidden bonus that Redwolf is talking about exists, it might only be computed either at the end of the set up phase or first orders phase. In that case, it wouldn't appear in all your tests.

A way to get around this is to set up a map with multiple guns and HQs, the latter all having the same bonuses, and run them through the test multiple times. That way, all variations due to luck, etc. will average out and give you a more reliable picture of whatever is going on anyway.

Michael

[ August 29, 2002, 04:41 AM: Message edited by: Michael emrys ]

[Redwolf]

Originally posted by Dschugaschwili:

If this observation is correct, there's still a way of testing this: Place one or more shooters and targets on the map, plus the HQ(s) you want to test. Now change the setup for the different tests: place the HQs to command all the shooters or place them in a remote corner of the map so they don't reach any of them. That way you have the same shooters in all tests, cancelling any possible variations within one unit category.

If I remember correcly, that is not neccessary. From my observation this "individuality" is per-battle, not per-scenario. That means if you start the same scenario a few times from scratch, you will have different results. If you restart several times from the same savegame, the "individuality" stays the same.

I would be grateful if somebody could test this and tell me whether I am paranoid or this is for real.

I found a good test to be a StuH versus a Wespe. Somebody claimed the Wespe was more accurate and I tested and sure enough it was. But the next time there was a diference in precision again, but this time the StuH was better. A quick test is also to drive a Pz IV along some enemy infantry in fast move. How likely is it to turn its turret or to staty focused. A third test is a StuG against infantry, how likely will it use its main gun?

As I said, I do not remember precisely whether this is per-battle or pe-scenario, so very valid testing would have to edit the scenario. But I think it was per-battle.

[Michael Emrys]

Y'know, Redwolf, what you are seeing could also be ascribed to statistical variance, of which there is a lot in CM.

To give a simple example of what I mean, take a coin toss. If you toss an honest coin 1,000 times, you can reasonably expect it to come up heads pretty close to 500 times and the same for tails. But if out of that series of tosses you were to randomly select ten consecutive tosses, you would probably find that either heads or tails predominated. In fact, it wouldn't even be all that surprising to get ten straight heads or tails. And that is even though your are using the same coin through all 1,000 tosses. So the physical behavior ot the coin hasn't been altered between tosses, you are just seeing randomness in action.

So, what this has to do with gun accuracy in CM is that in comparing, say, two guns is that one may look better in one set of trials and the other in another set, all without the program giving any bonus to either one. It's just randomness.

Michael

[ August 29, 2002, 11:19 PM: Message edited by: Michael emrys ]

[tecumseh]

Hey redwolf - I just ran some StuH vs Wespe tests. 10 regulars each on the map, area-targeting a square of rough exactly 300m away. I ran the map 6 times for 5 turns, so I got a good feel.

You may be right! I saved the game after I'd setup the fire orders, and wooooooooo there was this one Wespe that was hitting dead center on all the turns, AND continued to after I'd restored to the saved state to run the test again and again (he was third in from the east).

There was also a StuH near the middle that was abnormally good, and a wespe near the map edge that was abnormally bad. These were also consistent after restoring and running the test again and again.

However overall there was no conclusive evidence that Wespes are more accurate than StuHs. However i did see evidence that individual vehicles were more accurate than others, and this effect survived over save-games.

Of course, this was just a quick test. But it was very strange.

[edit]

I ran another test at 560m. Again there seemed to be "talented" units and "challenged" units. There also was a StuH near the map edge that would fire less than the others, but this may have been because he was near the edge (?!?). Anyway, these characteristics stayed through all 10 restored games.

I am thinking Steve will post soon and say "there is no such effect!!" and I'll look like a dip****, so I'll leave it there.

[ August 30, 2002, 07:54 AM: Message edited by: tecumseh ]

[xerxes]

An individual "skill" modifier would certainly be interesting if it exists. It would be quite hard to tell from statistical randomness thought Tec's findings are intriguing. It would explain some very odd things that I've encountered.

[Silvio Manuel]

If it's true, I think I like it...it allows for the variance between crews that have the same exp.level for the same AFV.

[Redwolf]

Originally posted by Michael emrys:

So, what this has to do with gun accuracy in CM is that in comparing, say, two guns is that one may look better in one set of trials and the other in another set, all without the program giving any bonus to either one. It's just randomness.

All you say is correct. The question here is when you you say "the probably that is happens accidentially is so low that there must be something in the code"?

I worked for 12 years as a programmer in market reasearch and I think I am very hard to bluff with statistics. I claim I have a well-tuned sense for effects that are real and those which are just conicidence.

When I discovered what I felt is "individuality" in CM vehicles, I most cerntainly defaulted to think that this is just an interesting turn in the random number generator or that I was falling victim to some pshychological trap. But I re-ran enough tests to satisfy my own first threshold of "this is real", and as I said I would guess I have an exceptionally high threshold (certainly 10 times higher than anyone else in market research )

And I think coding in such a feature makes sense. For one, it is realistic. Then, think about how hard it is to make a programmer's decision about engaging infantry with a tank, either the decision of diring the main gun or the decision to turn the turret. So, whatever threshold and decision mechanisn the programmer writes, it might be wrong and turn out to be a gameplay problem. But if you give each vehicle an individuality setting so that they all behave with some variance around a middle value, then you will equal out a lot of mistuning, resulting in a game which is more robust against slight misjudges in coding, so that the programemr can concentrate on more important issues.

OK, so I said "my first threshold". I would like to do more, but I just don't have the time to do any real CM testing this summer. That is why I asked other people to look out for what I have seen.

[Redwolf]

Well, I ran tests with 10 105mm woitzer, 5 under 2+ HQ command and 5 under no command.

The ones with HQ do have a tigther area fire pattern. However, the effect is minimal, at a range of 3600m the spread area is only a few meters tigther. The difference is clearly visible, but given the blast of the gun I would say it has no combat effect, much less at any realistic range.

[Redwolf]

I ran tests with zooks at 205-207m.

I cannot find test results that would indicate the HQ-controlled Schrecks have a higher hit probablity.

[Silvio Manuel]

Originally posted by redwolf:

I ran tests with zooks at 205-207m.

I cannot find test results that would indicate the HQ-controlled Schrecks have a higher hit probablity.

Silly question, but was that vs. Armored targets?

[Michael Emrys]

Originally posted by tecumseh:

You may be right! I saved the game after I'd setup the fire orders, and wooooooooo there was this one Wespe that was hitting dead center on all the turns, AND continued to after I'd restored to the saved state to run the test again and again (he was third in from the east).

There was also a StuH near the middle that was abnormally good, and a wespe near the map edge that was abnormally bad. These were also consistent after restoring and running the test again and again.

However overall there was no conclusive evidence that Wespes are more accurate than StuHs. However i did see evidence that individual vehicles were more accurate than others, and this effect survived over save-games.

These are interesting and suggestive data. Thanks for running and reporting the tests, tecumseh.

Michael

[Redwolf]

Originally posted by Silvio Manuel:

Silly question, but was that vs. Armored targets?

Yes, the Schrecks were shooting at immobile Sextons.