strange tank spotting behaviour

[Narses]

Anyone playing and expecting, or even hoping to come out ahead, has a screw loose, I'll agree. I can see the attraction of taking the money you might've spent on show tickets/coke/hookers or the like and choosing to blow it on the gee gees or the tables instead: there's a thrill in thinking you might get lucky after all Personally, I'd rather drop a ton on a good restaurant; good food (even if it's a shortcut to the cardiac ward) is healthier than gambling debts.

Or dropping it at the Vegas gun range firing a MP40 or Thompson. Then eating is next !

[siffo998]

the idea is that each time you roll a dice, the chances that you roll 1 are the same. it doesn't matter if you have rolled 1 a hundred times in a row previously. the chances are the same for each roll.

spotting a tank (for example) is different. crews are trained specific scanning techniques for spotting targets. the chance that a crew spots an enemy tank in any given second increases by every second they do scanning. it's a bit like if you had four boxes and one of them contained a cake -- each time you open one box the probability that you find the cake from the next box increases.

what i mean regarding the game is that i suspect it works in some areas by the dice method, not the cake-in-a-box method. this is why there's a huge variation in results for example in your spotting test, as with bad luck the Sherman may never spot the Panther during the 30 minute battle (while it may spot it on the first second if you are lucky) -- each time the game "rolls the dice" to see if Sherman spots the Panther the chance of spotting remains the same.

i don't mean that there wouldn't be logic in the game or that it would all be just randomness. i just mean that there's so much randomness that you need to do those 2000 tests to have good statistics. and because of the same reason it's almost irrelevant when you play the game (you would need to play 2000 battles to see the difference, just like when running tests).

i don't mean that the game would be broken. i'm just pointing out that there's so much variation that it's practically impossible to draw conclusions even after running 50 tests.

I understand your point and iam familiar with probability calculation nonetheless thanks for your explanation. the problem i had with blackcats answer was to understand what he wanted to say to me regarding the results of my test ( a) your results of the first test are bad, your results of your second test are bad, c) you are an idiot that just cant run tests ;-) ).

I`ve tried to eliminate most of the variables that cause randomness (all crews have the same experience and the same motivation etc., they are standing on complete flat ground, facing each other with a cover arc in direction of the other tank). After all its a game and it should show some continuity in the results at least if you start 50 tests...! I for myself think that the results were different because the testing and counting method in the first 3 tests (which showed some proofable continuity by the way) were different than in my last test were i focused mainly on who spotted faster (not who shoots first).

After all do you really think that the design crew would run 2000 tests for each behaviour ingame to show if their input is similar to the real world ? i dont think so... (i want to see the eyes of the tester that runs 2000 test in a row ;-) )

[siffo998]

ok to make it quick i`ve rerun the last test with the same conditions but without the cover arcs for the tanks and it showed nearly the same results regarding spotting time (test runs total: 50):

Sherman spotted Panther faster:

23 times (46%)

Panther spotted Sherman faster:

25 times (50%)

Draw (both spotted at the same time or both didnt spot each other after one minute at all):

2 times: (4%)

----------------------------

Panther spotted Sherman not at all after one minute:

13 times

Sherman spotted Panther not at all after one minute:

12 times

the test results are in my opinion absolutely compareable to the results of the previous test. my first 3 test runs showed different results because of different testing conditions and different things i`Ve looked at.

i can conclude from my actual test and the last one that actually:

the panther has an slight advantage in first sighting the sherman like it should be.

but: (and this part is just from what i`ve seen and is not tested) the panther needs a slightly longer time from spotting to actually firing than the m4a3. (dieseltaylor has already assumed this)

afterall dieseltaylor was right with his initial concers that my inital 3 test had some test flaws in it...at least regarding the spotting time.

by the way i can conclude that...50 tests show reproduceable results and you do not need to run 2000 tests... ;-)

[undead reindeer cavalry]

the test results are in my opinion absolutely compareable to the results of the previous test.

the results seem very different to me. :confused: in previous test Panther had 100% higher chance to spot than Sherman. in this second test the chances are equal. if you run a third test, will Sherman have 100% higher chance to spot?

by the way i can conclude that...50 tests show reproduceable results and you do not need to run 2000 tests... ;-)

i guess you are talking about scores of spotting duels? the results of those duels are so random that 50:50 results are expected (it's just about flipping the coin with such small dataset). please notice that you had 53:47 result in the previous test where Panther had 100% higher chance to spot (so 100% difference in chance to spot lead to 3% difference in outcomes). then compare with the results from the second test. i find it hard to conclude anything other than the difference in the chance to spot has almost nothing to do with who wins most of 50 such duels.

so i guess what i am saying is that it might be more productive to ignore the outcomes of the duels and just look at the data you get for spotting from the duels (what is the chance for a tank to spot the opposing tank, per second).

[undead reindeer cavalry]

what comes to what i wrote about suspecting things about how the game engine handles "randomness" internally, let me use another example to make the point a bit clearer.

in the spotting duel test the Panther had 100% higher chance to spot, yet the 3% difference in the end results of the duels was of expected scale.

now another scenario. you have Iraqi T-72 that has 48% chance to hit US M1. you have US M1 that has 96% chance to hit the Iraqi T-72. now you make them duel it out, which one hits first, and you run the test 50 times. while we have the same 100% difference in chance to hit, do we expect 3% difference in the end results?

there's a huge fundamental difference between these two scenarios and i really wish is that CMBN engine design has methodology to deal with this and what we see in the game and trials like those of this thread are just statistical abnormalities.

[undead reindeer cavalry]

just in case someone doesn't get what i mean: if CMBN would check spotting ten times per second, then the M1 has 282 times greater advantage than Panther per each individual "dice roll" (96-48 vs 0.33-0.16, or 48 vs 0.17).

if M1 fires only once and Panther keeps spotting indefinitely, the advantage would be nullified in 28 seconds and in 56 seconds Panther would have twice the advantage. this is why casino wins in the end.

the game also wins if it, and the developers, has means to manage the issue. if it doesn't do it well, the end result is that different parts of the game have different levels of randomness. it would make some parts of battles extremely random while others barely random at all.

hopefully the game handles this well, but i fear the game doesn't have enough "dice rolls" for the small percentage stuff and thus we get some unintuitive randomness.

[LemuelG]

but: (and this part is just from what i`ve seen and is not tested) the panther needs a slightly longer time from spotting to actually firing than the m4a3. (dieseltaylor has already assumed this)

Well that sounds proper to me. Panther gunner has no periscope, whereas Sherman gunner does - Sherman gunner has much better awareness (no contest, US gunner does not need to use his sight to acquire the target, using his wide-FOV periscope while German gunner must pan around using his 2.5x with laughably narrow FOV) due to this and can lay gun on the target quicker.

The quality of the German optical sight may have been better, but that's not particularly helpful for target-acquisition unless the target happens to be directly in front of the turret.

[siffo998]

Well that sounds proper to me. Panther gunner has no periscope, whereas Sherman gunner does - Sherman gunner has much better awareness (no contest, US gunner does not need to use his sight to acquire the target, using his wide-FOV periscope while German gunner must pan around using his 2.5x with laughably narrow FOV) due to this and can lay gun on the target quicker.

The quality of the German optical sight may have been better, but that's not particularly helpful for target-acquisition unless the target happens to be directly in front of the turret.

actually iam not complaining about this situation... i just wanted to show the results...

but as far as i know the panther A also had a periscope... (only modell D had none)

[siffo998]

just in case someone doesn't get what i mean: if CMBN would check spotting ten times per second, then the M1 has 282 times greater advantage than Panther per each individual "dice roll" (96-48 vs 0.33-0.16, or 48 vs 0.17).

if M1 fires only once and Panther keeps spotting indefinitely, the advantage would be nullified in 28 seconds and in 56 seconds Panther would have twice the advantage. this is why casino wins in the end.

the game also wins if it, and the developers, has means to manage the issue. if it doesn't do it well, the end result is that different parts of the game have different levels of randomness. it would make some parts of battles extremely random while others barely random at all.

hopefully the game handles this well, but i fear the game doesn't have enough "dice rolls" for the small percentage stuff and thus we get some unintuitive randomness.

nice conclusion... even as iam intended german college my math knowledge (especially probability calaculation) was never the best... so it is good to have someone here that is such a math crack...

but could you please explain to me how you get those hits per second results out of my hitting time results (Sherman: 1.6% Panther: 3.3% posted on page 2 from you) thanks in advance.

[dieseltaylor]

One thing about the spotting should be the range if we are going to compare the comparative spotting abilities of the tanks. We could also get cute and see if actual tanks differ. And whether looking inot the sun is significant, and in low light do the superior optics make a difference.

However what I am finding interesting from a players point of view is the amount of variation in spotting times. There are two things here:

1. Confidence as a player in what one might expect in spotting terms.

2. Can it be right that in identical circumstances there is such huge variation

Whist siffo's figures show wh o spotted first graphing them shows there is something interesting. Panthers do not spot at all in 17 instances and Shermans 4 times. Off hand I cannot recall the rnage. Siffo am I right in what I observe?

[siffo998]

One thing about the spotting should be the range if we are going to compare the comparative spotting abilities of the tanks. We could also get cute and see if actual tanks differ. And whether looking inot the sun is significant, and in low light do the superior optics make a difference.

However what I am finding interesting from a players point of view is the amount of variation in spotting times. There are two things here:

1. Confidence as a player in what one might expect in spotting terms.

2. Can it be right that in identical circumstances there is such huge variation

Whist siffo's figures show wh o spotted first graphing them shows there is something interesting. Panthers do not spot at all in 17 instances and Shermans 4 times. Off hand I cannot recall the rnage. Siffo am I right in what I observe?

it is the opposite in my first test (panthers does not spot the sherman at all after one minute: 4 times, sherman does not spot the panther at all after one minute :17 times). in comparison in the second test: Panther spotted Sherman not at all after one minute: 13 times, Sherman spotted Panther not at all after one minute: 12 times)

[dieseltaylor]

Sorry I have not got all your stats. Are they available anywhere? I do have the test set-up and the first stats.

As to using other tanks it might establish if there is a generic spotting regardless of tank. And for that matter if you have less crew : )

If this seems bizarre to want to know it really is not if you want to have confidence in moving and fighting knowing how often and big serious differences are changes how many men you put to the task.

[siffo998]

no problem dieseltaylor... the second valid test is on page 3 nearly the middle (i havent provided exact seconds until spotting here sorry for that)

[undead reindeer cavalry]

nice conclusion... even as iam intended german college my math knowledge (especially probability calaculation) was never the best... so it is good to have someone here that is such a math crack...

but could you please explain to me how you get those hits per second results out of my hitting time results (Sherman: 1.6% Panther: 3.3% posted on page 2 from you) thanks in advance.

i by accident bounced on the Excel file i used in the calculations and realized i forgot to answer this question of yours (yes, i get the rhetorical nature of it).

i didn't make any real probability calculations, just very basic statistics.

basicly i took your spotting data from post number 6 and i calculated for each second (1-59) the number of tanks (separately for the two types) that had spotted during that second and the number of tanks that had still been spotting during that second.

so for example for Shermans on second 1 there are 50 tanks spotting and 2 of them spot. so 4% of the Shermans did spot the Panther during that second. or by stretching the meaning of the percentage, a single Sherman had 4% chance of spotting the Panther (with 6.79% margin of error lol). on the second number two there were 48 Shermans still spotting and 3 of them did spot (so 6.25% did spot, or a single Sherman had a 6.25% chance of spotting --- with a 8.39% margin of error).

the average for seconds 1-59 for Shermans was 1.78%. then i just calculated standard deviation for the whole bunch (2.41%) and confidence interval (0.61%). so again in other words, this crude method gives a single Sherman, during a single second, a 1.78% "chance of spotting" the Panther, while the margin of error is 0.31%. or in other words in similar 50 test runs in 95% of cases the average should be in between 1.17 and 2.40 percentages.

for Panthers it was 3.99% to spot Sherman (not 3.3 as i stated on the post -- i simply remembered the percentage wrong), with 0.75% margin of error. thus in 95% of cases of 50 test runs the percentage should be between 2.49 and 5.49.

so the Panther had 2.24 times greater chance for spotting per second. yet it was meaningless for the end results of your duels, due to the considerable variance, or randomness.

this is of course highly simplistic and contains fundamental errors (starting with the available data stopping at 59 seconds). so no, i didn't make any real probability calculations or such. just very simplistic basic statistics.

[drrowley]

another thing to consider is the random number generator. math generators have tendancy to generate streaks of high's and low's. this could explain the single panther vs sherman oddity.

[siffo998]

i by accident bounced on the Excel file i used in the calculations and realized i forgot to answer this question of yours (yes, i get the rhetorical nature of it).

i didn't make any real probability calculations, just very basic statistics.

basicly i took your spotting data from post number 6 and i calculated for each second (1-59) the number of tanks (separately for the two types) that had spotted during that second and the number of tanks that had still been spotting during that second.

so for example for Shermans on second 1 there are 50 tanks spotting and 2 of them spot. so 4% of the Shermans did spot the Panther during that second. or by stretching the meaning of the percentage, a single Sherman had 4% chance of spotting the Panther (with 6.79% margin of error lol). on the second number two there were 48 Shermans still spotting and 3 of them did spot (so 6.25% did spot, or a single Sherman had a 6.25% chance of spotting --- with a 8.39% margin of error).

the average for seconds 1-59 for Shermans was 1.78%. then i just calculated standard deviation for the whole bunch (2.41%) and confidence interval (0.61%). so again in other words, this crude method gives a single Sherman, during a single second, a 1.78% "chance of spotting" the Panther, while the margin of error is 0.31%. or in other words in similar 50 test runs in 95% of cases the average should be in between 1.17 and 2.40 percentages.

for Panthers it was 3.99% to spot Sherman (not 3.3 as i stated on the post -- i simply remembered the percentage wrong), with 0.75% margin of error. thus in 95% of cases of 50 test runs the percentage should be between 2.49 and 5.49.

so the Panther had 2.24 times greater chance for spotting per second. yet it was meaningless for the end results of your duels, due to the considerable variance, or randomness.

this is of course highly simplistic and contains fundamental errors (starting with the available data stopping at 59 seconds). so no, i didn't make any real probability calculations or such. just very simplistic basic statistics.

thanks for the explanation undead !

[undead reindeer cavalry]

another thing to consider is the random number generator. math generators have tendancy to generate streaks of high's and low's. this could explain the single panther vs sherman oddity.

there are some strange things with the results of the test, but it's almost impossible to speculate about them because we only have those 50 runs. it could be that the game uses a more advanced system than just one single percentage (by which i mean that the percentage may change over time). there also may be a system which makes only a part of the units to spot at any given moment in time, and the system may select the active units somewhat randomly. there may also be some bugs.

[dieseltaylor]

Thinking ahead to the RoW Tournament, whenever it occurs, the large variability in spotting times and results might provide significant swings in battles where there will be few tanks.

OK so it is realistic but in terms of comparing performance it does seem uncomfortably variable. My gut feeling from hundreds of CMAK battles is that the variability in spotting simply was either not there or not as great. To be fair I do not have huge figures for that as replaying battle moves is not something one generally does.

I can suggest the hit figures were more variable particularly in CMBB.

[undead reindeer cavalry]

OK so it is realistic

i think it's very unrealistic. there's nothing in the scenario that prevents the tanks from spotting each other on second 1. some spottings lagging till around second 10 could be tolerable, but not being able to spot in the whole minute is ridiculous.

there are numerous field trials about this very subject and none of them show anything like this. buttoned tanks (without using thermals or similar), do routinely spot enemy tanks in 20 seconds in complex real world terrain and at considerable ranges.

like i said the trial data has some peculiarities and it may be that spotting suffer from some "optimisations" made to conserve CPU cycles.

My gut feeling from hundreds of CMAK battles is that the variability in spotting simply was either not there or not as great.

i haven't played CMAK so much, but CMBB does not have the variability. i used to hate the instant spotting, especially spotting of non-moving tanks in good ambush positions.

[dieseltaylor]

URC - do you have any links to these studies?

However I do nead to get time to run the tests under 1.01 to see if it has been tweaked.

[stoex]

urc,

I totally agree with you, spotting under the circumstances in the tests should be near instantaneous 90% of the time. And it should never take more than 20 seconds max.

[undead reindeer cavalry]

URC - do you have any links to these studies?

unfortunately i don't seem to have any. it's been at least a couple of years since i looked into this. if i have bookmarks and documents, they are in some old harddrive in the closet.

but i did have two papers somewhat related to the subject. i was using them for estimating battlefield effects of illumination lamps in WW2. the main effect and purpose of the lamps is of course not that of aiding in target detection, but i was curious about the theoretical or practical effects on the ability of tanks to detect enemy tanks (yes, a most silly subject).

the papers are:

AN ANALYSIS OF TARGET ACQUISITION BEHAVIOR FOR OBSERVERS IN TANKS EQUIPPED WITH THERMAL OR OPTICAL SIGHTING SYSTEMS

http://www.dtic.mil/cgi-bin/GetTRDoc?Location=U2&doc=GetTRDoc.pdf&AD=ADA162208

i had been trying to use it in determining "standard" abilities of non-thermal equippent tanks (WW2 tanks) to detect enemy tanks at night. it seems to contain some basic probabilities and times for detecting tanks at various conditions.

and

The Effects of Observer Location and Viewing Method on Target Detection with the 18-Inch Tank-Mounted Searchlight

http://www.dtic.mil/cgi-bin/GetTRDoc?AD=AD445050&Location=U2&doc=GetTRDoc.pdf

i had been trying to use it to determine how much the lamps could help in detecting tanks at night.

However I do nead to get time to run the tests under 1.01 to see if it has been tweaked.

did you run any tests?