Frustration with CMCW - Russian side

[The_Capt]

17 minutes ago, Artkin said:

You didn't post anything contrary to my point of view originally. You wrote something completely irrelevant to the point of someone, and I called you out on it.

Completely irrelevant?  In describing the real world conditions of trying to spot in combat?  I will admit to the buttoned/unbuttoned confusion but you keep slipping from one point to the next it is impossible to keep up what is relevant.  Let’s see, we went from T72s, to T34s and T72s, to buttoned vs unbuttoned, to T64s and now 5 min nil spot outliers.  You have done “thousands of test” yet have not posted a link to any of them and then have been insulting and downright prickish throughout the discussion…all we know from your point of view is “something is wrong”.  Oh and then you had the audacity to ask for a “bone” to do any real work on this.

Well that is very helpful, we will get the lads working on that right away. Why don’t you go play another CM title for awhile then until we get it all fixed up for you.

[Grey_Fox]

25 minutes ago, Artkin said:

You're the first person I've ever heard this from. In my experience they're utterly useless at all times in CMCW

Were they opened-up?

[Pelican Pal]

4 hours ago, The_Capt said:

The other area of improvement is the effect of artillery on armoured vehicles.  The current CM engine is not reflecting realities we are seeing in the war in Ukraine, I do believe a revisit is in order on just how vulnerable tanks are to heavy indirect fires.  This will have a big impact on CW as the Soviets were an artillery heavy force.

IMO this is the key problem with CM not modelling artillery fragments against tanks. It results in a feast or famine situation where you either knock tanks out of the fight with direct hits or do nothing with near misses. Artillery gives you no degradation over time and a CM abstracts a lot of system knock outs into chip damage so not having any chip damage from arty really hurts anyone who goes heavy into it.

 

[Artkin]

31 minutes ago, The_Capt said:

Completely irrelevant?  In describing the real world conditions of trying to spot in combat?  I will admit to the buttoned/unbuttoned confusion

That's what I meant. There was no need for it to go this far. I've been ready to move on.

 

31 minutes ago, The_Capt said:

 You have done “thousands of test”

Where did I say this?

32 minutes ago, The_Capt said:

insulting and downright prickish throughout the discussion…

If someone's going to be a smartass for no good reason, or not even read what was posted, then sure. I'm ready to admit when I'm wrong.

 

33 minutes ago, The_Capt said:

Oh and then you had the audacity to ask for a “bone” to do any real work on this.

It wasn't entirely serious. I'm not asking for a bone. I wasn't even asking to be on the CW beta team, I figured I was already a no-go person for my no-bull**** attitude toward issues within this engine. I'm not doing hours worth of testing or buying Excel to prove this point. It's very obvious without ever having to do testing. Testing just confirms the suspicions. It's like driving your car every day and then noticing when it takes a little more force to turn the wheel. Your tires are low on air. These sort of things you just know after driving your car so often. Humans are THAT good at things.

36 minutes ago, The_Capt said:

Well that is very helpful, we will get the lads working on that right away. Why don’t you go play another CM title for awhile then until we get it all fixed up for you.

You have proof above in this thread. All you have to do is acknowledge it maturely and dissect what exactly is going wrong. If the soviet crews are staring at birds for five minutes, why aren't the american crews doing it? There's no chance the difference between optics causes a 300% in maximum spotting times. I can't see ergonomics doing it too at these ranges.

If we were talking about 100m, I would believe it. That is a danger close range. If the Soviet optics don't have variable power settings (I doubt they do) it would be really hard to spot something that close, which is a distance inbetween your optimum optics range and the glass vision blocks. It would be pretty hard to spot something in that iffy middle ground. 800m is a totally different story. The gun should be ranged for 800-1200m, this is (I'm assuming here) the typical engagement range Soviet engineers designed their tank for.

38 minutes ago, Grey_Fox said:

Were they opened-up?

No, because the distance to target was around 2000m, it would probably be better to stay turned in and use the more powerful Shturm optics. They're just blind as hell. I noticed it originally in CMBS and the problem remained in CMCW.

[Vanir Ausf B]

6 hours ago, Artkin said:

I'm specifically talking about the 300% maximum difference.

I have explained why that is a contrived number. I'm done talking about it.

6 hours ago, Artkin said:

You're the first person I've ever heard this from. In my experience they're utterly useless at all times in CMCW and CMBS. Once I had at least a full US battalion of vehicles in a massive blob approaching a shturm platoon. Both sides had clear LOS to each other but my shturm platoon was hidden in woods with grass tiles underneath. The shturms were completely incapable of spotting the massive blob of vehicles, but instead were picked off one by one. I don't usually play with thermals so I don't think there were any TTS on the US side. I think shturms are one of the weakest spotters in all of CM.

I understand the games are running the exact same engine, but there appears to be a problem with the CW soviet tanks specifically. It's come up a bunch of times, shturm too iirc.

Sturm-S vs M60 RISE+ @800 meters. Outliers not excluded.

• Median
  • M60: 29 seconds
  • Sturm: 17.5 seconds
• Average
  • M60: 29.8 seconds
  • Sturm: 25.5 seconds

Since you attach so much meaning to the single largest time, for the Sturm it was 88 seconds.

That's a 49.4% advantage in median spot time for the Sturm-S. WTF @The_Capt, how do you sleep at night?  

Edited May 15, 2023 by Vanir Ausf B

[The_Capt]

1 hour ago, Vanir Ausf B said:

I have explained why that is a contrived number. I'm done talking about it.

Sturm-S vs M60 RISE+ @800 meters. Outliers not excluded.

• Median
  • M60: 29 seconds
  • Sturm: 17.5 seconds
• Average
  • M60: 29.8 seconds
  • Sturm: 25.5 seconds

Since you attach so much meaning to the single largest time, for the Sturm it was 88 seconds.

That's a 49.4% advantage in median spot time for the Sturm-S. WTF @The_Capt, how do you sleep at night?

Thanks for running these.  Well I think that pretty much puts to bed this round of “CM Spotting is broken!”.  As a game lead/designer I make it a policy never to ignore a player but there is one that is really testing me on this thread.

As to how do I sleep? - on piles of dirty cash and hookers, just like Steve and Charles (but smaller piles and uglier hookers).

[Artkin]

Not for nothing I didnt say the Shturm was at 800m. I will test longer ranges tomorrow. I wouldnt be engaging targets with the Shturm at 800m since it has a 400m minimum range iirc. 800m is a desperate suicidal distance for a shturm to fight at. 

And no, that doesnt put this to rest. I understand what the outlier means and I'm not interested in the average. The outliers are the problem. It doesnt affect US tanks as it does with Soviets. You have two pairs of optics in T-72. Theres no reason we should be seeing a dramatic difference in spotting times at 800m from the frontal arc between these mbt's on a flat clean range. Almost 5 minutes to spot a tank sized target is insane. Especially when a M60 is a LOT taller than a T-72. 

Good job Capt blowing over my last post. Where did I say I did "thousands of tests" again? Are you smoking it too? 

[IdontknowhowtodoX]

On 5/14/2023 at 5:56 AM, Vanir Ausf B said:

Perhaps there should be an option to disable the RNG and make all spotting checks 100% successful, LOS permitting. I would not use it but I suspect there is a vocal minority of users who would.

No dude don't do that. We already have Wargame Red dragon. Don't need another one. RNG spotting is what makes CM different. Even though it does frustrate players at times.

[IdontknowhowtodoX]

5 hours ago, Vanir Ausf B said:

I have explained why that is a contrived number. I'm done talking about it.

Sturm-S vs M60 RISE+ @800 meters. Outliers not excluded.

• Median
  • M60: 29 seconds
  • Sturm: 17.5 seconds
• Average
  • M60: 29.8 seconds
  • Sturm: 25.5 seconds

Since you attach so much meaning to the single largest time, for the Sturm it was 88 seconds.

That's a 49.4% advantage in median spot time for the Sturm-S. WTF @The_Capt, how do you sleep at night?  

Crazy. I thought Soviet's lovely tiny ATV were completely blind. This is truly surprising. 

[Butschi]

Let's calm down a little, shall we?

This unpleasant "discussion" made me dig out a test I did some time ago. I've been doing data analysis for basically all my professional life, so I always suffer a bit when reading these threads. So, as we've been discussing on page one or so, anecdotal evidence ("here look at this battle, spotting is broken!!!!") is meaningless. But also doing "experiments" is only as good as the experiment itself plus the evaluation afterwards. First of all, if you do experiments, control the variables! Eliminate everything you are not directly interested in. If you want to look at spotting, do it on a flat surface and make both opponents hold fire. Because, as I often see, if you measure time until first shot, or kill, you are skewing the spotting process. Next, don't look at averages or medians alone. Look at distributions. And don't eliminate outliers. My text book about statistical data analysis said that, eliminating outliers, although done often, should really only be done if you know what you are doing, e.g. when you know that your outlier is actually some measurement error and not some rare event.

So, here's what I did:

I put an M60 TTS and a bog standard T72 (the exact models aren't relevant for the method) on a flat map with paved ground, roughly 2 km apart. I set both vehicles to hold fire. I then measured the time it took for each tank to first get a partial contact and then a full contact.

Here is the raw data:

t72 = [84, 17, 78, 4, 174, 65, 77, 321, 289, 444, 31, 3, 290, 2, 40, 120, 40, 159, 57, 69, 15, 54, 80, 95, 19, 58,
       23, 672, 154, 154, 17, 14, 342, 12, 386, 43, 84, 12, 378, 123, 30, 44, 240, 311, 110, 2, 68, 181, 137]

t72_id = [7, 21, 7,28, 7, 35, 14, 0, 56, 0, 35, 0, 7, 70, 42, 7, 35, 35, 0, 49, 0, 7, 0, 0, 0, 63, 28, 7, 14, 0,
          7, 7, 7, 49, 7, 21, 14, 21, 14, 0, 7, 14, 28, 0, 14, 7, 28, 56, 77]

m60 = [89, 68, 41, 38, 71, 10, 73, 0, 20, 9, 4, 55, 91, 34, 31, 8, 14, 116, 64, 4, 18, 63, 116, 38, 3, 18, 71, 132,
       39, 73, 43, 73, 116, 210, 207, 36, 180, 27, 88, 48, 102, 3, 52, 77, 176, 22, 18, 80, 24]

m60_id = [7, 7, 7, 0, 14, 14, 0, 14, 14, 7, 14, 0, 7, 14, 35, 14, 7, 0, 7, 14, 7, 0, 14, 7, 7, 7, 14, 35, 7, 0, 7, 0,
          0, 7, 0, 7, 14, 14, 7, 7, 7, 0, 7, 0, 0, 7, 21, 7, 7]

Times are in seconds, the entry t72 is time until partial contact for the T72 trying to spot the M60, the one with  "_id" is the difference between partial and full contact. The same for m60. And in order to get the distributions I made histograms with 30s bins for plots 1 and 3 and 7s for plot 2.

So, what do we see here? Well, first of all, I should have taken at least ten times the data or make make larger bins. I didn't have the patience for the former and doing the latter would mean that we don't see much of a distribution. Anyway, from the raw data we see: time until partial contact can be any number, time to ID (which is what I call time to go from partial to full contact) is always a multiple of 7.

The histograms tell us the following: Although it is not possible to get the exact distribution, this is definitely not something symmetrical where average or media are easy to interpret. If you look at the bins with the highest counts, those are at low times. That means, players will usually see that their T72 or M60 are often quick to spot their target. Nothing to complain about or make a forum "rant". But for both (!) tanks it is quite possible that it takes several minutes - more likely for the T72 but also the M60 TTS had an event with over 3.5 minutes. The time until partial contact is consistent (no more, no less) with an exponential decay which you would expect when counting the number of dice rolls necessary to roll a specific number - only that the M60 TTS rolls with a D6, while the T72 rolls a D20, so to speak.

Now, is spotting "broken" or not? For me, this is really not a meaningful thing to say. Because by "broken" people usually mean "takes too long" without saying what "too long" is and why. With the above distributions, it is possible to never spot the target. Right? Wrong? Broken? This is a game/simulation and as any such thing, at least if you want it to run in reasonable time on a consumer machine, it is simpler than real life and abstracted. A meaningful discussion would be "Is this spotting model adequate? Could CM do better by applying model XYZ, instead?". I'm not sure if the long tails (what some call "outliers") are working as intended (although I find @The_Capt analogy with the barrel quite convincing) or if it is a model that is just designed to get the "center", the common situations, right, accepting that every now and then it produces something odd. But getting the tails of an exponentially decaying distribution right is brutally difficult - in fact, come to think of it, my whole PhD thesis was about modelling the tails of a similar distribution correctly.

 

[Vanir Ausf B]

2 hours ago, Butschi said:

So, what do we see here? Well, first of all, I should have taken at least ten times the data or make make larger bins. I didn't have the patience for the former and doing the latter would mean that we don't see much of a distribution.

Back when I was doing this for beta testing I would not submit a result to BFC with less than n=300, but I don't have the patience for it anymore. There is a guy who wrote a program in Python to automate it and would do n=1000 but that's not a skill set I possess.

2 hours ago, Butschi said:

 So, as we've been discussing on page one or so, anecdotal evidence ("here look at this battle, spotting is broken!!!!") is meaningless. But also doing "experiments" is only as good as the experiment itself plus the evaluation afterwards. First of all, if you do experiments, control the variables! Eliminate everything you are not directly interested in. If you want to look at spotting, do it on a flat surface and make both opponents hold fire. Because, as I often see, if you measure time until first shot, or kill, you are skewing the spotting process.

Exactly right. I also make sure all spotters are looking at the same thing, i.e. rather than timing the test groups spotting each other I time them spotting an identical third group (I use T-72As as my "target" group in all tests). I also eliminate C2 information sharing by making every unit of the test group in a different battalion and spacing them by more than 32 meters from each other.

2 hours ago, Butschi said:

Anyway, from the raw data we see: time until partial contact can be any number, time to ID (which is what I call time to go from partial to full contact) is always a multiple of 7.

I've never noticed that before, but it would make sense given CM spotting cycles are in 7 second intervals, most of the time.

2 hours ago, Butschi said:

 I'm not sure if the long tails (what some call "outliers") are working as intended (although I find @The_Capt analogy with the barrel quite convincing) or if it is a model that is just designed to get the "center", the common situations, right, accepting that every now and then it produces something odd. But getting the tails of an exponentially decaying distribution right is brutally difficult - in fact, come to think of it, my whole PhD thesis was about modelling the tails of a similar distribution correctly.

BFC has never commented on the long right tails, but they are a consistent feature of the CM spotting model across titles. I mentioned earlier that I have seen results longer than 10 minutes at ranges up to 1000m in CMBN but they are very rare.

Thanks for running the tests! I would not have been surprised by an even larger M60 advantage given the thermal imager. I picked the RISE+ in my test as I felt that was more apples to apples vis-a-vis the T-72A.

Edited May 15, 2023 by Vanir Ausf B

[Butschi]

35 minutes ago, Vanir Ausf B said:

BFC has never commented on the long right tails, but they are a consistent feature of the CM spotting model across titles. I mentioned earlier that I have seen results longer than 10 minutes at ranges up to 1000m in CMBN but they are very rare.

I guess it can be as simple as calculating a probability for a vehicle (or probably for every crew member) to spot the target and then every second they draw a random number to check if the spotting process was successful (the calculation of the probability may be very much not simple!). In that case the probability to not have a spot after n tries is (1-p)^n, which is (chi-by-eye...) consistent with the distributions above. It is never 0 even after 100 minutes, just very small.

[The_Capt]

4 hours ago, Butschi said:

Let's calm down a little, shall we?

This unpleasant "discussion" made me dig out a test I did some time ago. I've been doing data analysis for basically all my professional life, so I always suffer a bit when reading these threads. So, as we've been discussing on page one or so, anecdotal evidence ("here look at this battle, spotting is broken!!!!") is meaningless. But also doing "experiments" is only as good as the experiment itself plus the evaluation afterwards. First of all, if you do experiments, control the variables! Eliminate everything you are not directly interested in. If you want to look at spotting, do it on a flat surface and make both opponents hold fire. Because, as I often see, if you measure time until first shot, or kill, you are skewing the spotting process. Next, don't look at averages or medians alone. Look at distributions. And don't eliminate outliers. My text book about statistical data analysis said that, eliminating outliers, although done often, should really only be done if you know what you are doing, e.g. when you know that your outlier is actually some measurement error and not some rare event.

So, here's what I did:

I put an M60 TTS and a bog standard T72 (the exact models aren't relevant for the method) on a flat map with paved ground, roughly 2 km apart. I set both vehicles to hold fire. I then measured the time it took for each tank to first get a partial contact and then a full contact.

Here is the raw data:

t72 = [84, 17, 78, 4, 174, 65, 77, 321, 289, 444, 31, 3, 290, 2, 40, 120, 40, 159, 57, 69, 15, 54, 80, 95, 19, 58,
       23, 672, 154, 154, 17, 14, 342, 12, 386, 43, 84, 12, 378, 123, 30, 44, 240, 311, 110, 2, 68, 181, 137]

t72_id = [7, 21, 7,28, 7, 35, 14, 0, 56, 0, 35, 0, 7, 70, 42, 7, 35, 35, 0, 49, 0, 7, 0, 0, 0, 63, 28, 7, 14, 0,
          7, 7, 7, 49, 7, 21, 14, 21, 14, 0, 7, 14, 28, 0, 14, 7, 28, 56, 77]

m60 = [89, 68, 41, 38, 71, 10, 73, 0, 20, 9, 4, 55, 91, 34, 31, 8, 14, 116, 64, 4, 18, 63, 116, 38, 3, 18, 71, 132,
       39, 73, 43, 73, 116, 210, 207, 36, 180, 27, 88, 48, 102, 3, 52, 77, 176, 22, 18, 80, 24]

m60_id = [7, 7, 7, 0, 14, 14, 0, 14, 14, 7, 14, 0, 7, 14, 35, 14, 7, 0, 7, 14, 7, 0, 14, 7, 7, 7, 14, 35, 7, 0, 7, 0,
          0, 7, 0, 7, 14, 14, 7, 7, 7, 0, 7, 0, 0, 7, 21, 7, 7]

Times are in seconds, the entry t72 is time until partial contact for the T72 trying to spot the M60, the one with  "_id" is the difference between partial and full contact. The same for m60. And in order to get the distributions I made histograms with 30s bins for plots 1 and 3 and 7s for plot 2.

So, what do we see here? Well, first of all, I should have taken at least ten times the data or make make larger bins. I didn't have the patience for the former and doing the latter would mean that we don't see much of a distribution. Anyway, from the raw data we see: time until partial contact can be any number, time to ID (which is what I call time to go from partial to full contact) is always a multiple of 7.

The histograms tell us the following: Although it is not possible to get the exact distribution, this is definitely not something symmetrical where average or media are easy to interpret. If you look at the bins with the highest counts, those are at low times. That means, players will usually see that their T72 or M60 are often quick to spot their target. Nothing to complain about or make a forum "rant". But for both (!) tanks it is quite possible that it takes several minutes - more likely for the T72 but also the M60 TTS had an event with over 3.5 minutes. The time until partial contact is consistent (no more, no less) with an exponential decay which you would expect when counting the number of dice rolls necessary to roll a specific number - only that the M60 TTS rolls with a D6, while the T72 rolls a D20, so to speak.

Now, is spotting "broken" or not? For me, this is really not a meaningful thing to say. Because by "broken" people usually mean "takes too long" without saying what "too long" is and why. With the above distributions, it is possible to never spot the target. Right? Wrong? Broken? This is a game/simulation and as any such thing, at least if you want it to run in reasonable time on a consumer machine, it is simpler than real life and abstracted. A meaningful discussion would be "Is this spotting model adequate? Could CM do better by applying model XYZ, instead?". I'm not sure if the long tails (what some call "outliers") are working as intended (although I find @The_Capt analogy with the barrel quite convincing) or if it is a model that is just designed to get the "center", the common situations, right, accepting that every now and then it produces something odd. But getting the tails of an exponentially decaying distribution right is brutally difficult - in fact, come to think of it, my whole PhD thesis was about modelling the tails of a similar distribution correctly.

 

Excellent, far better than anything I ever did up.  So from this I am less concerned with the far outliers - crews doze off or any number of reasons can have long lapses (eg miscommunication between crew is a big one).  What is odd is that the far outliers are only happening to the T72s.  I mean US crews can doze off or think the other guys is keeping eyes out as well. (however you do mention a 3.5 min lag, which for a TTS equipped system is pretty darn long).  Perhaps the extremity of the outliers is linked to the platform equipment?

I think the “too long” impulse is 1) based on unrealistic assumptions with no real world experience and/or 2) 3 mins is an eternity in a tense game.  The reality is that in combat conditions spotting is hard and it is very possible to roll a natural “1”.  The answer to this is not to scratch off natural “1”s as it would make the game less realistic, not more.

[Vanir Ausf B]

16 minutes ago, The_Capt said:

crews doze off or any number of reasons

The TC had to piss, and no it could not wait.

[chuckdyke]

Tank Crews can be unionized. 4PM and it is time for a Devonshire Tea with scones cream and homemade jam. 

[IdontknowhowtodoX]

10 hours ago, Butschi said:

Let's calm down a little, shall we?

This unpleasant "discussion" made me dig out a test I did some time ago. I've been doing data analysis for basically all my professional life, so I always suffer a bit when reading these threads. So, as we've been discussing on page one or so, anecdotal evidence ("here look at this battle, spotting is broken!!!!") is meaningless. But also doing "experiments" is only as good as the experiment itself plus the evaluation afterwards. First of all, if you do experiments, control the variables! Eliminate everything you are not directly interested in. If you want to look at spotting, do it on a flat surface and make both opponents hold fire. Because, as I often see, if you measure time until first shot, or kill, you are skewing the spotting process. Next, don't look at averages or medians alone. Look at distributions. And don't eliminate outliers. My text book about statistical data analysis said that, eliminating outliers, although done often, should really only be done if you know what you are doing, e.g. when you know that your outlier is actually some measurement error and not some rare event.

So, here's what I did:

I put an M60 TTS and a bog standard T72 (the exact models aren't relevant for the method) on a flat map with paved ground, roughly 2 km apart. I set both vehicles to hold fire. I then measured the time it took for each tank to first get a partial contact and then a full contact.

Here is the raw data:

t72 = [84, 17, 78, 4, 174, 65, 77, 321, 289, 444, 31, 3, 290, 2, 40, 120, 40, 159, 57, 69, 15, 54, 80, 95, 19, 58,
       23, 672, 154, 154, 17, 14, 342, 12, 386, 43, 84, 12, 378, 123, 30, 44, 240, 311, 110, 2, 68, 181, 137]

t72_id = [7, 21, 7,28, 7, 35, 14, 0, 56, 0, 35, 0, 7, 70, 42, 7, 35, 35, 0, 49, 0, 7, 0, 0, 0, 63, 28, 7, 14, 0,
          7, 7, 7, 49, 7, 21, 14, 21, 14, 0, 7, 14, 28, 0, 14, 7, 28, 56, 77]

m60 = [89, 68, 41, 38, 71, 10, 73, 0, 20, 9, 4, 55, 91, 34, 31, 8, 14, 116, 64, 4, 18, 63, 116, 38, 3, 18, 71, 132,
       39, 73, 43, 73, 116, 210, 207, 36, 180, 27, 88, 48, 102, 3, 52, 77, 176, 22, 18, 80, 24]

m60_id = [7, 7, 7, 0, 14, 14, 0, 14, 14, 7, 14, 0, 7, 14, 35, 14, 7, 0, 7, 14, 7, 0, 14, 7, 7, 7, 14, 35, 7, 0, 7, 0,
          0, 7, 0, 7, 14, 14, 7, 7, 7, 0, 7, 0, 0, 7, 21, 7, 7]

Times are in seconds, the entry t72 is time until partial contact for the T72 trying to spot the M60, the one with  "_id" is the difference between partial and full contact. The same for m60. And in order to get the distributions I made histograms with 30s bins for plots 1 and 3 and 7s for plot 2.

So, what do we see here? Well, first of all, I should have taken at least ten times the data or make make larger bins. I didn't have the patience for the former and doing the latter would mean that we don't see much of a distribution. Anyway, from the raw data we see: time until partial contact can be any number, time to ID (which is what I call time to go from partial to full contact) is always a multiple of 7.

The histograms tell us the following: Although it is not possible to get the exact distribution, this is definitely not something symmetrical where average or media are easy to interpret. If you look at the bins with the highest counts, those are at low times. That means, players will usually see that their T72 or M60 are often quick to spot their target. Nothing to complain about or make a forum "rant". But for both (!) tanks it is quite possible that it takes several minutes - more likely for the T72 but also the M60 TTS had an event with over 3.5 minutes. The time until partial contact is consistent (no more, no less) with an exponential decay which you would expect when counting the number of dice rolls necessary to roll a specific number - only that the M60 TTS rolls with a D6, while the T72 rolls a D20, so to speak.

Now, is spotting "broken" or not? For me, this is really not a meaningful thing to say. Because by "broken" people usually mean "takes too long" without saying what "too long" is and why. With the above distributions, it is possible to never spot the target. Right? Wrong? Broken? This is a game/simulation and as any such thing, at least if you want it to run in reasonable time on a consumer machine, it is simpler than real life and abstracted. A meaningful discussion would be "Is this spotting model adequate? Could CM do better by applying model XYZ, instead?". I'm not sure if the long tails (what some call "outliers") are working as intended (although I find @The_Capt analogy with the barrel quite convincing) or if it is a model that is just designed to get the "center", the common situations, right, accepting that every now and then it produces something odd. But getting the tails of an exponentially decaying distribution right is brutally difficult - in fact, come to think of it, my whole PhD thesis was about modelling the tails of a similar distribution correctly.

 

It's MATH! RUN FOR YOUR LIFE!

Just joking. As for CMCW ability of spotting is just one part of the game. US tanks can't convert ability of spotting to ability of killing completely. Even if m60s can spot t64s/t72s/t80s faster, it would usually take two or three rounds to kill or just immobilize opponent. However soviet tanks can do one shot one kill thing. It's useless that just seeing enemy tanks if you can't kill them.

[Brille]

By the way: has someone a good source (or multiple) that specifically focuses on optics of the various vehicles, including commanders sight? Maybe even with some pictures through the actual things to get an understanding as a non-tanker. 

I often read that soviet style optics have a poor field of view while they may have some decent magnification.

I had found some technical data on T55 and T72 commanders sights which would confirm this. 

Haven't found anything (despite the gunners sight) on a T34/85 to compare them and see the evolution of optics there. 

So magnification is good and all but if you are looking through a toilet paper roll it might be hard to actually spot something because you have to be right on point. 

And this shows in CMCW I guess. 

Edited May 16, 2023 by Brille

[The_Capt]

9 hours ago, Vanir Ausf B said:

The TC had to piss, and no it could not wait.

Minutes go by

”Hey you guys still awake?  Anybody see anything?”

”Nothing sarge.  Except a rabbit that I am pretty sure is inbred.  I watched the stupid thing run headfirst into a fence pole…twice”

”Heh, I think I saw the same one, its fur is messed up, bald patch on its butt.”

Pause

”Hey what are you covering right now?”

”90 to 12 degrees like you said sarge.”

”I said to cover ‘9 to 12 o’clock!”

”Ah crap..”

Electro-mechanical sounds

“Ah ffffuuu…Contact!!”

And we get 700 seconds in a spotting delay.

 

[slysniper]

When is someone going to point out CM has a program switch that once you complain enough on the forum.

they activate the switch on that gamers program which adds unpredictable events in their game just to irritate them all the more.

We can see who has had their switch activated here.

[Artkin]

We can see the problem finally gets the attention it deserves*

[Artkin]

3 hours ago, The_Capt said:

Minutes go by

”Hey you guys still awake?  Anybody see anything?”

”Nothing sarge.  Except a rabbit that I am pretty sure is inbred.  I watched the stupid thing run headfirst into a fence pole…twice”

”Heh, I think I saw the same one, its fur is messed up, bald patch on its butt.”

Pause

”Hey what are you covering right now?”

”90 to 12 degrees like you said sarge.”

”I said to cover ‘9 to 12 o’clock!”

”Ah crap..”

Electro-mechanical sounds

“Ah ffffuuu…Contact!!”

And we get 700 seconds in a spotting delay.

 

Why isnt this happening to US crews in the same way then....? 

In a forested area it's believable.

In an open grass area it's not. 

If there's no way for the game to distinguish between the two... thats a problem. 

[Butschi]

21 minutes ago, Artkin said:

Why isnt this happening to US crews in the same way then....? 

It is. Same mechanics, different probability.

[The_Capt]

3 hours ago, Artkin said:

Why isnt this happening to US crews in the same way then....? 

In a forested area it's believable.

In an open grass area it's not. 

If there's no way for the game to distinguish between the two... thats a problem. 

I am pretty certain the game can tell the difference between forest terrain and open grass.  Open grass is not really "open", Steve has said this many times.  It is filled with dimples and dips etc.  He has said this since CMx1.  My guess (engine IP is held very tightly and I do not blame them) is that there are layers that effect the spotting curves.  Heavier terrain makes spotting dice rolls harder.

The only thing I can see about the US crews is that they have the TTS system.  The M60 has far better FCS and ergonomics (it is why it is so big) so they likely factored that into the outliers.  However 3.5 minutes for an M60 TTS is still a pretty damned long time.

[Butschi]

13 minutes ago, The_Capt said:

I am pretty certain the game can tell the difference between forest terrain and open grass.  Open grass is not really "open", Steve has said this many times.  It is filled with dimples and dips etc.  He has said this since CMx1.  My guess (engine IP is held very tightly and I do not blame them) is that there are layers that effect the spotting curves.  Heavier terrain makes spotting dice rolls harder.

The only thing I can see about the US crews is that they have the TTS system.  The M60 has far better FCS and ergonomics (it is why it is so big) so they likely factored that into the outliers.  However 3.5 minutes for an M60 TTS is still a pretty damned long time.

I can't prove it, of course, but I really don't think that the "outliers" are the result of any explicit modelling.

A I said, it looks like my data is consistent with something like (1-p)^n (p being a probability and n the number of dice rolls).

@domfluff probably described it more accurately:

On 4/22/2023 at 7:25 PM, domfluff said:

Also, of course there's an RNG aspect to spotting - that's how spotting has been modelled since there have been models for spotting.

As the basic example, Koopman in Search and Screening (1946), who theorised that the detection rate is proportional to the solid angle subtended at the point of observation of the target.

Since your chances of finding something is going to be harder the larger an area you're looking at, that's a base-e relationship of some kind. The Koopman probability of a detection in time t is P(t) = 1 - e^(-yt), with your y in Koopman theory being y=kh/r^3, h and r describing the height and distance to the target, and k being a value for how complex the search operation is.

Call that a "dice roll" if you like, but that's how modelling this kind of thing usually goes.

 

His formula basically comes done to the same, though. If true, the outliers are the result of the same process as the "normal events". There would simply be a different gamma (or p in my case) for the M60 than for the T72, making the distribution for the M60 steeper and thus generate those outliers less often.

[The_Capt]

3 hours ago, Butschi said:

I can't prove it, of course, but I really don't think that the "outliers" are the result of any explicit modelling.

A I said, it looks like my data is consistent with something like (1-p)^n (p being a probability and n the number of dice rolls).

@domfluff probably described it more accurately:

His formula basically comes done to the same, though. If true, the outliers are the result of the same process as the "normal events". There would simply be a different gamma (or p in my case) for the M60 than for the T72, making the distribution for the M60 steeper and thus generate those outliers less often.

Could you not make k variable, perhaps even dynamic?  The complexity of the observation is mitigated by a series of factors both hard (equipment) and soft (experience, condition).  So the k for the M60 would be different than the T72 based on these factors?