Frustration with CMCW - Russian side

[Butschi]

2 minutes ago, The_Capt said:

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?

That's what I meant, I think.

Btw. This is all about gnerating the partial contact. The "ID step" seems to work a bit different. I'd guess that it is a random process abstracting performing some task, something like zooming to the partial contact, switching on thermals, etc. Really only blindly speculating here, but maybe something like where you get an additional die for your dice roll which you perform every 7 seconds. You have chance to roll the required result right away but the longer it takes the likelier it gets.

[domfluff]

In terms of a Koopman search, absolutely. Depending on the model, how you define and vary K is pretty much the definition of the level of precision of the model (and more precise isn't necessarily better).

The main thing is really that there's an e in there somewhere - the larger the area there is to search, the harder it is to find. That kind of self-referential thing usually needs an e or a natural log.

Now, I have no idea if CM spotting uses a Koopman search - that's just the oldest and one of the most basic models that could be used. The outputs are likely to look really similar though.

Edited May 16, 2023 by domfluff

[Lethaface]

On 5/10/2023 at 11:13 PM, Sarjen said:

Just some info regarding the IR capabilities of the T-80s: the early T-80s uses the TPN-3 which is equipped with a much more sensitive infrared image converter assembly than its predecessor. The passive range of the early versions is between 500 and 800m. With the support of the modernized infrared headlight L4A, which had received a more luminous lamp, the visibility in active operation increased to about 1200 m. Still worse than the western technology but still more than 20m as in my video example. 

IIRC in CM the IR searchlights aren't ever turned on for any side, because they would stand out too much. So the IR headlight isn't in operation AFAIK.

[Lethaface]

On 5/14/2023 at 5:46 PM, 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.

That's good news I didn't read before and IMO well warranted.

To be fair we did have plenty of discussion about that one, but no changes to the simulation even though quite some work was done on the subject pointing towards the simulation being too generous on tanks. I did a bit of testing / work for that myself, especially CMBS M1A2 and Oplot-M dealing with 152/203mm direct and indirect hits. 

I might have missed it but until now I didn't read about plans to change this behavior. Better late than never ;-), and thanks for the update! I feel such significant decisions should/could be posted/collected in a sticky adminonly post on the General Forum.
If anything they will help motivate people to do work when they anecdotally start to develop an intuition that there is something off in the simulation (they care about).

On the spotting subject: I don't have much issues with spotting with RED after who knows how many hours of CMx2 since 2007. And while atm I can't put many hours, almost all of my playtime is in PBEM in some organized format; there's certainly players who can make the RED spotting 'work for them'.  
Perhaps the outliers or assessed capabilities of vehicles like a T-90AM in CMBS, or the Khrizantema, feel glass half-empty while they appear half-full for the M1A2SEPv2/3. 
IMO a lot of the spotting is often impacted by C2 sharing, that's more difficult for RED in CM (which I'd say is fair) but is manageable as a player and to large affect. There are some c2 issues for at least Syrians in CMSF2, but that's another subject.

Edited May 16, 2023 by Lethaface

[Andrew Kulin]

23 hours ago, Butschi said:

Let's calm down a little, shall we?

Thank you for this post.  Interesting way to do it and I think I will attempt something similar at my end.

My take on your histogram plots are:

• The M60 partial spots the T72 within 1 minute half the test (n=27/50) and within 90 seconds in almost four-fifths of the test (n=39/50)
• The T72 managed about 40% partial spots within 60 seconds (n=21/50), and about 60% within 90 seconds (n=29/50).  No issue with either of these, seems reasonable to me.
• Both vehicles convert partial spots into confirmed spots generally within 60 seconds after making the partial spot.
• What I found odd in the data is the longer tail for the partial spotting of the T72 vs. M60.  The M60 was at 100% partial spotting success by 210 seconds (3:20 minutes), whereas the T72 was at 80% with the remaining spots taking essentially up to about 6 minutes, plus that almost 700 second outlier.  I am wondering if you assign any significance to the lack of a long tail with the M60 results?  I guess a similar question could be asked of the data pertaining to converting a partial spot to a full spot.  Again I see a longer tail with the T72 results than the M60.

Were the tanks buttoned up or unbuttoned when you ran your tests?  I did not see reference to that in your post.  My frustrations lay in the perceived poor spotting when my tanks were unbuttoned.  I am fully expectant that the western technology related to spotting (thermals, night vision, lasers etc.) would be superior to Soviet equivalents, so I am not surprised in buttoned up spotting is superior on the American side.  And it is the reason why I had my vehicles, sitting inside the tree line waiting in ambush to try and improve the odds of spotting the American forces that were advancing on my positions.

And as a statistician, which I presume you are as I think you stated you have a PhD and did this sort of thing for a living (at least that is how I interpreted your posts), if I had say a half-dozen vehicles with eyes on the area where my opponent's Bradleys and M1 were moving through open field, on a clear sunny day, would you expect it to be statistically reasonable that all my vehicles had the same poor (tail end of the distribution) spotting success?  As in not a single one seeing a single enemy vehicle within 5+ minutes of game time.

 

[Lethaface]

On 5/15/2023 at 6:57 AM, IdontknowhowtodoX said:

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.

Exactly. Even though it isn't always perfect, sometimes very far from perfect, it's exactly what keeps me/us playing this simulation which was originally released in 2007 (although significantly developed on since).

[Lethaface]

6 hours ago, Butschi said:

That's what I meant, I think.

Btw. This is all about gnerating the partial contact. The "ID step" seems to work a bit different. I'd guess that it is a random process abstracting performing some task, something like zooming to the partial contact, switching on thermals, etc. Really only blindly speculating here, but maybe something like where you get an additional die for your dice roll which you perform every 7 seconds. You have chance to roll the required result right away but the longer it takes the likelier it gets.

Interesting discussion by the way

[Butschi]

1 hour ago, Andrew Kulin said:

Were the tanks buttoned up or unbuttoned when you ran your tests?

Buttoned up.

1 hour ago, Andrew Kulin said:

What I found odd in the data is the longer tail for the partial spotting of the T72 vs. M60.  The M60 was at 100% partial spotting success by 210 seconds (3:20 minutes), whereas the T72 was at 80% with the remaining spots taking essentially up to about 6 minutes, plus that almost 700 second outlier.  I am wondering if you assign any significance to the lack of a long tail with the M60 results?  I guess a similar question could be asked of the data pertaining to converting a partial spot to a full spot.  Again I see a longer tail with the T72 results than the M60.

Why do you think this is odd? It's really more like a too small sample size. I'm not claiming that what I say is exactly true - in fact, back at university I'd have been flogged for making such bold statements with so little data. I did a "simulation", meaning, "made a dice roll", i.e. I generated a random number, repeatedly, until it was < 0.01 or 0.02, respectively. That is about the order of magnitude (very roughly! I don't know the exact values!) for the spotting probability in the scenario I described (0.02, or 2% for the "simulated" M60). I repeated this 50 times for each "tank". The result looked like this:

Different histogram but same phenomenon. Long tail for orange and seemingly no tail for blue. Now, exactly same parameters and setup but 5000 "experiments" for each tank.

Here you see that both have long tails, and if I were to repeat the experiment a million times each, you would probably see that both get events out to 700. Only that orange gets way, way, way more of them. Just by having 1% instead of 2% probability for each dice roll.

1 hour ago, Andrew Kulin said:

And as a statistician, which I presume you are as I think you stated you have a PhD and did this sort of thing for a living (at least that is how I interpreted your posts), if I had say a half-dozen vehicles with eyes on the area where my opponent's Bradleys and M1 were moving through open field, on a clear sunny day, would you expect it to be statistically reasonable that all my vehicles had the same poor (tail end of the distribution) spotting success?  As in not a single one seeing a single enemy vehicle within 5+ minutes of game time.

Wish I was the statistics expert, I'm a particle physicist and some statistical data analysis was part of my PhD and later in my job. But in the land of the blind, the one-eyed man is king, I guess. What you describe may be unlikely (though that depends on what the actual parameters governing this situation are) but remember this: There are thousands of players out there making thousand upon thousands of dice rolls each day. The probability that someone observes such a situation (and then makes a frustrated post on the forums) is actually not that small.

[Artkin]

8 hours 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.

Yes I noticed the 3.5 minutes immediately. That does seem incredibly long for any tank in CMCW.

What I meant was: you're going to end up with these massive outliers whether your tank is in an open field or in a dense forest, both with a clear FOV to target. Perhaps the outliers exist because the crew is searching around them. It makes no sense in an open grass field, but it would make sense if there were trees surrounding the vehicle (But a clear LOS to target still). It makes sense that the game would be unable to differentiate between the two situations so either way there would be outliers. I guess it could be a trade-off is what I'm saying. If so, I'm not sure if it's a good solution - hence me claiming there's a problem.

Of course flat grass is not exactly flat, but we can set elevation in the editor at 1m intervals. So it's pretty safe to assume the grass is not deviating up or down 1m, which is pretty negligible at these distances with these sized targets.

Edited May 16, 2023 by Artkin

[Butschi]

14 minutes ago, Artkin said:

Yes I noticed the 3.5 minutes immediately. That does seem incredibly long for any tank in CMCW.

What I meant was: you're going to end up with these massive outliers whether your tank is in an open field or in a dense forest, both with a clear FOV to target. Perhaps the outliers exist because the crew is searching around them. It makes no sense in an open grass field, but it would make sense if there were trees surrounding the vehicle (But a clear LOS to target still). It makes sense that the game would be unable to differentiate between the two situations so either way there would be outliers. I guess it could be a trade-off is what I'm saying. If so, I'm not sure if it's a good solution - hence me claiming there's a problem.

Of course flat grass is not exactly flat, but we can set elevation in the editor at 1m intervals. So it's pretty safe to assume the grass is not deviating up or down 1m, which is pretty negligible at these distances with these sized targets.

How else would you suggest to model the spotting process? As I tried to outline, the "outliers" are probably in no way special, just the results of a statistical process. For simplicity, let's assume you turn your head with constant velocity, making a full sweep of the horizon every 10 seconds. You can, of course, remove the randomness completely but how is that realistic, there is never a 100% probability to spot something. Perhaps you blink in exactly the wrong moment. So, instead it makes more sense to give this "spotting test" some probability p. If you do it like that, you end up with a probability to not have spotted a target sitting there somewhere after n sweeps that is something like (1-p)^n. Now, if you do the math, you will see the probability drops exponentially but is never 0, even if you set p to 99.9%. So you will still end up with events where you haven't spotted a target after 3.5 minutes. It just won't happen as often. Now, you suggest that 3.5 minutes are a problem but are they just happening to often or should they never happen at all? If the latter then you would have to introduce some kind of cut-off. How would you justify that? There is no law of nature that says that after having looked in the direction of something you have to spot it, no matter what, after max. 30 seconds. Even were you to suggest that the individual probability to spot the target were to increase with every sweep (and I don't see why you should be more likely to spot something with the 101st try that you didn't see the last 100 times), you would still have a non-zero probability that it takes 3.5 minutes. 

I'm not saying that this way of modelling spotting is the best way to do it but so far, nobody here has come up with a more convincing model...

[Pelican Pal]

The best improvement to the system would be an increase in spotting cycles. Right now its every 7 seconds(?) or every stance change. So a T-72 will attempt to spot an enemy tank 8 times per turn. 3.5 minutes is like 28 spotting attempts. If you were to cut the spotting cycle in half that 3.5 minutes would end up being like 1.75 minutes.
 

[Artkin]

15 minutes ago, Butschi said:

Even were you to suggest that the individual probability to spot the target were to increase with every sweep (and I don't see why you should be more likely to spot something with the 101st try that you didn't see the last 100 times), you would still have a non-zero probability that it takes 3.5 minutes.

I would suggest this, increase the spotting probability. I'm not suggesting taking away randomness entirely, but it appears to happen so often that a few of us have complained. And then this brings me back to my point... the US probability is significantly different in comparison to the Soviet. It produces a large difference at times

Edited May 16, 2023 by Artkin

[Artkin]

On 5/14/2023 at 2:46 PM, Vanir Ausf B said:

Not all Soviet vehicles are worse at spotting than all US vehicles. For example the Shturm-S is very good at spotting, much better than a M60.

So I ran a very quick test with sturms at 2km. They did well and spotted tanks almost instantly. Now, I put them inside "Tree A" and their spotting quickly suffers. They have good LOS and are unable to spot the tanks after 3 minutes. This is exactly how I had them set up when the "blob of us vehicles" was coming toward me. Since the shturms sights are low to the ground I didn't expect Tree A to have such a debilitating effect on spotting.

At the end of the range there are 3 pattons facing opposite my three shturms (I've yet to turn the pattons around and see what they see). one patton was killed already, you can see the smoke. This shturm has good LOS but is unable to see the pattons. I'll continue running the test scenario and see what happens. It's been a few minutes since I've restarted it and the shturms failed to see a single patton this time.

Flat grass tests only tell so much, IMO. And what you see apparently isn't what you get, still.

Edited May 16, 2023 by Artkin

[Artkin]

After five minutes of being in the same spot, running the same test three times for five minutes, but on the final test adding flavor objects so I'm sure of the patton's location.

It sure does seem like something wacky is going on, and this exactly matches (in regards to shturm spotting specifically) my previous experience. Neither side sees each other despite having good LOS:

Edit: Not every position in the treeline is safe. I tried moving forward the other two shturms (The ones that I didn't draw LOS with) and they were quickly wiped out. The one shturm featured in the above post, and down below with the LOS tool is still unable to see anything after 10 minutes, despite having good LOS and both pattons firing and exposing their positions.

Edited May 16, 2023 by Artkin

[Vanir Ausf B]

1 hour ago, Artkin said:

It sure does seem like something wacky is going on, and this exactly matches my previous experience. Neither side sees each other despite having good LOS:

There is some wackiness going on but it is due to the way foliage affects spotting. Judging from your screen shots I suspect LOS is being degraded by tree branches, despite appearances to the contrary and despite what the LOS line says.

In Combat Mission tree trunks are accurately represented visually but tree canopies are significantly abstracted. In my experience tree canopies are both less opaque and lower to the ground "under the hood" than their visual representation suggests.

This unintuitiveness is compounded by the target line lacking LOS context. When checking LOS with the target command the LOS line is binary -- you either have it or you don't. But under the hood LOS through trees is non-binary. Tree branches and leaves degrade LOS proportional to how much tree canopy the line passes through "under the hood". The target command line will show clear LOS up to a certain amount of degradation, then at some point will change to show LOS blocked even though spotting is still possible.

Edited May 17, 2023 by Vanir Ausf B

[Artkin]

24 minutes ago, Vanir Ausf B said:

The target command line will show clear LOS up to a certain amount of degradation, then at some point will change to show LOS blocked even though spotting is still possible.

If I understood correct, the exact opposite is happening in my screenshots right? I have a clear target line but no LOS. I've never seen it this way.

I have experienced when my vehicle spots something (Has LOS) and is unable to draw a target line to it, but that's typically only when one of the vehicles is in defilade or behind tall grass.

[Vanir Ausf B]

39 minutes ago, Artkin said:

If I understood correct, the exact opposite is happening in my screenshots right? I have a clear target line but no LOS. I've never seen it this way.

You have LOS, it's just that your spotting checks are being heavily penalized. If you wait long enough the units may spot each other. Even when the target line shows no LOS through trees units can sometimes spot each other. That's why you shouldn't  trust trees to hide your units unless there is a LOT of foliage between you and the enemy.

At least that's what I think is most likely happening. The lack of 1 to 1 graphical representation of tree canopies make it difficult to know for certain how LOS is being affected in any situation.

[domfluff]

We know the broad outlines of how this works in CM.

LOS and line of fire are different calculations, and this includes the concept of "hindrance", as per Advanced Squad Leader or similar.

Broadly, you take the terrain features of each tile between firer and target, and the most-blocking terrain (e.g., tall grass) will negatively affect the spotting calculations.

The actual firing is calculated explicitly. Mass, air resistance, etc. are all modeled essentially 1:1, but spotting is a slightly more abstract model.

This, again, is how spotting is modeled in everything that tries to model spotting, since spotting models have existed. Stochastic simulation is a close match for real world behaviour, especially when you're looking at a game with the level of fidelity of cm (as in, platoon level).

The detection/identification step is part of that, naturally - cm's identification model presumably isn't particularly complex, since partial spots are clearly "tank" or whatever, and not more granular than that.

 

[domfluff]

Which is a long way to say "yes, terrain affects line of sight", which I'd have thought would be obvious.

[AlexUK]

It does seem like a problem to me. Isn’t it advised tactically to place your armour within the tree line when setting up ambush positions? If CM current setup means this is penalised then it would be a problem. Intuitively you would expect the opposite to occur. The ambusher spots the moving target in open ground much easier and the ambushed having a very hard time spotting the stationary, in shadow, and partially concealed by foliage ambusher. 

[slysniper]

Since cm is not perfect as to if the unit can see or not see out of the woods, the easy answer is. When its time to spring the attack. move the units forward just out of the woods and the issue is solved.

And since there is no spotting penetly for the moving unit, it clears up most of the issues here. 

And no, cm does not have anything built in it presently for making it harder to spot something that is covered in shadows or has its outline  blocked by vegetation hidding its outline. So no it will not give them type of advantages that would be there in real world spotting.

 

[Andrew Kulin]

2 hours ago, slysniper said:

Since cm is not perfect as to if the unit can see or not see out of the woods, the easy answer is. When its time to spring the attack. move the units forward just out of the woods and the issue is solved.

My experience with the modern CM titles is that movement into LOS of the enemy usually leads to a quick death.  Particularly as to regards with Soviets movement.  On defensive scenarios I set up and/or move my Soviet/Syrian units into positions (hull down, inside tree lines, etc.) before the NATO side arrives on scene if I can.

 

2 hours ago, slysniper said:

And since there is no spotting penetly for the moving unit, it clears up most of the issues here.

Did not realize that (or forgot it).

[A Canadian Cat]

15 hours ago, Butschi said:

If you do it like that, you end up with a probability to not have spotted a target sitting there somewhere after n sweeps that is something like (1-p)^n. Now, if you do the math, you will see the probability drops exponentially but is never 0, even if you set p to 99.9%. So you will still end up with events where you haven't spotted a target after 3.5 minutes. It just won't happen as often.

Thanks for that explanation. Its always nice to see the math for what you intuitively understand.

 

15 hours ago, Butschi said:

Now, you suggest that 3.5 minutes are a problem but are they just happening to often or should they never happen at all?

This is an issue for sure. Even if tweaks were made to reduce the number of long blind periods we would still see a post every now and then complaining that spotting is broken. Honestly I suspect we would see that even if there was a cut off implemented. It's just the way we humans are wired.

 

15 hours ago, Pelican Pal said:

The best improvement to the system would be an increase in spotting cycles. Right now its every 7 seconds(?) or every stance change. So a T-72 will attempt to spot an enemy tank 8 times per turn. 3.5 minutes is like 28 spotting attempts. If you were to cut the spotting cycle in half that 3.5 minutes would end up being like 1.75 minutes.

I don't think that would help though. Well not for this issue. If the number of spotting cycles was increased the probability of spotting things would also have to drop.  As @Butschi points out that will still mean there will be data points on the log tail.

Unless you are actually saying that everything should be easier to spot and the game should just be changed to make spotting easier. Honestly I think the feeling I get from seeing all this combat footage from Ukraine is that spotting the enemy is harder IRL than in the game now.

[Pelican Pal]

My thought would be that more checks would ameliorate some of the edge cases. If the game checked every 3 seconds you'd nearly triple the number of dice rolls and be less likely to suffer from a string of "bad rolls". 

 

1 hour ago, IanL said:

Unless you are actually saying that everything should be easier to spot and the game should just be changed to make spotting easier. Honestly I think the feeling I get from seeing all this combat footage from Ukraine is that spotting the enemy is harder IRL than in the game now.

Having watched a fair amount of footage from Ukraine what I've generally seen is that while spotting is very hard it seems generally easier to put fire down near the enemy than CM allows. CM generally requires a full spotting resolution before a model will fire unless the player interferes. It might be more accurate to have models firing at the ? before they know exactly what or where it is. You'd get less accurate fire but more fire.

[Butschi]

35 minutes ago, Pelican Pal said:

My thought would be that more checks would ameliorate some of the edge cases. If the game checked every 3 seconds you'd nearly triple the number of dice rolls and be less likely to suffer from a string of "bad rolls". 

It doesn't work like that. Imagine CM to be a board game. Every turn you get to roll a die (the normal 6 sided one) and if it shows a 6 you spot the enemy tank. On average it will take your tank 6 turns to spot the tank but 16% of the time or so it will take 10 turns or more. Of course, if you roll 2 dice per turn, the probability that it takes you 10 turns or more to spot is lower. But also the average is lower, so you effectively make spotting easier. To mitigate that you can use dice with more sides but then the probability that you need more than 10 turns increases again...

50 minutes ago, Pelican Pal said:

Having watched a fair amount of footage from Ukraine what I've generally seen is that while spotting is very hard it seems generally easier to put fire down near the enemy than CM allows. CM generally requires a full spotting resolution before a model will fire unless the player interferes. It might be more accurate to have models firing at the ? before they know exactly what or where it is. You'd get less accurate fire but more fire.

Remember, though, that, contrary to real life, your Pixeltruppen never misidentify one of their own as an enemy. So, if not requiring a positive ID, blue on blue kills should be a thing, too. (And I think that would be awesome! The chaos, the drama!)