Russians Underpowered, US Overpowered in CMBS?

[IMHO]

2 hours ago, Rinaldi said:

It's lead to considerable differences in results from those who did considerably less testing; which basically leads me to throw your reasoning out the window immediately.

Well don't throw me just yet  Because when doing off the cuff sampling for discrete sets you needn't look at the total sample size but at the share of EACH of the discrete values in the sample size. So if you say that you want to understand the probability for a vehicle to bog, you ran 200 tests and the vehicle was immobilized just once - I'd say keep trying  But if you ran 20 test and had 10 immobilizations I'd say may be it's discussable. But that's off the cuff - for important cases you don't stop even there. You have to understand how "the system works" and isolate the significant factors to build a robust sample relative to these factors. Because when you run test after test in CM you make one implicit assumption - that this set is random. And in real life there may well be some unknown factor that introduces bias so that even millions of events do not give you a representative sample  Like when you say that for spotting you need to run the test at the same distance, weather, crew experience level etc. you actually make an assumption that this list of factors is exhaustive. And in real life it may well not be so and the next thing - you read an article about someone who lost hundreds of millions of dollars on careless approach to statistics 

2 hours ago, Rinaldi said:

...you're suggesting that the game really doesn't have many variables and ultimately if x then y in all instances

No. From what I've seen it's multi-step calculation though each step is an obvious function with randomization. It's not a straightforward: "Take experience, weather, distance etc. feed it into one function, here you go - spotting or no spotting..." You can crudely compare it to a neural network with very simple function nodes. As in any neural network the end result taken per se is hard to interpret - here come these heated discussions  But as the nodes are simple functions - if you take a case apart it becomes cristal clear. Disclaimer: I tested what I tested - see above...

2 hours ago, Rinaldi said:

I wonder if I did the RPG test 200 times in a row precisely as you asked, would the result actually model your hypothesis? Perhaps when I'm home I'll indulge you...

Indulge me with CMSF circa 2011  Just joking. Actually my feeling is the whole approach hasn't changed (luckily because it's quite logical ). I did some limited testing on CMBS - I see what looked to me like behaviour patterns similar to CMSF 2011 just at a well more refined and complex level. I've never done an exhaustive testing on CMBS similar to CMSF 2011. The objective of that old one was to understand how much and what kind of suppression I need to throw before I rush a team against a building in a urban terrain. So in the end I had a model where you can simply input your teams, target building parameters and the distance to it, nearby building parameters, probable enemy teams and probability of their presence in buildings and it would give you a distribution of loss probabilities for the attacking team depending on the amount of fire time for each suppressing team. But that's kind of killed the fun

2 hours ago, Rinaldi said:

ordinance effectiveness versus spotting effectiveness are very different beasts; the latter has many more variables

Spotting back then was quite simple and logical. Perceived complexity in spotting comes from the fact that many nodes generate binary results - so when you look from outside the end result seems incomprehensible. It's not - it's just not a straightforward single function calculation and you have a certain in-node randomization. So what you might see are several peaks/plateaus in the end results depending on how these binary nodes perform.

Edited March 21, 2017 by IMHO

[Sgt.Squarehead]

13 minutes ago, IMHO said:

So in the end I had a model where you can simply input your teams, target building parameters and the distance to it, nearby building parameters, probable enemy teams and probability of their presence in buildings and it would give you a distribution of loss probabilities for the attacking team depending on the amount of fire time for each suppressing team. But that's kind of killed the fun

Can't argue with that. 

Edited March 21, 2017 by Sgt.Squarehead

[IMHO]

14 minutes ago, Sgt.Squarehead said:

Can't argue with that. 

CMSF=Combat Mission Science Fiction?

[Sgt.Squarehead]

Combat Mission: Statistical Furore? 

or

Combat Mission: Statistician's Fighting! 

Edited March 21, 2017 by Sgt.Squarehead

[IMHO]

Combat Mission Six-dimensional (w)Formhole to the enemy occupied building?

PS @Sgt.Squarehead, Combat Mission Statisticians-style Fighting  Like in Wu-Shu

Edited March 21, 2017 by IMHO

[Sgt.Squarehead]

Combat Missions.....Sounds Fun!  

Let's find out.....Off to play some! 

[IMHO]

@Sgt.Squarehead, e.g. for CMBS I have a load-out model for the Russian seven-team infantry squad  Number of enemy teams vs. total building size facing the probable route to clear up a town vs. the total length of the open terrain to run vs. distance distribution to the buildings you need to suppress (affects GL and RPG grenades flight time - they have different velocities so it affects the practical rate of fire and ammo use) -> gives you suppression time with "Target light" vs. "Target" with heavy weapons when you detect an enemy and you switch into full-blast, then the strategy you use - two full team suppression plus one-team attack vs. three scout teams split vs. three assault teams split  Actually the only discernible unpredictability is TacAI now sometimes makes soldiers in suppression team run from a window to window so it affects the time each soldier spends firing so affects the optimal load-out  

Edited March 21, 2017 by IMHO

[Vanir Ausf B]

9 hours ago, IMHO said:

@c3k,

1. Why 200 tests and not 20 or 2000? I mean you know the variance and distribution rule?  Actually some years ago I ran hundreds of tests of one of the old versions of CMSF engine. I was interested in spotting/accuracy/fire effects for infantry. The variance is actually quite low - the set of outcomes is predefined and quite limited. Certainly many things could have changed since then.

The variance in spotting is high, particularly at long ranges. For example, in the WW2 CM2 games the average time for a Panther A (mid) to spot a Pz IV at 1200 meters is 125 seconds (n=300, 95% confidence interval: 115.5 thru 138.9) with median of 107 and a standard deviation of 103.

EDITED to add: Dug up some old Black Sea data. BMP-3 spotting M1 Abrams at 700 meters, n=220

Mean = 67.5 seconds

95% confidence interval for Mean: 59.11 thru 75.87

Standard Deviation = 59.0

Hi = 326 Low = 1

Median = 49

Average Absolute Deviation from Median = 41.1

Edited March 21, 2017 by Vanir Ausf B

[Sgt.Squarehead]

2 hours ago, IMHO said:

@Sgt.Squarehead, e.g. for CMBS I have a load-out model for the Russian seven-team infantry squad  Number of enemy teams vs. total building size facing the probable route to clear up a town vs. the total length of the open terrain to run vs. distance distribution to the buildings you need to suppress (affects GL and RPG grenades flight time - they have different velocities so it affects the practical rate of fire and ammo use) -> gives you suppression time with "Target light" vs. "Target" with heavy weapons when you detect an enemy and you switch into full-blast, then the strategy you use - two full team suppression plus one-team attack vs. three scout teams split vs. three assault teams split  Actually the only discernible unpredictability is TacAI now sometimes makes soldiers in suppression team run from a window to window so it affects the time each soldier spends firing so affects the optimal load-out  

I'd probably just use a tank. 

23 minutes ago, Vanir Ausf B said:

The variance in spotting is high, particularly at long ranges. For example, in the WW2 CM2 games the average time for a Panther A (mid) to spot a Pz IV at 1200 meters is 125 seconds (n=300, 95% confidence interval: 115.5 thru 138.9) with median of 107 and a standard deviation of 103.

EDITED to add: Dug up some old Black Sea data. BMP-3 spotting M1 Abrams at 700 meters, n=220

Mean = 67.5 seconds

95% confidence interval for Mean: 59.11 thru 75.87

Standard Deviation = 59.0

Hi = 326 Low = 1

Median = 49

Average Absolute Deviation from Median = 41.1

Sometimes it's possible to lose sight of the obvious in a sea of numbers.....The problem demonstrated in my test is what I'd call the 'WTF? Factor'.

The higher the 'WTF? Factor' of a given incident is, the more likely an inquisitive player like myself is to question it and possibly to draw the wrong conclusions.  Given the political rows that can break out over almost nothing in this forum, the last thing we need is mistaken conclusions leading to accusations of bias (this is precisely why I think the admin team should stand above the factional ranting that sometimes takes place here).

Edited March 21, 2017 by Sgt.Squarehead

[IMHO]

@Vanir Ausf B,

I'll try to remember 2011  I meant variance in the each function node results. Let's assume 2D map, a soldier has a sector of so many degrees and he spends so much time scanning this sector. "Scanning" outcome could be no spotting, full spotting or semi-spotting and the latter two are grouped as "positive". If I remember correctly the probability of EITHER of the two positive outcomes for this sector scanning timeslot was quite stable. You get a large variance in the end time to spot finally and reliably because if one sector scan does not result in spotting (and it's randomized) then you'll need to wait for the next sector scanning or even the one after that etc. "Sector", "timeslot" etc. are certainly terms related to exact implementation so that's a kind of speculation for the sake of explanation  The actual implementation method may well be far away.

What I haven't researched much was full-spotting vs. semi-spotting and I never digged deep on how an angle to the target affects the probability. I had a crude proxy for the latter and I say crude as you have many soldiers in the squad so this smoothes out the effect of where exactly this particular guy is looking at this very moment. And the former one seemed to be a more complex function but luckily it was not so important for the task of a building assault optimization. You'll never be wrong with blasting with all force the moment you get a semi-spot on an OPFOR infantry lurking in the building.

I never tried it with vehicles.

Edited March 22, 2017 by IMHO

[DreDay]

I have actually done extensive testing on RPG-7 and RPO-M behavior back in a day. Here is the gist of my findings - PG-7VL will always miss its targets past 250 meters. It absolutely does not matter how experienced/unsuppressed  your operators are and how good good the weather is. In real life, it is capable of being effective at twice that range under perfect conditions.

RPO-M will always miss a small house at distances over 200m, again its range far surpassed that in real life when dealing with area targets. When it does manage to hit a small house with an infantry squad inside (from point-blank range of course) it fails to inflict any casualties or even heavy suppression 90% of the time. It is also completely incapable of penetrating light armor of BTRs and MTLBs; which makes it completely useless in this game.

Edited March 22, 2017 by DreDay

[IMHO]

And what were the targets for RPG? I have two different setups where a seven men veteran/crack RUS infantry squad inside a building fire at another building at 289m. I haven't calculated exact hit ratio but my feeling is they are on target with the correct floor with a 30-40% probability. It's a daytime, dry weather and gentle dead wind.

Edited March 22, 2017 by IMHO

[DreDay]

5 minutes ago, IMHO said:

And what were the targets for RPG? I have two different setups where a seven men veteran/crack RUS infantry squad inside a building fire at another building at 289m. I haven't calculated exact hit ratio but my feeling is they are on target with the correct floor with a 30-40% probability.

I was aiming for vehicular targets, obviously it can hit a broad side of a barn at longer ranges than the ones that I had listed

[IMHO]

OIC, yes my buildings are pretty wide. I assume building vs. vehicle may affect targeting logic as well.

[DreDay]

21 minutes ago, IMHO said:

OIC, yes my buildings are pretty wide. I assume building vs. vehicle may affect targeting logic as well.

I would love to be mistaken here, but I don't think that there is much of a "targeting logic" involved; just a basic dispersion coefficient...

[HerrTom]

I did my own tests on spotting, which I admit I don't have a ton of data, but as Vanir said, there is a metric ton of variation.  The M1A2 definitely spots faster, but there were a couple of cases where it took significantly longer.  I haven't filtered out the outliers from this data set if there are any, but there doesn't appear to be a huge difference between the two tanks...

Who spots first shoots first though, and the M1 tended to take about 2 seconds to aim and fire compared to the T-72's 4 seconds.  Anyway, read into it what you will...

Map looks like this - flat terrain

Both tanks were Regular, Normal, +0, etc... middle of the line.

[HerrTom]

I just noticed that the standard deviations were practically the same between the two vehicles in all cases.  Perhaps it would indicate that the spotting follows a normal distribution that's shifted by factors left or right?  This is just pure speculation, I don't have nearly enough data to say anything definitive, and it's too late for me to really put brain cells to work!

Edit: And for those unfamiliar with statistics - the standard error on the bottom line of the table is the "standard deviation of the mean" - a simpler way of saying that, in theory, if you took this test again, the standard deviation of the calculated means of any identical test you run will likely be that value.

Edited March 22, 2017 by HerrTom

[c3k]

HerrTom,

It looks like a single tank and a single target. Cool.

Of course, to have fun and test the digital targeting links, it'd be great to run that test a LOT more. Then, introduce a platoon of tanks trying to spot the same, single truck, target. You know, because we're curious.

Ken

[DMS]

Large st. deviataion explains that stories about T-72, that coudn't see Abrams standing at distance 900 metres. It seems that M1 are better because of communications, 1 tank of 4 spots and give info to others, while T-72 will share info much slower.

[IMHO]

@HerrTom, if CMBS algorithm is eyes, sectors, time-per-sector there's little point in looking at overall scatter - it will not reveal internal mechanics. The meaningful thing is a quality of clusterization achievable over the dataset. But for full-crewed tanks you'll need hundreds of tests at least, I bet  The funny thing that I'd pay attention to - two outliers for the time between spot and contact for T-72

Edited March 22, 2017 by IMHO

[panzersaurkrautwerfer]

On 3/20/2017 at 3:50 PM, Sgt.Squarehead said:

If we are going to deal in facts and accuracy, this one should definitely be taken into consideration:

http://www.popularmechanics.com/military/a20917/us-crews-fail-to-place-in-nato-tank-competition/

US crews aren't in fact even third rate.....That honour goes to the Poles. 

I've got almost no time to write so I'm turning and burning on this one:

That whole competition needs to be ignored when speaking of tank skill levels for US forces.  Basically every other competitor had months to practice and prepare for the event and to send picked crews, while the US team was basically from the rotational unit with a few weeks notice.  It's like if your local college suddenly had to field an Olympic level team the week prior to the opening ceremonies, it's your pretty darn good vs the best who've been practicing.

As to the rest of the thread, this game tries to simulate a lot of soft advantages or variables using hard numbers.  In that regard, a US tank with a crew with on average several years more experience than their Russian counterparts, using more, and better sensors, with more robust communications should virtually always spot and shoot first.  However what CMBS does because it's not reality is somewhat randomizes those results to add in all the various frictions of the battlefield.  In that it will generate unrealistic (or at least unrealistic looking outcomes!) because it "rolls" the low numbers for the US tank, while rolling the high numbers for the Russians.  This spread will always affect the US platforms less because they rightly have a superior spotting-engagement time, so even when they roll poorly, the Russian's performance has to be above average to truly exploit that.

It's certainly not fair, but it's about as a reasonable approximation of engagement times as I've ever seen achieved in what amounts to a game.

As far as Abrams 3.5, every nation in the game got gear that still hasn't been fielded.  Looking at the US stuff, it's mostly kit that's either just around the corner (AMP) or a pretty short stretch (APS), vs entirely fictional Russian hardware, and Ukrainian stuff that exists in small numbers.

I'd like a "reality" DLC that basically offers TOEs that match where the various factions are at right now though.  Would be fun given how much of a non-thing APS is in 2017, and having the "real" M1A2 SEP v2 and the less scifi Russian tank line up would be cool too.  

[Artkin]

1 hour ago, panzersaurkrautwerfer said:

having the "real" M1A2 SEP v2 and the less scifi Russian tank line up would be cool too.  

The T90M isn't too sci fi for us, right!? 

[shift8]

The in game tank is a SEPv2. Don't understand where this 3.5 stuff is coming from. Minus the APS and ERA, which dont mean much in the context of a tank v tank debate. 

Edit: Re-read post, saw that you appeared to be implying what I just stated. So never mind. 

Edited March 22, 2017 by shift8

[Sorrow_Knight]

5 hours ago, panzersaurkrautwerfer said:

 vs entirely fictional Russian hardware, and Ukrainian stuff that exists in small numbers.
 

And what Russian hardware in game you think so "entirely fictional"? 

Ukrainian army at moment almost hasn`t anything really numerous, so some "small numbers" hardware may be in the field.

[Vanir Ausf B]

4 hours ago, shift8 said:

The in game tank is a SEPv2. Don't understand where this 3.5 stuff is coming from. Minus the APS and ERA, which dont mean much in the context of a tank v tank debate. 

Plus laser warning receiver and AMP round. The in-game Abrams is functionally an SEP v4 minus the Gen III FLIR and maybe the upgraded armor. It's scheduled for 2021 or thereabouts 

http://www.armyrecognition.com/february_2017_global_defense_security_army_news_industry/u.s._army_will_begin_the_development_of_m1a2_abrams_sep_v4_main_battle_tank_11302171.html

I don't have an issue with it being in the game, I just wish the real SEP v2 was also present.

 

Edited March 22, 2017 by Vanir Ausf B