Combat damage

[unmerged(14603)]

I performed 3 tests Inf vs Inf SA = 2 GD = 1. All conditions remain what I have previously stated. No org damage was observed in any of the tests.

Test 1 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.850 ----- 99.925
2 ------ 99.775 ----- 99.700
3 ------ 99.625 ----- 99.550
4 ------ 99.500 ----- 99.175
5 ------ 99.325 ----- 99.100
6 ------ 99.250 ----- 98.875
7 ------ 99.175 ----- 98.650
8 ------ 98.100 ----- 98.425
9 ------ 98.950 ----- 98.200
10 ----- 98.725 ----- 98.050

Test 2 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.850 ----- 99.850
2 ------ 99.625 ----- 99.625
3 ------ 99.400 ----- 99.325
4 ------ 99.325 ----- 99.100
5 ------ 99.175 ----- 98.725
6 ------ 98.800 ----- 98.575
7 ------ 98.425 ----- 98.500
8 ------ 98.350 ----- 98.425
9 ------ 98.200 ----- 98.200
10 ----- 98.050----- 97.975

Test 3 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.925 ----- 99.925
2 ------ 99.850 ----- 99.550
3 ------ 99.775 ----- 99.400
4 ------ 99.625 ----- 99.175
5 ------ 99.550 ----- 98.950
6 ------ 99.325 ----- 98.800
7 ------ 99.025 ----- 98.575
8 ------ 98.875 ----- 98.350
9 ------ 98.575 ----- 98.125
10 ----- 98.500 ----- 97.675

Note: There was not 1 hour in the entire test with zero damage.

 

[unmerged(13914)]

Can you believe it?

Ah, I love when I can post a question in the morning and there's data when I come back! Thanks Mikel!

Okay, let's add up the damage.

Test 1 -- 1.275 attacker, 1.95 defender
Test 2 -- 1.95 attacker, 2.025 defender
Test 3 -- 1.5 attacker, 2.325 defender

Total damage 11.025 divided by 0.171 average damage (from the last test series) gives 64.5 hits.

For this many combat rounds in the first series we had 12.5 hits. This is a HUGE difference. If you expect the first attack point to perform the same as before, because it still has a defense point to contend with, then the second attack point is inflicting 52 hits in 60 rounds, or 0.867 probability of hit.

And that also means one defense point gives (0.867-0.208) = 0.659 or about 65.9 % chance of stopping one attack point.

Yikes, are we back to that? Could the original manual text referring to a 2/3rds chance of stopping a hit actually be right? I'm really amazed by this outcome but it would explain a whole lot.

1) In each attack round only ONE defense point gets a chance to stop an attack (see my comments about this in the other thread).

2) The one defense point has a 66.67 % chance of succeeding.

Result: EXACTLY what Mikel's formula says. Not an approximation, that really is the rule.

 

[unmerged(13914)]

Some more on this model.

I have some hypotheses that should also be true, IF the points in my last post are true.

3) When a unit is outnumbered, extra defense DOES matter. For example, Mikel did a trial with two units each of 30 SA and varying GD. In that trial the extra GD was of no use. But I suspect in a battle between two 30/60 units and one 30/60 unit, the outnumbered unit will use all 60 points of GD.

4) When effectivity or strength or both are low, extra defense can also matter. For example, take a unit with 30 SA and 60 GD, with 50 % damage. The 60 GD are cut by half due to strength, so the unit only has 30 GD. This is still enough to fend off a comparable unit at full strength.

5) I may have been wrong to guess (although it was only a guess) that the +5 % bonus to ground_def_eff granted by tech has the effect of multiplying by 1.05. If one point of GD stops 66.7 % of hits then the +5 % may take us to 71.7 %. Again, this needs to be tested but if it's true it means we can adjust GD effectiveness to any value we want -- even 99 %. Very exciting.

All three of these hypotheses need checking, and on that subject I have a little speech to give. Mikel, let me say again, thanks for your fantastic work in testing. I am reminded of the old rule that assuming something is always perilous. I did some rough testing for my own hypothesis, and saw that a proper, thorough test would require the painful process of saving dozens and dozens of times (once per hour) and then opening dozens and dozens of savegame files and retrieving the relevant information. I decided that my rough estimate was "good enough" for the time being, and didn't do the necessary work. You did the work, proving my estimate was wrong, and your hypothesis about what the real rule was has turned out to be really superb -- I think given this latest round of testing it's probably right on. We all owe you one. Thanks.

 

[FrEDa]

Re: Can you believe it?

Originally posted by Math Guy
Ah, I love when I can post a question in the morning and there's data when I come back! Thanks Mikel!

Okay, let's add up the damage.

Test 1 -- 1.275 attacker, 1.95 defender
Test 2 -- 1.95 attacker, 2.025 defender
Test 3 -- 1.5 attacker, 2.325 defender

Total damage 11.025 divided by 0.171 average damage (from the last test series) gives 64.5 hits.

For this many combat rounds in the first series we had 12.5 hits. This is a HUGE difference. If you expect the first attack point to perform the same as before, because it still has a defense point to contend with, then the second attack point is inflicting 52 hits in 60 rounds, or 0.867 probability of hit.

And that also means one defense point gives (0.867-0.208) = 0.659 or about 65.9 % chance of stopping one attack point.

Aren't you confusing TO_HIT with IS_BLOCKED here? I don't know if there is a to_hit check but 52/60 (where there is no block chance) should be a rough estimate of the TO_HIT probability. Which gives :

1st A P(hit)*P(not blocked by 1 D) = 50/240

=>

1st A 52/60* not_blocked = 50/240

=> P(not blocked by 1 D) = 3000/12480 = 25/104 = 0,240384615 = 24%

So the probability of a D to block should be as high as 76% (roughly).

/F

 

[FrEDa]

Originally posted by Math Guy
Some more on this model.

I have some hypotheses that should also be true, IF the points in my last post are true.

5) I may have been wrong to guess (although it was only a guess) that the +5 % bonus to ground_def_eff granted by tech has the effect of multiplying by 1.05. If one point of GD stops 66.7 % of hits then the +5 % may take us to 71.7 %. Again, this needs to be tested but if it's true it means we can adjust GD effectiveness to any value we want -- even 99 %. Very exciting.

This would be most interesting, and could explain why the P(block) in Mikhels trials were >2/3.

Mikhel can you check if any of the sides in your trials have any eff_modifier techs?

/F

 

[unmerged(14603)]

Re: Can you believe it?

Originally posted by Math Guy
Result: EXACTLY what Mikel's formula says. Not an approximation, that really is the rule.

If true then I must have stumbled into a goldmine... better yet the sultan's harem. Should have played the lottery that day.

I also performed 3 more tests Arm vs Arm yesterday but noticed that I did not post the results as I thought I had. I must not have hit the submit button!

They were Arm vs Arm HA = 2 GD = 1. All conditions remain what I have previously stated. No org damage was observed in any of the tests.

Test 1 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.500 ----- 99.700
2 ------ 99.325 ----- 99.475
3 ------ 99.100 ----- 99.250
4 ------ 98.950 ----- 99.100
5 ------ 98.875 ----- 98.950
6 ------ 98.725 ----- 98.725
7 ------ 98.350 ----- 98.650
8 ------ 98.200 ----- 98.500
9 ------ 97.825 ----- 98.425
10 ----- 97.750 ----- 98.350

Test 2 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.150 ----- 99.850
2 ------ 99.475 ----- 99.550
3 ------ 99.250 ----- 99.325
4 ------ 99.175 ----- 99.175
5 ------ 99.025 ----- 99.025
6 ------ 98.875 ----- 98.725
7 ------ 98.575 ----- 98.350
8 ------ 98.425 ----- 98.200
9 ------ 98.200 ----- 98.050
10 ----- 97.975----- 97.825

Test 3 strength values:

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.850 ----- 99.775
2 ------ 99.625 ----- 99.700
3 ------ 99.475 ----- 99.400
4 ------ 99.325 ----- 99.175
5 ------ 99.250 ----- 99.025
6 ------ 99.175 ----- 98.950
7 ------ 99.100 ----- 98.725
8 ------ 98.800 ----- 98.500
9 ------ 98.575 ----- 98.425
10 ----- 98.350 ----- 98.050

Note: Damage occurred in every hour of all 3 tests..

 

[unmerged(14603)]

Originally posted by FrEDa
Mikhel can you check if any of the sides in your trials have any eff_modifier techs?/F

Effectivity was held constant at 100% for both sides. Ground_defense_eff set to 0.000 for both sides.

 

[unmerged(14603)]

Originally posted by Math Guy
5) I may have been wrong to guess (although it was only a guess) that the +5 % bonus to ground_def_eff granted by tech has the effect of multiplying by 1.05. If one point of GD stops 66.7 % of hits then the +5 % may take us to 71.7 %. Again, this needs to be tested but if it's true it means we can adjust GD effectiveness to any value we want -- even 99 %. Very exciting.

I think we have the problem on the run.

I have tried to test your idea. The first set of tests have Inf vs Inf SA = 1 GD = 1. I set the ground_def_eff to 33.333 for the attacking Inf.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 100.000 ----- 100.000
2 ------ 99.775 ----- 100.000
3 ------ 99.775 ----- 100.000
4 ------ 99.775 ----- 100.000
5 ------ 99.550 ----- 99.850
6 ------ 99.400 ----- 100.000
7 ------ 99.700 ----- 100.000
8 ------ 99.725 ----- 100.000
9 ------ 99.775 ----- 100.000
10 ----- 99.475 ----- 99.925

The attacker (with ground_def_eff = 33.333) received damage similar to older tests when ground_def_eff = 0 for both sides.
However, the defender (with ground_def_eff = 0.000) suddenly received significantly less damage. Could ground_def_eff be helping the blocking ability of the opponent?

So I ran another series of tests. This time the attacker had ground_def_eff = 0 while the defender's = 33.334. Notice I used 33.334 instead of 33.333.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 100.000 ----- 99.850
2 ------ 100.000 ----- 99.850
3 ------ 100.000 ----- 99.550
4 ------ 100.000 ----- 99.775
5 ------ 100.000 ----- 99.250
6 ------ 100.000 ----- 99.775
7 ------ 100.000 ----- 99.475
8 ------ 100.000 ----- 99.850
9 ------ 100.000 ----- 99.550
10 ----- 100.000 ----- 99.850

The ground_def_eff modifier is definitely working for the opponent. Also the attacker received no damage in all 10 tests. Changing the modifier from 33.333 to 33.334 may have caused blocks to = 100%. Math Guy it seems that you were correct!

In the final test I let the game run beyond 10 hours. On the first night time hour (where effectivity drops from 100% to 60%) the attacker immediately took his first hit (strength went from 100.000 to 99.850). Looks like the 100% block was reduced by the 60% effectivity.

 

[unmerged(14603)]

BTW if I'm right and gound_def_eff increases help the opponent then:

Tech 1205 Basic Div Sig Command System +5
Tech 1501 Improved Div Sig Command System +5
Tech 1802 Advanced Div Sig Command System +5

are all causing perverse results.

Traitors in the Signal Corps!

 

[unmerged(14603)]

I decided to test the idea that we could block 100% of hits by manipulating ground_def_eff.

The following tests are Inf vs Inf SA = 30 GD = 30. I set the ground_def_eff to 33.334 for the defending Inf.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 99.400 ----- 91.675
2 ------ 99.700 ----- 90.925
3 ------ 99.325 ----- 91.975
4 ------ 99.775 ----- 91.975
5 ------ 99.925 ----- 88.975
6 ------ 99.625 ----- 91.375
7 ------ 99.700 ----- 88.450
8 ------ 99.500 ----- 89.575
9 ------ 98.950 ----- 90.550
10 ----- 99.250 ----- 90.550

As you can see although the modifier substantially reduced damage (for the opponent) damage was never 0 so 100% blocks was not achieved.

 

[unmerged(13914)]

Reviewing the math -- more mysteries

First let's look at the modifications to ground_def_eff. These are truly perverse. I can't begin to explain why increasing GD effectiveness should benefit the other side. (Although I do remember a while back in another set of tests on another thread, that I got confused about which side Mikel meant by "attacker" . . . any chance that might be happening here?)

What is really striking is the result Mikel mentions when it's nighttime -- low effectivity should round down GD to zero, IF rounding down works. But see below for some ???? about that.

Now on the point FrEDa made about my computation of 65.9 %. I agree I kind of jumped over several steps here.

First, I didn't think about one implication of Mikel's data. He shows hits in hour 0 in his original data series. That means that we're not really dealing with 10-hour trials, but rather 11-hour trials. So now our base is 264 hours, not 240. However, in his new series he shows no hits in hour 0 even though the probability of hits is much higher overall. So I think his later series are still based on 10 hours.

So let's recompute the original 1/1 or 1/10 hit probability -- 50/264 = 18.9 %

Now when we come to figure out how much we've gained by having a second attack point, we need to adjust all our numbers.

Let's start with Mikel's latest test data:

Total damage in 6 runs: 11.7 % / 0.171 = 68.4 hits in TEN hours.

Subtract 12.5 * (60/66) = 11.4 hits, that leaves 57 hits.

Averaging with the other data (54.5 - 11.4 = 53.1) we get 55.05/60 = 91.75 %

So now the estimated block probability would be (1 - 0.189/0.9175) or 79.4 %. As you observed, FrEDa, my estimate yesterday wasn't done correctly, assuming it really is a two-stage process. But of course if the block probability isn't 66.67 %, then what the heck is causing the 100 % block that Mikel later got?

However, there's an interesting problem here that we need to solve before we can get much further. When strength falls below 100 % for a unit with SA = 2, we should get rounded down. So the number of hits should fall dramatically once we have damaged units, IF the rounding down rule is actually in effect.

The number of hits in the very first hour of the original tests was 6, and in the first three hours, 23. There were 12 tests and 2 opponents for each, total 24, so probability of a hit in the first hour was 25 % and in the first three hours, 31.9 %. Meanwhile in the last three hours of the original tests there were 15 hits and in the last hour, 5, for 20.8 %. That seems like a significant finding.

For the later runs of 2 SA/1 GD we have to estimate hits from damage. In the first hour, 2.85 % / 0.171 = 16.67 hits. In the first three hours, 7.125 % / 0.171 = 41.67 hits. In the final three hours, 6.825 % /0.171 = 39.9 hits, and in the final hour, 2.475 % / 0.171 = 14.47 hits.

So there is some evidence of declining hits in the later hours, but perversely the evidence is much better where the strength is 1 (and presumably rounded up) and it's much less convincing where the strength is 2 (and presumably rounded down). These numbers aren't consistent with any scheme of rounding down that I can think of. They might just be random chance, but 23 hits as opposed to 15 is a pretty big difference.

So is there no rounding down, or what? Yet ANOTHER thing I'm wrong about? More mysteries.

Also, some interesting clues here about the GD effectiveness variable. When Mikel had GD effectiveness at 99.999 %, there were two hits that got through in 100 hours of combat. That's ridiculously high if the value of GD effectiveness is carried over to all those decimal places, but maybe it's not. If it was rounded down to 99, there'd be a 1 % chance of getting through and then 2 hits in 100 hours would be very reasonable. The GD effectiveness variable is stored with two digits before the decimal place so maybe it gets rounded down to those two digits.

So Mikel, maybe try adding GD = 28.000 and running again, ten tests of 10 hours each. You should get 5 hits in 100 hours. Then try 27.999 and see how big a difference that makes. Even though we can't make sense of why this benefits the opponent, we can at least determine if there is rounding off. I'm suggesting these values instead of 33.000 and 32.999, just because the expected number of hits (1) if you use 33.000 is kind of low. It'll be easier to have confidence in the results if you use 28.000.

We could also look at units with SA = 1 and GD = 0, to see if (as FrEDa observed), HOI will allow GD to remain zero or will round it up. I'm betting it will allow it to be zero, and we should see the "true" hit rate in that trial.

 

[unmerged(14603)]

### ERROR ALERT ###

Originally posted by Math Guy
First, I didn't think about one implication of Mikel's data. He shows hits in hour 0 in his original data series.

BIG typing mistake on my part. In some series I put hour 1 into hour 0, hour 2 into hour 1 etc. Those series were easy to pick out because hour 10 always = hour 9 ( hour 9 had hour 10's results). All tests were 10 hours.

Hope those errors haven't caused too much trouble.

I've will edit the posts to correct the typing errors.

 

[unmerged(14603)]

Re: Reviewing the math -- more mysteries

Originally posted by Math Guy
First let's look at the modifications to ground_def_eff. These are truly perverse. I can't begin to explain why increasing GD effectiveness should benefit the other side. (Although I do remember a while back in another set of tests on another thread, that I got confused about which side Mikel meant by "attacker" . . . any chance that might be happening here?)

I am using the term "attacker" to mean the unit that entered the province already occupied by the "defender" causing combat to take place.

I've triple-checked the savegames and the results are as I explained. When I raise the ground_def_eff modifier of a unit, that unit's damage remains the same but its opponent's damage declines.

The attacker's opponent got the benefit in the first set of tests after I had increased the attacker's modifier. In the second set of tests the defender's opponent got the benefit after I had increased the defender's modifier.

So Mikel, maybe try adding GD = 28.000 and running again, ten tests of 10 hours each. You should get 5 hits in 100 hours. Then try 27.999 and see how big a difference that makes. Even though we can't make sense of why this benefits the opponent, we can at least determine if there is rounding off. I'm suggesting these values instead of 33.000 and 32.999, just because the expected number of hits (1) if you use 33.000 is kind of low. It'll be easier to have confidence in the results if you use 28.000. [/B]

I will run tonight. I'm assuming my error hasn't made this a mute point.

We could also look at units with SA = 1 and GD = 0, to see if (as FrEDa observed), HOI will allow GD to remain zero or will round it up. I'm betting it will allow it to be zero, and we should see the "true" hit rate in that trial. [/B]

I will also run this tonight.

 

[FrEDa]

Originally posted by mikel
I decided to test the idea that we could block 100% of hits by manipulating ground_def_eff.

The following tests are Inf vs Inf SA = 30 GD = 30. I set the ground_def_eff to 33.334 for the defending Inf.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 99.400 ----- 91.675
2 ------ 99.700 ----- 90.925
3 ------ 99.325 ----- 91.975
4 ------ 99.775 ----- 91.975
5 ------ 99.925 ----- 88.975
6 ------ 99.625 ----- 91.375
7 ------ 99.700 ----- 88.450
8 ------ 99.500 ----- 89.575
9 ------ 98.950 ----- 90.550
10 ----- 99.250 ----- 90.550

As you can see although the modifier substantially reduced damage (for the opponent) damage was never 0 so 100% blocks was not achieved.

I think we should get this to the attention of some Moderators, who can take it to the programmers. Surely this can't be WAD..

Great work, I don't know how you find the time for all those tests but we all thank you for it!

/F

 

[unmerged(14603)]

I've run the suggested tests.

The first test was Inf vs Inf SA = 1 GD = 0. I set the ground_def_eff to 0.000 for all units. Effectivities remained constant at 100% for the entire 10 hours.

Test 1 strength values. Note these are hourly values.

Hour -- Attacker ---- Defender
0 ------100.000 ----100.000
1 ------ 99.850 ----- 99.925
2 ------ 99.625 ----- 99.775
3 ------ 99.475 ----- 99.700
4 ------ 99.325 ----- 99.475
5 ------ 99.250 ----- 99.325
6 ------ 99.175 ----- 99.175
7 ------ 99.025 ----- 99.100
8 ------ 98.950 ----- 99.025
9 ------ 98.800 ----- 98.950
10 ----- 98.575 ----- 98.725

Damage occurred every hour for both units. The distribution of hourly damage was: .075 = 8, .150 = 8, .225 = 4, definitely skewed to the lower values.

I then ran a series of 10 tests using the same factors. Data was recorded after 10 hours.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 98.275 ----- 98.650
2 ------ 98.275 ----- 98.275
3 ------ 98.575 ----- 98.500
4 ------ 98.275 ----- 98.200
5 ------ 98.425 ----- 98.575
6 ------ 98.575 ----- 98.575
7 ------ 98.125 ----- 98.575
8 ------ 98.050 ----- 98.500
9 ------ 98.200 ----- 98.275
10 ----- 98.875 ----- 98.350

The average damage after 10 hours was 1.593, very close to one hit/hour at .150 per hit.

Apparently damage in the first test was lower than the average of the next 10 tests because of the skewed distribution of damage/hit in the first test.

I then ran another series of tests this time with Inf vs Inf SA = 1 GD = 1. I set the ground_def_eff to 28.000 for the defending unit.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 100.000 ----- 99.700
2 ------ 100.000 ----- 99.475
3 ------ 100.000 ----- 100.000
4 ------ 100.000 ----- 99.400
5 ------ 100.000 ----- 99.700
6 ------ 99.700 ------ 99.400
7 ------ 99.925 ------ 99.550
8 ------ 100.000 ----- 99.250
9 ------ 100.000 ----- 99.775
10 ----- 100.000 ----- 99.475

Again we are seeing the peverse effect of ground_def_eff. I raised the defending units modifier by 28.000 and its opponent got the advantage. However 1 shot got through in test 7 and at least 2 in test 6. So the +28.000 did not result in 100% blocks.

So I ran yet another series of tests with Inf vs Inf SA = 1 GD = 1. I set the ground_def_eff to 27.999 for the defending unit.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 100.000 ----- 99.325
2 ------ 100.000 ----- 99.550
3 ------ 100.000 ----- 99.400
4 ------ 100.000 ----- 99.925
5 ------ 100.000 ----- 99.550
6 ------ 100.000 ----- 99.700
7 ------ 100.000 ----- 99.775
8 ------ 100.000 ----- 99.475
9 ------ 100.000 ----- 99.700
10 ----- 100.000 ----- 99.925

As the Math Guy would say, YIKES!. I lowered the ground_def_eff to 27.999 to see the effect of rounding and instead I get 100% blocks.

On that happy note I will sign off.

 

[FrEDa]

Re: Reviewing the math -- more mysteries

Originally posted by Math Guy

What is really striking is the result Mikel mentions when it's nighttime -- low effectivity should round down GD to zero, IF rounding down works. But see below for some ???? about that.

Yes GD should be 0 (if round down) but there's still a possible did-we-hit probability for the attacks which could be lowered by STR and/or efficiency and/or nighttime. Which could explain why we don't get full hits.

Now on the point FrEDa made about my computation of 65.9 %. I agree I kind of jumped over several steps here.

Well we don't know if there's a did-we-hit, but I think you mentioned there was one in AA.

So now the estimated block probability would be (1 - 0.189/0.9175) or 79.4 %. As you observed, FrEDa, my estimate yesterday wasn't done correctly, assuming it really is a two-stage process. But of course if the block probability isn't 66.67 %, then what the heck is causing the 100 % block that Mikel later got?

Well, we do have to make some minor modifications, but check out Mikhels latest posts about round 0 etc.

However, there's an interesting problem here that we need to solve before we can get much further. When strength falls below 100 % for a unit with SA = 2, we should get rounded down. So the number of hits should fall dramatically once we have damaged units, IF the rounding down rule is actually in effect.

Might have a solution here, or parts of it. I observed that the STR displayed in the game is always rounded to the nearest integer (99.550 was still 100 in the game, 98.300 was presented as 98 etc).

We don't know what is used for STR efficiency, but it's likely it's the rounded value (to avoid extra decimals). It's also possible the STR efficiency rounding is done in a similar way (1 * 99 would still be 1 but 1*44 would be 0).

The only way to say for sure would be to get the ground_defence_eff to block 100% (or close.. 99% would do) and start the fighting around STR 50 and observe the results for 1/1 fighting. Doable. We know A can't be lower then 1 because otherwise we could get very long fight.. very long..

So is there no rounding down, or what? Yet ANOTHER thing I'm wrong about? More mysteries.

See above for a theory on how we could check it out.

Also, some interesting clues here about the GD effectiveness variable. When Mikel had GD effectiveness at 99.999 %, there were two hits that got through in 100 hours of combat. That's ridiculously high if the value of GD effectiveness is carried over to all those decimal places, but maybe it's not. If it was rounded down to 99, there'd be a 1 % chance of getting through and then 2 hits in 100 hours would be very reasonable. The GD effectiveness variable is stored with two digits before the decimal place so maybe it gets rounded down to those two digits.

Most likely, I doubt they intended for it to be possible to introduce undamageable units. We also know that techs only add full % so I think this is a fair assumption on our part.

We could also look at units with SA = 1 and GD = 0, to see if (as FrEDa observed), HOI will allow GD to remain zero or will round it up. I'm betting it will allow it to be zero, and we should see the "true" hit rate in that trial.

So, what do we have on the to-do wishlist?

1. Check SA=1 GD=0 for rounding and hit rate (did-we-hit).

2. Check efficiency rounding by using 1/1 units and initial STR around 50 (for normal rounding scheme).

More?

/F

 

[unmerged(13914)]

Next target: where does the transition occur?

* * * thump * * *

* * * thump * * *

This is me pounding my forehead against the monitor. Does this make any sense to anyone? Great job, Mikel, but now I'm more confused than ever. I agree with FrEDa on his two action items.

Hmmm. Okay, here's a third action item, an attempt at making sense of ground_def_eff. Maybe we're still in an overkill zone.

33.333 -- 2 hits /100
33.334 -- no hits / 100
28.000 -- no hits / 100
27.999 -- 3 hits / 100

If the base probability was 66.667, then 28 and 27 should have resulted in more hits no matter how things are rounded. If the base probability was 80, there shouldn't have been any hits anywhere. I can't explain how it could happen, but it still seems like there's some threshold that we're above, and whenever you're above it, the chance drops to almost (but not quite) nothing. We know for a fact that the system behaves normally when we add 0.000 to the base probability, so we need to probe for the threshold value where things change.

The fastest way to do this is a binary search.

Try adding 14.000. If it still results in either no hits or 2-3 hits/100, cut that in half to 7.000. If it results in a normal-looking range of hits, split the difference between 28.000 and 14.000, try 21.000. Keep splitting the resulting zone of uncertainty until we find a value of ground_def_eff that seems to be the transition point between almost zero hits and a reasonable percentage of hits. Mikel, are you up for this? Fabulous work so far, you need to let us know when you get tired of running tests.

 

[FrEDa]

Originally posted by mikel
I've run the suggested tests.

On the tests with 0 GD, good.. we can't be sure there is no rounding up for GD yet but most likely. On the other hand we can't be sure there *isn't* rounding up when the 'initial' value of GD > 0.

Would a 0/0 fight do damage? Is A always rounded up even if 0 from the start (thinking Militia and HA here).

So I ran yet another series of tests with Inf vs Inf SA = 1 GD = 1. I set the ground_def_eff to 27.999 for the defending unit.

Strength after 10 hours of combat. Note this is NOT hourly data.

Test -- Attacker ---- Defender
1 ------ 100.000 ----- 99.325
2 ------ 100.000 ----- 99.550
3 ------ 100.000 ----- 99.400
4 ------ 100.000 ----- 99.925
5 ------ 100.000 ----- 99.550
6 ------ 100.000 ----- 99.700
7 ------ 100.000 ----- 99.775
8 ------ 100.000 ----- 99.475
9 ------ 100.000 ----- 99.700
10 ----- 100.000 ----- 99.925

As the Math Guy would say, YIKES!. I lowered the ground_def_eff to 27.999 to see the effect of rounding and instead I get 100% blocks.

On that happy note I will sign off. [/B]

Well, 100% blocks in 10*10 rounds. We can't be sure it's actually 100% yet.

3 hits in 100 for the 28.000 ground_def_eff is an ok variance for a 99% block. So is 0 hits in 100 for 27.9999.

It suggests however that the standard block probability is around 72%. Or is very likely to be in that neighbourhood since we can't say for sure if 99% is the maximum and reached with 27.000 without very extensive tests and variance observing. Did that make sense?

/F

 

[FrEDa]

Re: Next target: where does the transition occur?

Originally posted by Math Guy
* * * thump * * *

* * * thump * * *

This is me pounding my forehead against the monitor. Does this make any sense to anyone? Great job, Mikel, but now I'm more confused than ever. I agree with FrEDa on his two action items.

Hmmm. Okay, here's a third action item, an attempt at making sense of ground_def_eff. Maybe we're still in an overkill zone.

33.333 -- 2 hits /100
33.334 -- no hits / 100
28.000 -- (no hits / 100) 3 hits/100
27.999 -- (3 hits / 100) no hits/100

If the base probability was 66.667, then 28 and 27 should have resulted in more hits no matter how things are rounded. If the base probability was 80, there shouldn't have been any hits anywhere. I can't explain how it could happen, but it still seems like there's some threshold that we're above, and whenever you're above it, the chance drops to almost (but not quite) nothing. We know for a fact that the system behaves normally when we add 0.000 to the base probability, so we need to probe for the threshold value where things change.

The fastest way to do this is a binary search.

Try adding 14.000. If it still results in either no hits or 2-3 hits/100, cut that in half to 7.000. If it results in a normal-looking range of hits, split the difference between 28.000 and 14.000, try 21.000. Keep splitting the resulting zone of uncertainty until we find a value of ground_def_eff that seems to be the transition point between almost zero hits and a reasonable percentage of hits. Mikel, are you up for this? Fabulous work so far, you need to let us know when you get tired of running tests.

Hehe, you managed to explain what I meant above much better. It will take alot of tests and guesses.

I was thinking of another way, to see if ground_def_eff is capped or if it's did-we-block. What will happen if a unit has 0 GD but a groun_def_eff of 50.000? Would that still block 50% or none at all? What if we raise it to 99%? 200%?

It wouldn't solve the default did-we-block value, but at least give us a hint of where the thing is capped.

Edit : Since ground_def_eff work in mysterious ways it would have to be added to the unit NOT having GD 0 unless both are GD 0 units.



/F

 

[unmerged(14603)]

Tried to see how ground_def_eff (gdeff) responded to changes in its value versus changes in attack/def values.

I wanted to avoid looking through savegames so I made the units powerful enough to view results (rounded to nearest whole number) on the combat screen.

The first line shows the strength of the unit whose opponent had gdeff = 0. The other lines show the strength of the unit whose opponent had gdeff > 0 . All numbers after 10 hours of combat. Note: 30/30 means SA=30 GD =30 etc.

gdeff -- 30/30 ------60/60 -----90/90
0 ------ 91----------- 83----------72
7 ------ 93----------- 87----------80
14----- 97----------- 95----------89
16 ---- 98----------- 96----------96
18 ---- 99----------- 98----------98
21 ---- 99----------- 99----------99
34 ---- 100--------- 99----------99
100 -- 99----------- 99----------99

No, the 100 was in reality 99.925 (I looked in the savegame) so there were no cases of 0 damage.

gdeff does not seem to reverse course after it reaches a certain level (unlike the AA variables).

gdeff has a substantial impact on damage until it reaches somewhere between 18-21. After that damage was too small for me to detect any differences using whole numbers.

gdeff seemed to improve overall blocking effectiveness in that as SA rose the benefit from gdeff (7 and 14) rose also.

I tried testing gdeff with SA = 1 GD = 1 (looking at savegames) but there were so few hits in all cases it would have taken many trials to predict any meaningful differences caused by different values of gdeff.