Printable View
Weapons are a bit different than nanos, one because AAO/AAD effects weapons (it doesn't effect nanos), and two most weapons are 100% def checks.Quote:
Originally Posted by Lheann
That being said, it's something like....
WeaponSkill1*(WeaponSkillCheck1/100)+WeaponSkill2*(WeaponSkillCheck2/100)+AAO
(Yes, my calculator has a % check for AAO/AAD, 'cause I'm lazy. :p )
Nano skills are similar to that.
(If you're an uber haxer you can find the formula's on my calculator page. ;) )
Well, my point was to make the formula really really obfuscated! I forgot the ;)Quote:
Originally Posted by Ebag333
Back on topic - did you observe any critical hit variations due to huge disparities in AR and DEF for weapons? It's a popular myth, but not busted so far. I do know that negative AAD+evades results in perma crit, though my testing has not been conclusive. Something I noticed along the years is that some mobs seem to have varying evade/dodge/duck (not the same values for all three) possibly topped off by an amount of AAD, since some monsters become perma crit with a shotgun but not for melee attacks after a certain amount of AAD debuffing.
One thing I did not collect that I should have was crit hits. :(Quote:
Originally Posted by eroz_c
My instinct from testing it is that crit hits are straight from your crit hit %, but I don't have a toon with a ton of +crit so it may have been that there was a difference, but too small to notice.
As far as mobs goes, the convention is that they have AAD but not evades. On the weapon side, we know for sure that they have only AAO and no weaponskills, so it'd be somewhat odd (though I could see FC doing it) to give them evades and AAD.
Do you remember the crit chance for the two people in your example?
I will put what I know about crit chance and diff in AR and DR. I had the fun of being part of testing on Test Server during the Evade fixes pre-SL (yes long time ago). First lets just say I died more in that testing then all my other AO time over 8 years.Quote:
Back on topic - did you observe any critical hit variations due to huge disparities in AR and DEF for weapons? It's a popular myth, but not busted so far. I do know that negative AAD+evades results in perma crit, though my testing has not been conclusive. Something I noticed along the years is that some mobs seem to have varying evade/dodge/duck (not the same values for all three) possibly topped off by an amount of AAD, since some monsters become perma crit with a shotgun but not for melee attacks after a certain amount of AAD debuffing.
It was explained that all weapons have a base crit chance. That crit mods like nanos and scopes get added to this and that forms the characters base crit chance. Then when the attack is made the attackers attack rating (wep skill +buffs+aao) is compared to the defenders defense rating (evade+buffs+aad adjusted by aggdefbar) and based on the difference the crit chance is adjusted up or down.
The devs even pointed out a great place to see this. Take any toon with 1000+ AR and attack a ql1 leet in the old backyards. Those leets have 0 AC and 0 Evades and non backyard ql1 leets do have some AC and Evades. Again not much on explaining to me why. You should crit nearly every hit. The math was not revealed at anytime.
I asked if mobs followed the same rules. So they warped my toon to some ql200 mob that one hit critted me to death. The point they said was to show it worked both ways and that mobs took the same math into account. I just think it was just there way of having some demented fun.
I asked some questions about it and this was the take away:
Weapons have a base crit chance
There was at one time a quaranteed miss %. Though this seems to have changed at some point.
There is a small % chance to not crit even if you have 1000+ AR and the target has 0 defense rating. But this is really small like 1 or 2%.
And lastly the biggest thing that got my attention was that mobs are configured not with Weapon and Evade skills but with AAO and AAD. It makes them easier to configure or so they said. I took away from that 1 AAO = 1 Attack Rating always and that 1 AAD = 1 Defense Rating always.
@Ebag333, when I asked if you had a formula for the defense rating I was asking if you had figured out how to compute the players defense rating. My current effort has resulted in the following:
Defense Rating = ( ( ( BaseEvade + Implant/Symb Buffs + Nano Buff + Perks + Procs + Evade Debuffs ) * Aggdef Modifier ) + AAD ) + AAD Debuffs
The Aggdef bar modifier appears to be 1.0 at full def and something less than 1.0 at full agg with everything in between scaled. This works with your charts so far as I read them. Have not played with the tools yet. Maybe tonight.
Great post! Thanks for sharing that.
Ah, that. I just use the number displayed in the stats window (which, yes, I realize is not always actually accurate).Quote:
Originally Posted by Lheann
If I was doing a formula from scratch, it'd be something like....
Evades = BaseEvade + Implant/Symb Buffs + NanoBuff + Perks + Procs + EvadeDebuffs
AAD = AAD + AADBuffs + AADDebuffs
TotalDefense = (Evades*DefCheck)+AAD
From there you get into other factors, such as the attackers offense (a similar number). I (ab)use the Agg/Def modifier to actually adjust the curve of the land chance, so that actually applies after I've calculated the Diff AR %.
That formula is (roughly)
Land % = <NumberBasedOnAgg/Def>(<AR Diff %>^<AnotherDifferentNumberBasedOnAgg/Def>
The second one is actually static for weapons, but variable for nanos. I may play around with it and see if I can make it static for nanos as well, but I don't think that the formula would be as accurate if I do that.
For weapons, between -100 and 100 you have essentially the same curve, just adjusted left to right. For nanos, the shape of the curve actually changes (gets steeper the more you go aggro), meaning you have to have a different exponential at the end to compensate.
Clear as mud?
I use -100 to 100 for the range on the agg/def bar. It seems to work. You could use something else though, -1 to 1 would work the same, just the formula's would be slightly different. -100 to 100 seemed to give the best granularity without making the formulas silly (like multiplying something by .0000000000002).Quote:
Originally Posted by Lheann
Thanks Ebag333.
It really is interesting to figure this stuff out. Good work on all you have done. Testing this stuff can be a pita and time sink. So I feel your pain.
Would have been very interesting if someone had done this pre-LE and seen if there was any difference in the formula...
Based on the limited testing I can find pre-LE, it doesn't seem to be much different.Quote:
Originally Posted by ithacana
The problem with using that info is most folks idea of an "accurate" test is up to 100 attacks. My idea of the minimum of an accurate test is five times that, some of my tests are fifty times that.
There has been elaborate testing before, you did not invent the wheel here.Quote:
Originally Posted by Ebag333
What do you base your minimum amount of events on for a stable result? 500 events is probably not enough to get a stable result, while 5000 might be overkill.
A thread with a lot of weapon testing: http://forums.anarchy-online.com/sho...d.php?t=528996, it's post-LE and most tests are from before the base hit/miss chance got removed (which means it is also representative for a few more years back, the hit/miss formula is unlikely to have changed after the SL turmoil was over). There has been other larger tests as well before this, but I don't remember any collection with this many different samples.
I also know there's an old thread somewhere with the results of testing the influence of AAD on resisting nanos. The end result was that aad had a very minuscule effect on resisting nanos. I can't find back a link to it unfortunately, even though I've searched several times for it already.
Nope. Though I doubt many tests had over 50,000 attacks in them. :pQuote:
Originally Posted by Josephina
Nor do I recall any tests that resulted in a calculator for hit rate. ;)
Simple testing. Run a bunch of tests with 100 attacks, look at the variance of the results. Run a bunch of tests with 250 attacks, do the same. Run a bunch of tests with 500 attacks, do the same. Etc etc until you have a reasonable margin for error.Quote:
Originally Posted by Josephina
Much of my initial testing was done with smaller groups. In other words I'd use the same exact settings (so should get the same results) but after 50-100 attacks I'd record the hits and misses, and see how different the two were. I quickly found that 100 attacks was far too inaccurate, so bumped it up. And kept going until I found that 500 was a reasonable number, though of course more was better.
I ran several tests to absurdly large as a sort of 'control', for example one set was 6,666 attacks. That test was only .03% hit rate more than a test (same base AR/Def) with only 1,372 attacks, and only 7.28% more than a test with 60 attacks.
Over my course of testing, I found that 500 attacks would give you margin of error of about 5%. Considering that my time is somewhat limited, I decided that a 5% margin of error was reasonable, and preceded from there.
That still didn't stop me from running longer tests, many of the tests I ran at the end were 1k+. :p
Yup, seen that one, and while that's a better one for the number of hits per test, it still has too little in the way of total tests.Quote:
Originally Posted by Josephina
AAD has a curious effect on nanos. From my testing of adjusting my AAD by a lot (at least a lot for me :p ), I haven't seen a difference that couldn't be explained by the margin of error.Quote:
Originally Posted by Josephina
However, if you get enough AAD it seems to break landing nanos completely. The amount of AAD required is ridiculous and I doubt that any player could achieve it (IIRC it's in the tens of thousands), so I suspect there's some sort of obscure bug in there.
It's possible that profs with high AAD perks could pop everything they have and effect nano land rate, but those are going to be very special circumstances/cases, and don't really apply as a general rule. So since even 1k AAD didn't make an appreciable difference on landing nanos, I decided to ignore it to greatly simplify the equation.
Would you mind sharing your actual test results with us?Quote:
Originally Posted by Ebag333
The thread I linked also resulted in a few approximative formulas for hitrate, anyone that wants to can make a calculator for them. Personally I would be more interested in the formulas (clearly written out) behind your calculator so others can check if you might be on to something or not. As it is you've talked a whole lot already, but you're yet to prove anything and so your calculator may just as well be bogus.
Actually, I've already shared my actual test results for weapons. Perhaps if you spent a bit more time reading what I've talked about a whole lot already, you'd see that. ;)Quote:
Originally Posted by Josephina
I also have several pages with more detail on them, but that's really all the relevant info right there.Quote:
Originally Posted by Ebag333
As for the actual formulas themselves, anyone can see them if they take the time to right on the calculator page.
So there you have it. I've actually offered up a whole lot, and I've proven a whole lot. Feel free to disagree with me if you like, but you really should have an actual reason behind your disagreement with me. I've opened up most of what I have as proof that my calculator works, what do you have to offer up that it doesn't work?
You're yet to prove anything and so your suggestion that my calculator may just as well be bogus may just as well be bogus. :p
Those are not your actual test results.
How many hits+misses were involved in each? What was the AR (or weaponskill and AAO)? What was the involved evade skill at? How much aad was there in the different tests? The only 2 variables you give or the position of the slider and the measured hit%, the rest is also relevant ...
Your data is useless for anyone that wants to attempt to think up or check a formula. Your data bears no prove whatsoever about how correct your formula is for different AR and defence ranges because you don't give any. Because you don't give any details about how many hits/misses there were, your "average hit difference" means nothing because it might be off a test of just 500 hits (which you have said yourself showed an average error of 5% compared to a limitless test. If I interpret the way you read this right, this gives a maximum inaccuracy of (2x5) +10 on scale of 100. That's one large error.).
Right now it is impossible to disprove your formulas on the basis of your tests, because you don't share the test results themselves. Being unable to disprove anything because you don't have anything to go by is not bogus. Claiming that your results are accurate while you don't show any sufficient proof, that's bogus.
And I take it if others want to check up on your formula, they'll have to distil if from the source and hope that they don't make a mistake. This works as well, but it could have been done much easier+clearer.
Actually, yes, they are.Quote:
Originally Posted by Josephina
I can add that. As a matter of fact, it's done:Quote:
Originally Posted by Josephina
http://pvp.aodb.us/ARDEF/display.php
Actually, that's not the only two variables I give.Quote:
Originally Posted by Josephina
The actual weapon skill, evade, AAO, and AAD doesn't matter. For one, weapon skill = AAO, and evades = AAD. For another, 100 AR vs 200 Def will give the same results as 500 AR vs 1000 Def.
The most important variable there is the AR Diff %, which I give the formula for. That's a simple--yet extremely important--variable that gives you the difference between the attackers rating and the defenders rating.
All the data is there for anyone who wants to replicate the tests. The only thing they need to understand is how to calculate the AR Diff %. Once they find that, it doesn't matter if they have 100 AR or 1000 AR, the results will be the same if they have the same AR Diff %.Quote:
Originally Posted by Josephina
5% is a fairly conservative number that I choose, based on the entire scope of the data.Quote:
Originally Posted by Josephina
Because I have far more data points for full def (-100), I have refined the formula quite well on that side. If you add up the total of how far I'm off, you end up only being off by -1.57%, a very low number in anyones book.
So you could say that for using the calculator for full def (which the vast majority of people play), the margin of error is less than 2%. I tend to err on the side of extreme conservatism, so I'd rather claim too high a margin of error and always fall within that, than claim a lower margin of error and occasionally fall outside it.
I've given many details about hits/misses in anecdotal comments in the posts above, including illustrating the differences between different sample sizes for the same Diff AR %. Plus I've gone ahead and added the the number of attacks, no big deal there.
It would be very simple to disprove, and it's the same way I disproved other formula's that people claimed were accurate.Quote:
Originally Posted by Josephina
- Take the formula, and calculate out a scenario.
- Replicate the scenario, and gather the actual results.
- Compare the two. Are they reasonably close?
- Repeat as needed.
I've seen some formula's that don't even hold up for a single test. I've seen a couple that are actually surprisingly accurate, but only for the middle range (where granted most players tend to be), and are exponentially not accurate the higher/lower you get.
I've even made step 1 very easy for you as I've provided a calculator, so you don't even have to do the maths yourself. All that's left up to you is to replicate it. :p
I agree there are easier ways, but not by a whole lot. For one, the formula--while as simplified as I can make it--is still very complex, and adding the formula to the page will likely just confuse people. Those who are clever enough to decipher the formula from the source will also likely be clever enough to understand how to apply it properly. All too often I've seen formula's misused, and incorrect information put out there by someone who doesn't understand how it works correctly. So consider that a minimum bar to filter for those who are likely to have a good understanding of it. ;)Quote:
Originally Posted by Josephina
The second is that I still intend to refine the formula further. Just because I have something that I'm willing to call good with a 5% margin of error doesn't mean I'd like to lower that margin of error even further. Especially for full aggro the margin of error is larger, I averaged a +4.65% compared to only -1.57% for full def. That's a bit larger than I'd like, so I'd like to see it get under 3%.
But further testing is needed, and testing is expensive time wise. And right now there's been a major run on my time, and it's of limited supply. So more testing (and more refining) is put off until time frees up a bit. (If you don't think that testing is expensive, I suggest you go out and farm 50,000+ attacks worth of data. Let me know what you think then. :p )
Such is life.
Incorrect. As Intelknight has shown on the basis of more and more accurate tests than yours, there are base values for both attack rating and defence (unvisible in game). So no, you cannot simply extrapolate results from the 100s to the 1000s. And even had Intelknight not already proven this claim of yours to be wrong, it would have been proper to first prove this claim before you start touting it around as the absolute truth.Quote:
Originally Posted by Ebag333
And on what do you base this claim? By chance you happen to be right on the AR part: there have been tests with different skill and AAO values that have shown that only the final AR matters. But you're most definitely wrong with your defence "calculation". Long ago there have been tests using ga4 tl3 and tl4 fixers to see which is most effective: evades or aad. Their conclusion was that evades are more effective than aad on full def and aad more effective on full agro. This is in line with my experiences and my own very limited testing at the time. So again: No, you cannot simply assume that 1 aad = 1 evade and then start touting it around as an absolute truth.Quote:
Originally Posted by Ebag333
I know that this is the internet, but simply saying that it is so, does not make it so. You'll have to do better than this if you want your calculator to have any credibility.
Well gee. You got me there.Quote:
Originally Posted by Josephina
Oh wait....Quote:
Originally Posted by Belith
And, no, it hasn't changed. How do I know? Because I've tested it, in multiple ways. (I've also asked. ;) )
As you can see, a very small difference in land rate between the three, despite the first test being purely AAO, the second test being primarily weapon skill based, and the third being pure weapon skill based (the third is higher, but also has a 1% higher Diff AR %, so that is to be expected).Code:Test 1:
Weapon Skill: 0
AAO: 1455
Evade: 1577
AAD: 732
Diff AR %: 38%
Land Rate: 36.51%
Test 2:
Weapon Skill: 1689
AAO: 140
Evade: 1847
AAD: 1028
Diff AR %: 38%
Land Rate: 34.59%
Test 3:
Weapon Skill: 564
AAO: 0
Evade: 722
AAD: 135
Diff AR %: 39%
Land Rate: 38.98%
And no, that's not the only case I have testing this, simply the most extreme one.
I threw test 3 in there to illustrate that it's the Diff AR % that matters, not the base numbers.
So yes, I not only tested this but revisited it multiple times throughout my testing.
Intelknight's numbers are interesting, but I suspect a problem with them, and I'll explain why.
I specifically chose to test this with a toon with higher numbers, and there was a very important reason for this. AO has a calculation bug. It's not a big one, but it's there.
Multiple times I would start a test, and when I did I always recorded the base stat numbers (AR, weaponskills, etc). When I finished my test, before I recorded my hits/misses, I double checked my base stat numbers.
The two checks didn't always match up.
Most of the time they did. But relatively often they were off by a few points here, few points there. I never noticed them being off by more than one to three points, but I did notice a discrepancy.
This isn't an unknown issue either. I've confirmed that the bug exists with ARK (who evidentially has a way of seeing the actual base numbers, not the GUI numbers, and confirmed that they were off), and their answer is "zone" to fix it. Unfortunately, that's not a real solution when running massive amounts of tests and swapping equip like crazy. :p
I did compare the in-game stats to the calculated stats (using Auno), and again they were never more than a bit off (couple of points). So in the end I just shrugged and moved on. Even a couple dozen points isn't enough to effect a test when you're looking at 2k+ AR and Def. Lets say my total def is about 3k, and I have a 3 point error in there. That's a .1% error.
All that being said, how does that effect Intelknight's test? Well, at the most his total AR is 80 or less, and his def is under 70.
So lets say this bug strikes, and his AR is off by 3. That's a 3.75% error there. Lets say his Def is also off by three. That's a 4% error right there. Combined that's enough to throw his reported skills off by almost 8%! That's a very significant difference.
Now all that being said, there may be a base AAD (or evade) you automagically get. If it's there it's very small and would not effect anyone with a decent amount of skill. Might be a concern for the level 1 PvPers. ;)
For one, why would FC make AAO = WS, but AAD != Evade? Programatically, that simply does not make sense.Quote:
Originally Posted by Josephina
Secondly, what FC has publically stated about it agree's with this.
Thirdly, I've done enough testing to convince myself of it.
Incidentally, while I tried to go into this as unbiased and open minded as possible, I did not believe that AAD = evade, and AAO = weapon skill. My belief was that AAD = AAO, and evade = weapon skill, but they were not interchangeably equal.
So by my saying that 1 point AddAllDef = 1 point to duck/dodge/evade, I'm saying that I was wrong all the times I argued differently. Never much fun to admit you've been wrong the last couple years. I'd much rather make something up that better fit what I was saying, believe you me. :o
Yet, you can simply say that my calculator is wrong and that makes it so?Quote:
Originally Posted by Josephina
Honestly, post your full test results if you want to proof something. Once again the 3 tests you show actually prove nothing because you don't give all the data, in this case how accurate they were (and the agg/def stance, but I assume they were on full def). How many hits&misses were involved in them? And more importantly: if you want to (dis)prove the influence of one particular variable through a test, only vary that variable in the test.
To use an excellent test of Threeze as example:
The sum of evade+aad is the same each time, but the actual evade and aad vary, so does the slider position. Too bad that you can't draw any conclusions from these tests alone though, because the numbers don't work out, but still a good example of how one could attempt to (dis)prove the theory that 1 evade = 1 aad.Code:W-Skill AR Evade AAD Sum Slider #shots %hits
1763 1900 2209 1060 3269 50 2742 0,4486
1763 1900 1729 1540 3269 50 1747 0,3761
1763 1900 2209 1060 3269 88 1218 0,4376
1763 1900 1729 1540 3269 88 1215 0,4321
1763 1900 2209 1060 3269 0 1233 0,2798
1763 1900 1729 1540 3269 0 1336 0,3076
@"Yet, you can simply say that my calculator is wrong and that makes it so?"
You're dismissing several possibilities without explaining how or why. That means your calculator is standing on very shaky legs: this gives it a lack of credibility.
I still don't see any sufficient proof that 1 evade = 1 aad, nor do I see any proof for your claim that you can extrapolate data to different magnitudes of AR/defence.
This might only be limited to me, but I personally don't take anything FC says as truth without getting confirmation from reliable&independent sources or experiencing it myself first hand. And why would Funcom do as Funcom does? Admittedly I don't understand either, but seeing how they handled the rest of the game (bandaids to remedy bandaids), it wouldn't surprise me one bit if the formulas at the base of AO's combat system are hanging together with bandages as well. That's why I think it would be best to keep all reasonable options open and only dismiss them after having gathered sufficient proof for it.
On a final note: Would you mind making your posts more concise? :p You could make your point with a lot less words in a much shorter, much more pleasant to read post.
Honestly...I have. :pQuote:
Originally Posted by Josephina
All three were larger sample sizes (500+).Quote:
Originally Posted by Josephina
Why do you say that the numbers don't work out? They seem to work out just fine for me.Quote:
Originally Posted by Josephina
Incidentally, using my calculator, I get the following numbers:
Again, a 5% difference or less. The average difference is 2.53%.Code:Threeze Calculated Difference
44.49 39.45 5.04
37.61 39.45 -1.84
43.76 45.45 -1.69
43.21 45.45 -2.24
27.98 31.56 -3.58
30.76 31.56 -0.8
You're dismissing my entire calculator without giving a single example as to where it actually fails (outside of a reasonable margin of error).Quote:
Originally Posted by Josephina
That means your dissent is standing on very shaky legs: this gives it a lack of credibility.
Seriously, it my calculator is so wrong, why can't you provide a single example as to where it fails? (Reasonable example, so not 10 AR vs 1 Def. :p )
It's not hard to not see it when you cover your eyes.Quote:
Originally Posted by Josephina
How's this for short.Quote:
Originally Posted by Josephina
No matter how much proof I give you, you still won't believe me. You can't actually prove me wrong, so you're going to set up straw men arguements and simple repeat yourself over and over.
I have all the credibility I need. Feel free to disagree with me, that's the beauty of life. Until you can show me where my calculator is incorrect, you have absolutely nothing to go on.
I took the bait and filled in some numbers:
AR: 2400
Evades: 1900
AAD: 580
Full def, 1k tests.
Measured hit%: 57.4%
Calculated hit% with your calculator: 75.96%
AR: 2800
Evades: 1900
AAD: 580
Full def, 1k tests.
Measured hit%: 69.0%
Calculated hit% with your calculator: 89.99%
This is from tests from before the default hit/miss chances got removed, but they're still way off though.
You also don't need to take this personally. I would like to see one of the last great riddles of AO to be solved as well (or modeled with a very small margin of error: <1% for most AR ranges is what I'm hoping for), but I just find the way you are going about it to be very poor. You could share your test data (AR, evades, AAD remaining), but instead you chose not to share anything and you tell anyone that questions you (me so far :p) that you are right until proven wrong.
Those 3 tests that you shared held neither proof that 1 aad = 1 evade, nor that you can extrapolate the numbers. You mix everything together: "if you want to (dis)prove the influence of one particular variable through a test, only vary that variable in the test.". I also don't know if you have done more tests to check this (and I can't know whether you have or not, because you don't share anything out of yourself), but 3 tests of 500 hits is very little to base a theory on.
I've obviously been upsetting you, so I'll stop trying to drag information out of you. Good luck with your future tests and I hope your model will get more robust with time.
@"How's this for short."
Much better ty.
Outsite of a former Game Director telling us that 1 AAD = 1 Evade. Hmm.. Given the authority of the person that published that information I suspect it is correct.Quote:
Originally Posted by Josephina
You commented also about all the GA3 and GA4 testing and how it was found that evades are better at full def than AAD and that AAD is better at full agg. Finding one better at one setting vice another is not imperical evidence they are not equal. You have failed to take the entire equation into consideration. That equation is simply demonstrated by the math me and Ebag333 discussed in this very thread.
Total Defense Rating = ( (evade skill + buffs + implants + perks + procs + debuffs) * aggdef bar mod ) + ( AAD + AAD Buffs + AAD debuffs )
As you can see from the math evade skill is going to vary is effectiveness depending on your aggdef bar. It is discussed in other threads (see fixer fourms) that full def has a > 1.0 modifier and full agg has a < 1.0 modifier. The balance point were aggdef modifier is 1.0 is around 37.5% (0 to 100% scale). At that point 1 evade = 1 add as the modifer is 1.0.
Strangely the original box set from release manual even tells you that your Agg/Def bar effects the speed you attack at and how effective your evades are. Hmm........ Seems so far Ebag and I are track with how Evades and AAD work.
When Ebag comments that what matters is the difference AR %. He is already past the point of Total Defense Rating and Total Attack Calculations. You have to understand that at somepoint it boils down to Attack Rating and Defense Rating. The truth is his formula would be a very good one to use with a break point database table for hit miss calculations. Once you have the difference you look up in the break point table what the hit chance is and the server rolls a random % and if it is under then the attack is successful. The fact that his charts indicate a not perfrectly mathmatical curve would suggest a break point table where the scaling can be different along the whole of the curve.
I do not understand why you are so against his results. From what I know of Ebag and his testing methods he is very careful in what he does and his reasoning is supportable. In fact he has done a good job of supporting his reasoning here.
I further agree with Ebag that the alternate data set you suggested is effected by other bugs in the systems. I suspect that the % of error is no less than 10% and maybe as high at 15% with that data set.
If you are not aware of the calculation bug Ebag referred too then please feel free to test it. But I myself have asked why on test server and the short answer is the gui displays everything in integers and the data is stored in floats. The rounding math is setup to be in favor of the player as much as possible so that a player skill at 10.8 will be rounded up to 11 but a mob/check skill of 10.8 would be rounded down to 10. because of these rounding choices error can occur when combinations of them happen. The only fix they have is to have you zone to clear it (all skills are recomputed on zone as part of server to server validation) or remove all the rounding logic choices and go with a rounding scheme that is blind and in the end that would effect the game in ways people would not be happy with.