PC vs PC defensive parses (long!)

[Coprolith]

1. Introduction.

Parsing out defensive stats and skills is a lot harder in practice then parsing out offensive stats and skills. For parsing out offensive abilities there are a number of NPCs in the game that are rooted in place, don't fight back and are virtually unkillable. Parsing against these mobs can be safely done overnight; hit auto-attack, turn off the computer monitor, go to bed and in the morning you'll have a nice large sample size. These large sample sizes are necessary, EQ is a game of subtle increases in power and a not-so-subtle random number generator. All those subtle increases together can make a big difference, but to see the effects of a single upgrade/skill often requires a sample size of tens of thousands just to see the difference underneath the natural statistical variance. The conditions of the parse are very easy to control, so its easy to distinguish the effects of a single upgrade/skill.

When parsing out defensive abilities we don't have the luxury of fighting against mobs that don't fight back, since you obviously need to get hit by the mob. No amount of natural hp regeneration is enough to offset the damage output of blue con mobs ad infinitum. It is possible to parse out defensive skills during normal xp sessions, but this always comes at the cost of introducing additional variables and additional variation because the conditions aren't so easy to control. The best way to parse is against a single mob without ever killing it. That means constant healing is required (taking care not to drop below 20% health) while keeping agro and therefore it's a very work-intensive and tedious job. You have to be at the keyboard all the time and sacrifice time that could be spent xp'ing or raiding. It's no wonder there are far less defensive parses then offensive parses and those that do exist often suffer from bad statistics due to limited sample sizes.

A way around this is by using PC vs PC fights instead of PC vs NPC fights to parse out defensive abilities. Parsing PC vs PC instead of PC vs NPC has two distinct advantages:
1) The dps of the attacker can be controlled. For instance by swapping out some rusty weapons and removing all haste items the attacker's dps can be lowered to such a point that the natural regen of the defender is enough to counter it, which makes it possible to parse overnight.
2) The stats of both parties are known exactly, and they can be varied for more depth in the investigation.

The obvious disadvantage of PC vs PC parsing is that it isn't "real". But while the actual parsed numbers may not reflect real in-game combat situations, there's no reason to assume the mechanics behind it are very different. So we can still obtain qualitative information about the effects of the defenders AC vs the attackers ATK rating, and skills like Combat Agility and Combat Stability.

There's one obvious difference between PC damage and NPC damage tho. NPCs can hit you only for 20 different values, given by DB + X*DI. DB is is the Damage Base, DI the Damage Interval and X=1..20. Each mob has its own DB and DI values, which don't change under normal circumstances (some disciplines change them temporarily, and runes can cause hits other then the ones given by the formula as well). Increasing your AC, or debuffing the mob's ATK, will only change the average value of X but min hit and max hit (and all possible values in between) stay the same.
There is a remnant of this behaviour in PC damage. Your weapon's damage bonus takes on the role of DB, and the weapon's base damage divided by 10 is DI. On top of this damage distribution is a second distribution however, which is much broader and can take on any value between min and max hit. The max hit actually comes from this second distribution, and it changes with ATK. The above applies to mitigation of damage. If there are qualitative differences in avoidance as well remains to be seen. But because the PC's max hit changes with his ATK rating, there's no point in looking at mitigation as the ratio of average hit and max hit. Instead, we can simply look at how the average damage per hit changes with AC, ATK or AA skill.


2. Parse details.

For the PC vs PC parses i duelled my druid bot, Hollis. I would have preferred a melee class because the druid suffers from low offense and weapon caps, but the druid is the only high level char i have on my second account. I found an ideal spot in the Arena to perform the parses (impossible to get trained there, just dont duel in the PvP area :) ). Unhasted, Hollis was doing a whopping 3 dps at best, so i had no problems countering that with my natural Troll regen.

Most of the parses were done awhile back when the druid bot was level 63. Base stats of Hollis (L63 druid, attacker) are (no buffs):
- 1HS skill 160, Offense skill 195. ATK = 745.
- Weapon: Bone handled scimitar, 12/24 (no damage bonus)

Base stats of Coprolith (L65 BST, defender) are (no buffs, steins instead of weapons):
- Defense skill 240, dodge, block and riposte at the cap (no modifiers)
- 1049 AC, 137 AGI
- relevant defensive AAs: CA3, CS3, PE

I've performed 5 series of tests:

a) Mitigation/avoidance vs AC with Hollis at base stats
b) Mitigation/avoidance vs AC with Hollis' ATK rating increased. These required buffing (druid's wolf form and strength buff, beastlord Spiritual Vigor) the druid and therefore occasional keyboard intervention. Care was taken to remove all parts of the log where a buff had faded.
c) Effect of Agility, tested in 2 ways:
 c1) Encumbrance test. By moving 25 lbs. blocks of ore from a weight reducing bag to a normal bag, AGI was artificially lowered.
 c2) Mitigation/avoidance vs agility at constant AC
d) Frontal/rogue position
e) Effect of the AA skills Lightning Reflexes and Innate Defense

Series a and b are each split in two parts. Starting from Cop's base stats, AC was lowered by removing only gear with worn AC while keeping AGI constant at 137. Since i can only lower my AC by so much this way i did a second branch,where I start with a totally nekkid troll at 88 AGI, adding only gear with worn AC.

During series a, b and c Hollis was always in the Rogue position.

After series d this I took a break from parsing and did some xp'ing to get LR5 and ID5. During this time Hollis got to L65 (and got some skillups) and i got some gear upgrades on both toons, so for series e I made a new baseline parse before testing the LR and ID skill


3. Parse results and discussion.

3a. Mitigation/avoidance vs AC with Hollis at base stats (745 ATK)

test#	2	5	3	8	9	10	11	12	13
AC	1049	976	894	834	665	753	879	808	920
AGI	137	137	137	137	88	88	88	88	88
Swings	9439	5829	11817	7226	7801	6878	13255	9300	5739
Hits	4011	2464	4981	2969	3351	2912	5716	3901	2426
Misses	5428	3365	6836	4257	4450	3966	7539	5399	3313
Hit%	0.425	0.423	0.422	0.411	0.430	0.423	0.431	0.419	0.423
- error	0.010	0.013	0.009	0.012	0.011	0.012	0.009	0.010	0.013
Miss%	0.575	0.577	0.578	0.589	0.570	0.577	0.569	0.581	0.577
- error	0.010	0.013	0.009	0.012	0.011	0.012	0.009	0.010	0.013
Max hit	34	34	34	34	34	34	34	34	34
Avg dmg	8.18	8.41	10.12	11.72	19.04	15.02	10.54	12.38	9.59
- error	0.13	0.17	0.12	0.16	0.14	0.17	0.12	0.15	0.17

3b. Mitigation/avoidance vs AC with Hollis' ATK rating increased (879 ATK)

test#	14	15	16	17	18	19	20	21
AC	1049	976	894	834	665	753	819	879
AGI	137	137	137	137	88	88	88	88
Swings	6532	6521	6403	6513	6726	6531	6419	6644
Hits	2718	2741	2691	2748	2821	2862	2755	2733
Misses	3814	3780	3712	3765	3905	3669	3664	3911
Hit%	0.416	0.420	0.420	0.422	0.419	0.438	0.429	0.411
- error	0.012	0.012	0.012	0.012	0.012	0.012	0.012	0.012
Miss%	0.584	0.580	0.580	0.578	0.581	0.562	0.571	0.589
- error	0.012	0.012	0.012	0.012	0.012	0.012	0.012	0.012
Max hit	43	43	43	43	43	43	43	43
Avg dmg	11.70	11.98	13.86	16.17	22.56	19.27	16.69	14.53
- error	0.19	0.19	0.19	0.20	0.17	0.19	0.19	0.20

Graphed, the mitigation data (that is, average damage per hit) from a) and b) looks like this



The first graph shows the average damage per hit as a function of the listed AC. We see that with increasing AC the average damage decreases rapidly up to an AC of about 950 (about a factor 2 going from 665AC to 950AC). Beyond that, adding AC hardly changes the average damage anymore. This is what is known as the 'AC soft cap'. Let me emphasize right away that 950 AC is the soft cap when you are duelling a L63 druid; it's been known for a long time now that the soft cap is mob-dependent. Mobs in PoP, LDoN and GoD zones for instance have been known to have a much higher soft cap. So while its not worth increasing your AC over 950 when duelling a druid, it can definately be worth it when fighting mobs.

Increasing the druid's ATK (with str and pure ATK buffs) does not change the soft cap, but it does increase the average damage per hit dramatically. I can only guess that the AC soft cap depends on either the offensive skills of the attacker or even that it is hard coded as a mob-specific property. It would be interesting to test this using a melee character instead of a druid.
This observation agrees with parses posted on the Steel Warrior's board some time ago (done against a PoFire mob iirc in the 1500-2100 range. I've lost the url but i've still got the data on my HD). Going from 1500 to 2100 AC lowered the mob's average hit from roughly 290 to 260, or ~10%. A 100ATK debuff (druid's Hand of Ro) at 2000AC lowered the mob's average hit ~260 to ~220, a decrease of ~15%.
The conclusion is that ATK debuffs are still very useful even when the MT is well over the AC soft cap. Perhaps not in fast killing xp groups, where the mobs is at half health by the time Hand of Ro finishes casting, but in tough fights it can make a big difference. Also it makes a weapon like the Shinai of the Ancients which procs a 50 ATK debuff all the more valuable.

There's a strange difference between the 745ATK data and the 879ATK data. There'a a definite 'kink' in the 745ATK data at ~800AC. Below 800AC, the avg dmg decreases very rapidly with AC and from 800 to 950AC it decreases moderately fast. At 879ATK, the kink is not present at all and the avg dmg decreases steadily until the soft cap is reached. I have absolutely no clue what could be the reason for this behaviour.

The next graph shows avoidance (= miss rate when in rogue position) as a function of listed AC.



Its hard to see any differences in avoidance underneath the statistical uncertainty of the data, even at these sample sizes. It's certain tho that worn AC does not change avoidance, and increasing ATK with strength and pure attack buffs does not change avoidance either. Off course that last part we already knew from my offensive parses ;) . Only when we average out all the data obtained at 88 AGI and all the data obtained at 137 AGI (these numbers are represented by the dotted lines in the graph) can we see a very small improvement in avoidance at 137 AGI.

If worn AC affects only mitigation and agility affects only avoidance as is generally assumed then we would expect that the 2 branches of each avg dmg vs AC curve in figure 1 do not match up perfectly (if you recall the first branch was obtained at 137agi and then removing worn AC only, the second branch starts with no gear at all at 88AGI and adding only worn AC). In the region where there is overlap between the two branches the same listed AC would actually be made up of different amounts of 'avoidance AC' and 'mitigation AC'. However in the graph we do see the branches overlapping. Does this mean that agility affects mitigation as well as avoidance? No, its too early to tell from this data alone. The 49 pt difference in AGI translates to only 14 pts of listed AC, and on the scale of the graph a 15 AC pt difference would hardly show up beneath the statistical uncertainty. A larger AGI range is necessary, and i will dig deeper into this in the next section.

3c. Effect of Agility.

The first test of agility i performed is an encumbrance test. By moving heavy block of ore from a weight reducing bag to a normal bag. As my encumbrance increased, my agility lowered and the listed AC with it. I used a run at 137 AGI as baseline, and did 2 more runs just above and below the 75 AGI breakpoint

test#	3	6	7
AC	894	876	834
AGI	137	76	73
Swings	11817	7669	17004
Hits	4981	3215	7115
Misses	6836	4454	9889
Hit%	0.422	0.419	0.418
- error	0.009	0.011	0.008
Miss%	0.578	0.581	0.582
- error	0.009	0.011	0.008
Max hit	34	34	34
Avg dmg	10.12	9.97	10.13
- error	0.12	0.15	0.10

We see the large drop in listed AC when crossing the 75AGI breakpoint. What we don't see is any effect whatsoever on the average damage per hit. The drop in listed AC is now 60 pts, and if agility did have an effect on mitigation it would have shown up. So the conlusion has to be that AGI only affects avoidance as assumed. There's also no difference in avoidance in this test, but we've already seen that it takes a far larger sample size to test the effect of AGi on avoidance with statistical certainty.

I did just that by buying some cheap AGI-gear in the Bazaar and performing a very long parse at 188AGI, and combined the results with the average of all the 88AGi data and the average of all the 137AGI data.

test#	Sum88	Sum137	22
AGI	88	137	188
Swings	69293	60280	43941
Hits	29477	25323	17862
Misses	39816	34957	26079
Hit%	0.425	0.420	0.406
- error	0.004	0.004	0.005
Miss%	0.575	0.580	0.594
- error	0.004	0.004	0.005

And here's the data graphed:



We now see a small but significant increase in avoidance as a function of agility. Linear regression yields a (3.2+/-1.0)% increase in avoidance per 100 AGI (in this range that is). That's the same kind of number you'd expect based on the equations used by Magelo, with which the effect of AGI on listed AC are calculated. Assuming therefore that these equations scale properly over the whole range of possible AGI values, there's no reason to expect the benefits of AGI to continue above 200, since you get diminishing returns to your listed AC from AGI above 200. As it is, a 3% increase in avoidance for a 100 AGI increase is a joke anyway, SOE might as well remove the stat completely and reduce the CPU load of having to calculate its effect.

3d. Frontal/Rogue position

test#

1< Elder Coprolith III Trollie ferrul lawd of 65 levels (retired) Coprolith Savage Lord Posts: 787 Location: Netherlands Logged PC vs PC defensive parses (long!) #1 April 23, 2004, 11:19:22 AM (continued from main post) 3e. Lightning Reflexes and Innate Defense As i mentioned before i took a break from parsing before doing the LR and ID parses. I made a baseline when Hollis was L65, and there were several gear upgrades on both toons. For the baseline, Cop's AC was 1059 and AGI was at 144, with CA3, CS3 and PE. Hollis' ATK rating was 763 with a 166 1HS skill and 197 Offense skill. I didn't do parses at intermediate stages, but only of LR5 and ID5 (most if not all PoP AA abilities seem to be linear in level). By the time i had LR5, Hollis default ATK had increased because of some new gear with +STR on it, but since STR does not effect avoidance this does not affect the LR5 parse. For the ID5 parse i removed the new STR gear so his ATK was back at 763. But by this time Cop had gotten a few upgrades (mainly augments) and his listed AC was 1062 at 160 AGI. Since we're well above the soft cap and AGI doesnt affect mitigation anyway, this does not affect the ID5 parse either. Druid was back in the rogue position again. Here's the table with results: test# 23 24 25 Base LR5 ID5 AC 1059 1059 1062 AGI 143 143 160 ATK(*) 763 787 763 Swings 12566 12219 15779 Hits 5305 4487 5868 Misses 7261 7732 9911 Hit% 0.422 0.367 0.372 - error 0.009 0.009 0.008 Miss% 0.578 0.633 0.628 - error 0.009 0.009 0.008 Max hit 34 35 34 Avg dmg 8.09 8.20 7.49 - error 0.11 0.12 0.10 (*) Druid's ATK rating. - LR5: A very clear increase in avoidance from (0.578+/-0.009) at LR0 to (0.633+/-0.009) at LR5, a total increase of (9.5+/-2.1)% - ID5: The average damage per hit decreases from 8.09+/-0.11 at ID0 to 7.49+/-0.10 at ID5, a decrease of 7.5+/-1.8%. However it must be noted that this percentage will be lower under normal circumstsnces. Druids don't have a damage bonus. If they had a DB of 13 just like us, the percentage would be much lower, because damage bonuses can't be mitigated. In this case the numbers would be 21.09 and 20.49, a decrease of only 3%. And while this is an extreme case, against normal NPC mobs the effectiveness of ID will be decreased also. It all depends on the ratio of their DI and DB (or the ratio of their min and max hit if you like). Against most normal NPCs, ID5 would have netted a damage reduction of about 5-6 %. (Additional note: mind you it is the ratio of DI and DB, or min and max hit, that matters. I've seen posts stating that the harder a mob hits, the more valuable avoidance becomes compared to mitigation. This is not true. Against a mob with a max hit of 1000 but a min hit of 1 you'd get the full 7.5% from ID5, but against a mob with a min hit of 100 and a max hit of 101 ID5 is useless.) 4. In conclusion. As I've said before, PC vs PC defense parsing doesn't give mitigation and avoidance numbers that are applicable to PC vs NPC combat. You're not going to get a 60% avoidance unless you're fighting greens, the number will be more along the lines of 40% or less. But for qualitative parses, that is investigating the game mechanics and determining the value of stats and skills, PC vs PC defensive parses are very useful because of their controlled circumstances and superior statisitics. i now declare this paper open for discussion. ;) /hugs Elder Coprolith III Trollie ferrul lawd of 65 levels (retired) Kitvear Ravager Posts: 103 Location: Zebuxoruk Logged Thanks #2 April 23, 2004, 12:30:34 PM That's a long read and kinda hurts my befuddled brain lol but thank you very much for quantifying the affects of agi.  I will prolly raise my agi to 200 even though it is that much of a avoidence increase. I already have LR5 and am just starting on ND & CS so this data is timely and will help me stay focused on getting ID.  Thanks and rock on! http://www.magelo.com/eq_view_profile.html?num=706403 Kitvear 66  Beastlord Vearlis 67  Chanter Zebuxoruk Valse Brute Posts: 11 Logged ... #3 April 23, 2004, 12:42:11 PM Would be interesting to see parses with items with avoidance modifers (time gear) to see how they really effect you.  Maybe parsed at 5 then 10 20 30 and so on. Valse Feral Kitty of Appotus Dominus Vaz~ Aneya Savage Lord Posts: 313 Location: Canada Logged PC vs PC defensive parses (long!) #4 April 23, 2004, 02:47:06 PM I've always just assumed that ripostes, dodges, parries and blocks are 'converted hits', that is, the hit/miss check is made first, and only if it rolls a hit the r/d/p/b checks are made (thus avoiding a lot a r/d/p/b checks and reducing the CPU load). While this would reduce CPU load I beleive it makes less sence from a simulating combat perspective. In RL one would not wait to see if an attack hit before trying to block it. By that time it would be too late. Instead one would attempt to block all incoming attacks regardles of whether they would actualy hit if let through. So it does not seem strange to me that they check r/d/p/b before calculating misses. EQ Aneya 70 Beastlord Tarew Marr EQ2 Evalin Swashbuckler Mistmoore Hrann Savage Lord Posts: 662 Logged PC vs PC defensive parses (long!) #5 April 23, 2004, 02:56:02 PM So, this pretty much verifies what we've believed all along.  Agility has a an effect, but it is tiny and hardly worth paying attention to.  LR5 is usually better than ID5. Actually, I'm a little disappointed with the performance of ID5.  I don't see why the effect shouldn't be of the same magnitude as LR5 (i.e. ~9.5% less damage taken, at least under ideal situations - fighting a mob with a large hit range). I'm pretty impressed with the effects of AC and Atk.  Do you think it would be possible to correlate the two - something like 100 atk negates the effect of 100 AC or something?  Perhaps an analsys of that kind could lead to more insight on the atk/ac softcaps? And, as usual, Bravo Coprolith! http://www.magelo.com/eq_view_profile.html?num=469790&resize=true Jkal_Shihar Arch Animist - 75 Beastlord Savage Lord Posts: 404 Logged PC vs PC defensive parses (long!) #6 April 23, 2004, 04:26:24 PM im naming my next kid after Coprolith  :D but thanks for the brain twister of a post, it was very informative. had to cut and paste onto word page so i can sit and read it more throughily though  :wink: Animist Jkal Shi`har and tigger Arch Animist of the Tribunal Server My Magelo *still my main since dec. 2001* Coprolith Savage Lord Posts: 787 Location: Netherlands Logged PC vs PC defensive parses (long!) #7 April 23, 2004, 05:14:30 PM Actually, I'm a little disappointed with the performance of ID5. I don't see why the effect shouldn't be of the same magnitude as LR5 (i.e. ~9.5% less damage taken, at least under ideal situations - fighting a mob with a large hit range). Well technically they are of the same magnitude. (9.5+/-2.1)% is no different from (7.5+/-1.8 )% :wink: They dont give actual numbers for LR and ID in the AA description, but im betting they are the same just like CA and CS. The prob is that the 7.5% mitigation number is a "best case scenario" number, which you only get when there's no BD around. In practice this means ID is 15-30% less effective then LR I'm pretty impressed with the effects of AC and Atk. Do you think it would be possible to correlate the two - something like 100 atk negates the effect of 100 AC or something? Well that's the thing, there's no direct correlation between AC and ATK, as evidenced by the fact that increasing ATK increases the average damage, but not the AC soft cap. Adding 100AC is not the same as decreasing the attackers ATK rating by 100. In fact, you can't pin a single number on it, X AC for Y ATK, since it will depend entirely on where you are in the mitigation curve, pre-soft cap or post soft cap. First you'd have to completely break down both the ATK and the AC number to its constituent elements; mitigation AC, avoidance AC, to hit ATK, damage ATK, what part of each is affected by what skill, etc. I've done some of that with my ATK analysis, i'd have to do the same for AC, and then find a way to test all the variables one by one. That's a lot of work! While this would reduce CPU load I beleive it makes less sence from a simulating combat perspective. In RL one would not wait to see if an attack hit before trying to block it. By that time it would be too late. Instead one would attempt to block all incoming attacks regardles of whether they would actualy hit if let through. So it does not seem strange to me that they check r/d/p/b before calculating misses. True, but you dont have to simulate real-life combat to get a working game model. I've parsed over 200000 swings, of which ~120000 were misses. That's 360000 r/d/b checks wasted. And speaking of realism, my trollie backside is as broad as a barn, you'd have to be blind to miss when you're in the rogue position  :D /hugs Elder Coprolith III Trollie ferrul lawd of 65 levels (retired) Hrann Savage Lord Posts: 662 Logged PC vs PC defensive parses (long!) #8 April 23, 2004, 06:38:00 PM The "kink" in the graph you mentioned (on the 745 atk line), there seems to be one on the 879 atk line as well; and both of the "kinks" seem to right around the same AC as the atk.  Or am I just making patterns where there are none?  What do you think? http://www.magelo.com/eq_view_profile.html?num=469790&resize=true Kashmiir Battlekat WoW Fanboi Savage Lord Posts: 755 Location: World of Warcraft - Hyjal Server - Name: Khrash Logged PC vs PC defensive parses (long!) #9 April 23, 2004, 06:51:40 PM Moving to the Library. VERY interesting stuff man. Leaving Shadow so other can follow link. TerjynPovar Savage Lord Posts: 512 Location: Povar Logged PC vs PC defensive parses (long!) #10 April 23, 2004, 07:12:56 PM Ok, then I absolutely have to ask. Warriors have now parsed AC up to the 2200 range or so, and have shown that there is no universal soft cap, AC of 2200 still helps with some creatures and not with others. What can determine this and still fit in with your parse?  Level difference possible changing the soft cap point?  Or is it just that it really is different with plate classes vs. other classes? Terjyn, Retired Feral Lord on the Povar Server Noriko Primalist Posts: 202 Location: Xev Logged PC vs PC defensive parses (long!) #11 April 23, 2004, 07:34:40 PM I think it was mentioned before and Cop mentioned it again in the first half, the soft cap is mob dependent.  I would imagine the only way to test this is to run the parse against differen NPCs, which is what Cop is trying to avoid. http://www.magelo.com/eq_view_profile.html?num=774007">Noriko . http://www.celestialrising.com">Celestial Rising . http://pub228.ezboard.com/bxevserverboards">Xev . http://www.beastlords.org/forums/viewtopic.php?t=1593">BST65.80/Xv h++ 3khp 3km ft13 r9 G++<3 e++ o+ T+++ L++ H+++@ f g b- t++>++++ D- !X Coprolith Savage Lord Posts: 787 Location: Netherlands Logged PC vs PC defensive parses (long!) #12 April 23, 2004, 08:07:58 PM Warriors have now parsed AC up to the 2200 range or so, and have shown that there is no universal soft cap, AC of 2200 still helps with some creatures and not with others. What can determine this and still fit in with your parse? Level difference possible changing the soft cap point? Or is it just that it really is different with plate classes vs. other classes? Well first of all the average damage vs AC graph doesnt level off completely when you're past the soft cap, but the return is not even a 1/10th of what it is before. Most notably the % all hits that are min damage hits keeps increasing, and this decreases the chance to get 'one-rounded'. Having said that there are boss mobs where even at 2000+ AC there's a notable difference in average damage per hit. Mob level will be one factor for sure in determining the soft cap but its not the only one. I've seen L50 PoP mobs with a higher soft cap then 'old world' L50 mobs. Skill, or difference between mob offensive skill and player defensive skills could be another. However skill differences can only go so far, unless NPCs can have skill values much higher then the player cap. It wouldn't surprise me if mobs have an additional parameter or property, set at creation, that also affects the soft cap for that particular mob. This parameter could then be set to a higher value for boss mobs. To get a better idea of what's underneath i'd have to repeat the entire experiment several times preferably with a melee toon as attacker. For instance by deliberately neglecting a particular weapon skill the effect of weapon skills on the soft cap could be investigated. Effect of level could be done by using for instance a lvl 40 paladin with his skills capped. They will remain capped until L51 so this gives 10 levels at constant skill to play with. Parsing against a bunch of toons of the same level but with different Offense skill cap could shed some light on the effect of that skill. Prob is: I've just listed about 6 months of nightly parsing  :roll: The "kink" in the graph you mentioned (on the 745 atk line), there seems to be one on the 879 atk line as well; and both of the "kinks" seem to right around the same AC as the atk. Or am I just making patterns where there are none? What do you think? I don't think there is a kink in the 879ATK data, Hrann. Maybe i shouldn't have put in the dotted lines in the graph, they just connect the dots and were meant as a guide to the eye and not as an interpolation in between data points. Looking back i think it might be misleading you to think there is a kink in the graph. There's no data between 894 AC and 976 AC so there's no way to tell how the transition from the linear behaviour below 895AC to the soft cap evolves. Still, i can't rule it out either. I would rule out a kink at AC=ATK behaviour on principle tho. The listed AC number is made up of both mitigation AC and avoidance AC, and listed ATK is made up of a 'to hit'-rating and a 'damage'-rating. The graph looks only at mitigation, but as a function of listed AC and listed ATK. I would be happy if it turns out the kink showed up where mitigation AC = damage ATK rating, but I'd have to delve a lot deeper into the matter to see if that is even a remote possibility. im naming my next kid after Coprolith You'd name your kid Coprolith? How cruel is that!?  :mrgreen: /hugs Elder Coprolith III Trollie ferrul lawd of 65 levels (retired) Hrann Savage Lord Posts: 662 Logged PC vs PC defensive parses (long!) #13 April 23, 2004, 08:28:22 PM OK, one last crazy question - sorry it's a Friday. How sure are you that avg damage is linear prior to the particular mob's "soft cap".  Could it be a steep curve, perhaps something like if damage were porportional to x/AC+Atk, where x is some unknown modifier that possibly contains a portion of Atk as well? Just curious.  It would be nice to think that someday a universal, covers-all-bases, formula could be discovered.  Wait, were we discussing the universe or Everquest? http://www.magelo.com/eq_view_profile.html?num=469790&resize=true Coprolith Savage Lord Posts: 787 Location: Netherlands Logged PC vs PC defensive parses (long!) #14 April 23, 2004, 08:39:26 PM Put a ruler or even just a piece of paper against the the 879ATk curve, you'll see that the average damage vs AC graph is linear all the way from 660AC up to 900AC Oh and after seeing the complexity of the ATK rating unfold before my eyes i've pretty giving up hope of finding a simple universal covers-all-bases formula. I don't think we'll have a Grand Unification Theory of Everquest soon.  :) /hugs Elder Coprolith III Trollie ferrul lawd of 65 levels (retired) Print Go Up Pages1 2 User actions The Beastlords' Den Everquest 1 Library PC vs PC defensive parses (long!) User actions Print Help | Terms and Rules | Go Up ▲ SMF 2.1.4 © 2023, Simple Machines Hextech Theme Made By : TwitchisMental

[Coprolith]

(continued from main post)

3e. Lightning Reflexes and Innate Defense

As i mentioned before i took a break from parsing before doing the LR and ID parses. I made a baseline when Hollis was L65, and there were several gear upgrades on both toons.

For the baseline, Cop's AC was 1059 and AGI was at 144, with CA3, CS3 and PE. Hollis' ATK rating was 763 with a 166 1HS skill and 197 Offense skill. I didn't do parses at intermediate stages, but only of LR5 and ID5 (most if not all PoP AA abilities seem to be linear in level). By the time i had LR5, Hollis default ATK had increased because of some new gear with +STR on it, but since STR does not effect avoidance this does not affect the LR5 parse. For the ID5 parse i removed the new STR gear so his ATK was back at 763. But by this time Cop had gotten a few upgrades (mainly augments) and his listed AC was 1062 at 160 AGI. Since we're well above the soft cap and AGI doesnt affect mitigation anyway, this does not affect the ID5 parse either.
Druid was back in the rogue position again. Here's the table with results:

test#	23	24	25
	Base	LR5	ID5
AC	1059	1059	1062
AGI	143	143	160
ATK(*)	763	787	763
Swings	12566	12219	15779
Hits	5305	4487	5868
Misses	7261	7732	9911
Hit%	0.422	0.367	0.372
- error	0.009	0.009	0.008
Miss%	0.578	0.633	0.628
- error	0.009	0.009	0.008
Max hit	34	35	34
Avg dmg	8.09	8.20	7.49
- error	0.11	0.12	0.10

(*) Druid's ATK rating.

- LR5:
A very clear increase in avoidance from (0.578+/-0.009) at LR0 to (0.633+/-0.009) at LR5, a total increase of (9.5+/-2.1)%
- ID5:
The average damage per hit decreases from 8.09+/-0.11 at ID0 to 7.49+/-0.10 at ID5, a decrease of 7.5+/-1.8%. However it must be noted that this percentage will be lower under normal circumstsnces. Druids don't have a damage bonus. If they had a DB of 13 just like us, the percentage would be much lower, because damage bonuses can't be mitigated. In this case the numbers would be 21.09 and 20.49, a decrease of only 3%. And while this is an extreme case, against normal NPC mobs the effectiveness of ID will be decreased also. It all depends on the ratio of their DI and DB (or the ratio of their min and max hit if you like). Against most normal NPCs, ID5 would have netted a damage reduction of about 5-6 %.
(Additional note: mind you it is the ratio of DI and DB, or min and max hit, that matters. I've seen posts stating that the harder a mob hits, the more valuable avoidance becomes compared to mitigation. This is not true. Against a mob with a max hit of 1000 but a min hit of 1 you'd get the full 7.5% from ID5, but against a mob with a min hit of 100 and a max hit of 101 ID5 is useless.)


4. In conclusion.

As I've said before, PC vs PC defense parsing doesn't give mitigation and avoidance numbers that are applicable to PC vs NPC combat. You're not going to get a 60% avoidance unless you're fighting greens, the number will be more along the lines of 40% or less. But for qualitative parses, that is investigating the game mechanics and determining the value of stats and skills, PC vs PC defensive parses are very useful because of their controlled circumstances and superior statisitics.

i now declare this paper open for discussion. ;)

/hugs

[Kitvear]

That's a long read and kinda hurts my befuddled brain lol but thank you very much for quantifying the affects of agi.  I will prolly raise my agi to 200 even though it is that much of a avoidence increase.

I already have LR5 and am just starting on ND & CS so this data is timely and will help me stay focused on getting ID.  Thanks and rock on!

[Valse]

Would be interesting to see parses with items with avoidance modifers (time gear) to see how they really effect you.  Maybe parsed at 5 then 10 20 30 and so on.

Valse
Feral Kitty of Appotus Dominus
Vaz~

[Aneya]

I've always just assumed that ripostes, dodges, parries and blocks are 'converted hits', that is, the hit/miss check is made first, and only if it rolls a hit the r/d/p/b checks are made (thus avoiding a lot a r/d/p/b checks and reducing the CPU load).

While this would reduce CPU load I beleive it makes less sence from a simulating combat perspective. In RL one would not wait to see if an attack hit before trying to block it. By that time it would be too late. Instead one would attempt to block all incoming attacks regardles of whether they would actualy hit if let through. So it does not seem strange to me that they check r/d/p/b before calculating misses.

[Hrann]

So, this pretty much verifies what we've believed all along.  Agility has a an effect, but it is tiny and hardly worth paying attention to.  LR5 is usually better than ID5.

Actually, I'm a little disappointed with the performance of ID5.  I don't see why the effect shouldn't be of the same magnitude as LR5 (i.e. ~9.5% less damage taken, at least under ideal situations - fighting a mob with a large hit range).

I'm pretty impressed with the effects of AC and Atk.  Do you think it would be possible to correlate the two - something like 100 atk negates the effect of 100 AC or something?  Perhaps an analsys of that kind could lead to more insight on the atk/ac softcaps?

And, as usual, Bravo Coprolith!

[Jkal_Shihar]

im naming my next kid after Coprolith  :D

but thanks for the brain twister of a post, it was very informative. had to cut and paste onto word page so i can sit and read it more throughily though  :wink:

[Coprolith]

Actually, I'm a little disappointed with the performance of ID5. I don't see why the effect shouldn't be of the same magnitude as LR5 (i.e. ~9.5% less damage taken, at least under ideal situations - fighting a mob with a large hit range).

Well technically they are of the same magnitude. (9.5+/-2.1)% is no different from (7.5+/-1.8 )% :wink: They dont give actual numbers for LR and ID in the AA description, but im betting they are the same just like CA and CS. The prob is that the 7.5% mitigation number is a "best case scenario" number, which you only get when there's no BD around. In practice this means ID is 15-30% less effective then LR

I'm pretty impressed with the effects of AC and Atk. Do you think it would be possible to correlate the two - something like 100 atk negates the effect of 100 AC or something?

Well that's the thing, there's no direct correlation between AC and ATK, as evidenced by the fact that increasing ATK increases the average damage, but not the AC soft cap. Adding 100AC is not the same as decreasing the attackers ATK rating by 100. In fact, you can't pin a single number on it, X AC for Y ATK, since it will depend entirely on where you are in the mitigation curve, pre-soft cap or post soft cap.
First you'd have to completely break down both the ATK and the AC number to its constituent elements; mitigation AC, avoidance AC, to hit ATK, damage ATK, what part of each is affected by what skill, etc. I've done some of that with my ATK analysis, i'd have to do the same for AC, and then find a way to test all the variables one by one. That's a lot of work!

While this would reduce CPU load I beleive it makes less sence from a simulating combat perspective. In RL one would not wait to see if an attack hit before trying to block it. By that time it would be too late. Instead one would attempt to block all incoming attacks regardles of whether they would actualy hit if let through. So it does not seem strange to me that they check r/d/p/b before calculating misses.

True, but you dont have to simulate real-life combat to get a working game model. I've parsed over 200000 swings, of which ~120000 were misses. That's 360000 r/d/b checks wasted. And speaking of realism, my trollie backside is as broad as a barn, you'd have to be blind to miss when you're in the rogue position  :D

/hugs

[Hrann]

The "kink" in the graph you mentioned (on the 745 atk line), there seems to be one on the 879 atk line as well; and both of the "kinks" seem to right around the same AC as the atk.  Or am I just making patterns where there are none?  What do you think?

[Kashmiir Battlekat]

Moving to the Library. VERY interesting stuff man.

Leaving Shadow so other can follow link.

[TerjynPovar]

Ok, then I absolutely have to ask.

Warriors have now parsed AC up to the 2200 range or so, and have shown that there is no universal soft cap, AC of 2200 still helps with some creatures and not with others.

What can determine this and still fit in with your parse?  Level difference possible changing the soft cap point?  Or is it just that it really is different with plate classes vs. other classes?

[Noriko]

I think it was mentioned before and Cop mentioned it again in the first half, the soft cap is mob dependent.  I would imagine the only way to test this is to run the parse against differen NPCs, which is what Cop is trying to avoid.

[Coprolith]

Warriors have now parsed AC up to the 2200 range or so, and have shown that there is no universal soft cap, AC of 2200 still helps with some creatures and not with others.

What can determine this and still fit in with your parse? Level difference possible changing the soft cap point? Or is it just that it really is different with plate classes vs. other classes?

Well first of all the average damage vs AC graph doesnt level off completely when you're past the soft cap, but the return is not even a 1/10th of what it is before. Most notably the % all hits that are min damage hits keeps increasing, and this decreases the chance to get 'one-rounded'. Having said that there are boss mobs where even at 2000+ AC there's a notable difference in average damage per hit. Mob level will be one factor for sure in determining the soft cap but its not the only one. I've seen L50 PoP mobs with a higher soft cap then 'old world' L50 mobs. Skill, or difference between mob offensive skill and player defensive skills could be another. However skill differences can only go so far, unless NPCs can have skill values much higher then the player cap. It wouldn't surprise me if mobs have an additional parameter or property, set at creation, that also affects the soft cap for that particular mob. This parameter could then be set to a higher value for boss mobs.

To get a better idea of what's underneath i'd have to repeat the entire experiment several times preferably with a melee toon as attacker. For instance by deliberately neglecting a particular weapon skill the effect of weapon skills on the soft cap could be investigated.
Effect of level could be done by using for instance a lvl 40 paladin with his skills capped. They will remain capped until L51 so this gives 10 levels at constant skill to play with. Parsing against a bunch of toons of the same level but with different Offense skill cap could shed some light on the effect of that skill.

Prob is: I've just listed about 6 months of nightly parsing  :roll:

The "kink" in the graph you mentioned (on the 745 atk line), there seems to be one on the 879 atk line as well; and both of the "kinks" seem to right around the same AC as the atk. Or am I just making patterns where there are none? What do you think?

I don't think there is a kink in the 879ATK data, Hrann. Maybe i shouldn't have put in the dotted lines in the graph, they just connect the dots and were meant as a guide to the eye and not as an interpolation in between data points. Looking back i think it might be misleading you to think there is a kink in the graph. There's no data between 894 AC and 976 AC so there's no way to tell how the transition from the linear behaviour below 895AC to the soft cap evolves. Still, i can't rule it out either.

I would rule out a kink at AC=ATK behaviour on principle tho. The listed AC number is made up of both mitigation AC and avoidance AC, and listed ATK is made up of a 'to hit'-rating and a 'damage'-rating. The graph looks only at mitigation, but as a function of listed AC and listed ATK. I would be happy if it turns out the kink showed up where mitigation AC = damage ATK rating, but I'd have to delve a lot deeper into the matter to see if that is even a remote possibility.

im naming my next kid after Coprolith

You'd name your kid Coprolith? How cruel is that!?  :mrgreen:

/hugs

[Hrann]

OK, one last crazy question - sorry it's a Friday.

How sure are you that avg damage is linear prior to the particular mob's "soft cap".  Could it be a steep curve, perhaps something like if damage were porportional to x/AC+Atk, where x is some unknown modifier that possibly contains a portion of Atk as well?

Just curious.  It would be nice to think that someday a universal, covers-all-bases, formula could be discovered.  Wait, were we discussing the universe or Everquest?

[Coprolith]

Put a ruler or even just a piece of paper against the the 879ATk curve, you'll see that the average damage vs AC graph is linear all the way from 660AC up to 900AC

Oh and after seeing the complexity of the ATK rating unfold before my eyes i've pretty giving up hope of finding a simple universal covers-all-bases formula. I don't think we'll have a Grand Unification Theory of Everquest soon.  :)

/hugs