Dexterity's impact on critical hits

[Shamaya]

Cursed Soup is pretty good for -stats; -7. I used it while doing steelshells. Definitely the difficult thing is lowering str while being flexbile w/ dex...

I don't think ffxiah has a "sort by attribute" gear listing function. Somepage, being riddled with malware, is probably not a good place to go for gear browsing. But I did. They have a listing for dex and one for str (links are "xttp://"). Those lists also show which gears give -attributes. Ffxiclopedia has an article on strength but doesn't list gear; it does list gear for dexterity, but doesnt go into the minuses. Just a couple ideas for finding gear setups to do these tests.

[Masamune]

food isnot an option, considering this kind of test involve behing afk while sleeping and auto-whacking a mob.

[Massimo]

So you make a script that auto-uses food every 30 mins/3 hours, shouldn't be a problem.

Also:
http://images3.wikia.nocookie.net/ff...flame_Ring.png

[Shamaya]

That's a decent point, but this is a 3 hour food. There are ways to get around this, such as not sleeping while this is going, or setting your alarm clock, or as Mass said, making a script (though I wonder if a single round of hits out of the thousands would get off before the new food was consumed), etc. I can see this mattering a little if you want to do this from when you begin an 8 hour sleep, until the end of your 8 hour work day. A 16 hour data sample is pretty intense.

[pchan]

type win+R and enter the command :

shutdown -s -f -t 10800

this will shutdown automatically your PC after 3h

[Deadgye]

-22 is too much, it shouldn't be that low. It should be either -12 or -11.

Pretty sure it's -11.

(dstr+10/4) -> (-11+10)/4 -> -1

[Masamune]

a ok lol then my bad it s not about the fstr low cap yu were using, but ANY fSTR formula, as long using the exact correct str value making that formula output a -1 !

so that makes the current fSTR table @ wiki translated into :

dSTR range       fSTR approximation             max STR to get 0 dmg on:
                                         Ul'Hpemde lvl 68 69 70 71 72
12 or greater      (dSTR + 4) / 4 <=-1                 63 64 65 67 67
6 to 11            (dSTR + 6) / 4 <=-1                 61 62 63 65 65
1 to 5             (dSTR + 7) / 4 <=-1                 60 61 62 64 64
-2 to 0            (dSTR + 8) / 4 <=-1                 59 60 61 63 63
-7 to -3           (dSTR + 9) / 4 <=-1                 58 59 60 62 62
-15 to -8          (dSTR + 10)/ 4 <=-1                 57 58 59 61 61
-21 to -16         (dSTR + 12)/ 4 <=-1                 55 56 57 59 59
-22 or less        (dSTR + 13)/ 4 <=-1                 54 55 56 58 58

Highlighted the values matching the dSTR range from which used formula is. Means only the bolded values should wield 0 dmg.

I admit i didnot verify the level of the Hpemde i was testing this on :s so my blu75/sch1 with str64 could have been whacking at a level70 or lower, hence failing the test ?

brb to verify this...

EDIT: back from sea testing:

Doesnot work with BLU75/sch1 str64 (attack341) with IbushiShinai (DMG1 sword) on a Hpemde checked Neutral def @ 356 attack, ie either lvl71 or 72, which have VIT75 from earlier calculations. I got 1st hit dmg=1 and 4th hit = crit 2 dmg. So my dSTR were 64-75= -11. Note: i got intimidated with my CrimCuisses, which confirms the DRG job.

...Work @ str=60 ! work also with str=61, not 62 though, even increased my attack up to 347, still 0 dmg with spells set i got up to dex=64 and still have free slots. The only necessary things to make that work in my case is SnowRing+MorganaNeck+Cannonball to make str-3 (could use also that MalflameRing proposed earlier to free my neck slot^^). Checked mob further: its lvl71.

Corrected the table translation above with replacing "<0" by "<= -1", matching the tested str.

Sidenote: that fSTR method is a nice complementary method for confirming the determined couple {RaceVITrank;RaceDEFrank} from check+formulas method.

[pchan]

...Work @ str=60 ! work also with str=61, not 62 though, even increased my attack up to 347, still 0 dmg with spells set i got up to dex=64 and still have free slots. The only necessary things to make that work in my case is SnowRing+MorganaNeck+Cannonball to make str-3 (could use also that MalflameRing proposed earlier to free my neck slot^^). Checked mob further: its lvl71.

Corrected the table translation above with replacing "<0" by "<= -1", matching the tested str.

Sidenote: that fSTR method is a nice complementary method for confirming the determined couple {RaceVITrank;RaceDEFrank} from check+formulas method.

Worked for me too as PUP 75/ THF 37 and since I can equip mahatma body, I was able to get 60 STR. Unfortunaltely I don't have dmg1 club or dagger( trainee knife is all job) so I had to use a dmg1 staff with no skill...

so far @ 75 dex (ddex=8 ) : 231/2154 (10.72% with +/- 1.3% error)

If I had the dagger I could probably get higher sample sizes easily.

[Masamune]

O shit thanks god i just saw your post Pchan !

I just left my blu whacking afk one Hpemde lvl71 during all night to get 1147/12563=9.13%+-0.5% @ dDEX=6.
Just 1 hr ago, i woke up and stopped parse to add +2DEX for starting another parse @ dDEX=8, so far 101/1004=10.05%. I'll stop @ 7k hits ok? so can add my parse results to yurs.

I'm using ibushi shinai so high accu, and auto-refueling autoexec.
I'm curious how high can you go @ dDEX, personally i can barely reach dDEX=20 given my current gears ;;
Can't use mahatma/errant since when its indeed reducing str, its also reducing dex, greatly lowering yur dDEX testing opportunities.

Also did you check what level is the Hpemde you using as test subject? (lvl69&68 have lower AGI)
nvm yu said str60 so i assume yu are on a either a lvl71 or 72, same as me.

EDIT: so far my own linear fit matches pretty well that dDEX=6 point, with Crit%=dDEX/7 +29/350, it predicted 9.14% for dDEX=6 o.O. So for dDEX=8, predicts 9.43%, let's see...

Sidenote: Brutal ear parsed @ 5.14% at the dDEX6 test.

[pchan]

O shit thanks god i just saw your post Pchan !

I just left my blu whacking afk one Hpemde lvl71 during all night to get 1147/12563=9.13%+-0.5% @ dDEX=6.
Just 1 hr ago, i woke up and stopped parse to add +2DEX for starting another parse @ dDEX=8, so far 101/1004=10.05%. I'll stop @ 7k hits ok? so can add my parse results to yurs.

I'm using ibushi shinai so high accu, and auto-refueling autoexec.
I'm curious how high can you go @ dDEX, personally i can barely reach dDEX=20 given my current gears ;;
Can't use mahatma/errant since when its indeed reducing str, its also reducing dex, greatly lowering yur dDEX testing opportunities.

Also did you check what level is the Hpemde you using as test subject? (lvl69&68 have lower AGI)
nvm yu said str60 so i assume yu are on a either a lvl71 or 72, same as me.

EDIT: so far my own linear fit matches pretty well that dDEX=6 point, with Crit%=dDEX/7 +29/350, it predicted 9.14% for dDEX=6 o.O. So for dDEX=8, predicts 9.43%, let's see...

Sidenote: Brutal ear parsed @ 5.14% at the dDEX6 test.

Yes, STR 60 gives 1 dmg to some dolphin, and 0 dmg to others. So if I do 0 dmg I necessarely target one of the highest level (and therefore agi 67). My test @ ddex=8 (still not finished) guarrantees that you are > 9%, so if the crit% can only be an integer, 10% happens with some ddex<= 8. Your test for ddex=6 seems to show that you are still at 9%, so the crit% will jump at 10% at either ddex=7 or 8. So you are probably looking at

crit=9%+ floor(ddex/8 ) or floor(ddex/7) for early dex values.

[Masamune]

Thats exactly what i'm trying to prove: the linearity asumption CDF were talking about earlier. For that, only 2 possibilities:

- it can only accept integers like you suggesting, then our parses will show us ~1% jumps at some critical dDEX points,

OR

- it's a linear piece-wise function and then our parses should show very regular decimaled numbers following the regression line

BUT to prove that, needs very high confidence on our parse points, ie less than +-0.5% error (disproving "1% jumps").

EDIT: right now 554/5570=10.30% for dDEX8

[CDF]

My test @ ddex=8 (still not finished) guarrantees that you are > 9%, so if the crit% can only be an integer, 10% happens with some ddex<= 8.

Actually, to "guarantee" that something is not something else in statistics, you should really consider the power of your statistical test in some manner. (A confidence interval is basically one way to summarize the results of a statistical test applied to some data.)

Consider your given example. For n = 2154, if the crit rate was really 10%, the probability your two-sided 95% CI doesn't cover 9% is about 0.35, which is not really low enough to conclude that crit rate is discretized in 0.5% or 1% increments.

That aside, the outlined method with Hpemde is a pretty clever and convenient way to attain sufficient sample sizes to determine if crit rate (or any other melee-related rates) is discrete with minimal effort, considering all the trainee weapons are Lv 1 all jobs (Trainee Burin is 190 delay and the two daggers are 200 delay).

I have all these weapons and I was planning to check discreteness specifically, but I just downloaded Kparser and it doesn't seem to be working in Vista at the moment. The Alla equipment by job pages are down, too.

[Masamune]

Then please if you plan to contribute, let's at least allocate each other different dDEX points ok ?

Personally i plan to continue next with dDEX=10.

Just by eyeballing the upper/lower bounds i have currently, i suspect another jump occuring between dDEX=[11;14] (see image).

thoughts ?

EDIT: as shown in image, you can see if Pchan+my parses gets below +-0.5% error @ dDEX8, that will disprove the linear fit.
right now, i'm @ 10.42%=683/6560

[Shamaya]

W/ lvl1 subs I'm already at 59-60 str, so I'm in order. Base dex is 72. What points of dDex would you like to go for right now? I could see if I can get 40 or 48. Test some of the higher regions? 12dDex not so important now it seems. Prolly gonna go to sleep whackin one of these things tonight.
edit: after some fiddling w/ gear options, it looks like I can hit any point of dDex between 0 & 50 while maintaining -fSTr, for this; kinda cool
edit2: guess i'll do 39. could be nice to test the mettle of the current critical point candidate, and also contribute to testing for the tiered model

[pchan]

I'll do dDEX=8 for now for probably 50k sample, just to check if i'm around 10% with decent error, and then 7/6 to see if there is a jump. Just got traineee needle ... 90% hitrate on pup and 200 delay so i'm getting like 8x more melee hits than lolstaff.

[Masamune]

50k o.O wow is it that necessary ?

If we assume those "jumps" are 1%, an error of less than +-0.5% should be enough right ?

Also im getting more and more doubts if 2 parses can really be "added" together... especially regarding their corresponding errors :s

EDIT: nvm actually you right Pchan in the case of dDEX8 point, since the difference between 10% and the linear fit is 0.57% (to disprove it)

EDIT2: i'm now @ 991/9676=10.24%, which already disproves ANY linear asumption, with a lower error bound @9.64% (the best linear i could get is @ 9.43%, so out of bound).

[Masamune]

i'm wondering how to calculate a "tiered function" like Pchan suggested, in the form of :

instead of "y = a*x +b" -> y = floor(a*x) +floor(b)

knowing for example @ a "jump":

dDEX=7, crit%=0.09
dDEX=8, crit%=0.1

or would maybe be easier to just build a table with values, without any function ?

[CDF]

I'm doing 70 DEX (+3) with 60 STR to see if crit rate is floored there, more out of cheapness than curiosity. Sadly, dual wield 40% and 15% haste is offset by ~80% hit rate. While I can fix the accuracy issue later, I wish I had gear to proc Shinobi Ring, Koga Hakama, etc...

[Byrthnoth]

Can't you just use normal gear and get yourself beat on until you proc Shinobi Earring and Ring (24% Haste below 25% HP) and then kill the Hpmede you're fighting? If you're AFK with a D1 weapon, even assuming you somehow do damage and agro you're going to die anyway. Might as well have low HP. Just hide somewhere so people don't Cure bomb you as they're entering the palace.

[CDF]

That would work. All I would need then is for someone actually to put up both (they're never up when I actually look) so I can free up the other slots for adjusting DEX. I have blockaid on too to avoid interference as well.

At any rate, 5254 and counting in just under 3 hrs...