Dexterity's impact on critical hits

[Shamaya]

Feel free for sure. Most (all except 39?) of the 30's haven't been covered with this new method.

I had almost forgotten about the term "RNG" in gaming (random number generator). I wouldn't have high hopes for "cracking" such a code but I've always felt that the randomness of FF was 'wavy' in nature. Multiple hq's in a row, long stream of nq's. Long stream of normal hits in a row, then several crits. I could just be arguing from voodoo but it just seems like something's up with it. Idk what the best way to set up the parser for doing it would be.. perhaps just a table, but the table could get really big. Something like,

Miss, Hit, Dmg, Norm, Crit
0, 1, 42, 1, 0
0, 1, 80, 0, 1

Something like that would be a basic way, recording a new line for each different round of attacks. 1 for true and 0 for false. So w/ that example syntax, that would be a normal hit for 42 first, and then a crit for 80 next. You could then copy the table to excel or another program to analyze it. The goldfish test and something like this would be a really good way to test for an RNG in regards to accuracy and crit (I think). When not hitting for 0, you could try analyzing an an RNG for pDif. You could use something like this to look at RNG for drop rates. I remember when I did some testing awhile back (not for RNG) I had to fill out a chart on paper . Synthesis would be of course great but, yeah..
I'm not sure if successful analysis results would have practical use though *shrug*, but it would be very cool and very convenient to go along with the massive 10-100 thousand+ sample sizes we're doing here.


edit, dDex50
http://img.photobucket.com/albums/v7...te50dDex02.png
So 24% is within 0.5% of this *shrug*

Masa, have you been through verified-by-visa, etc, steps? Have you registered your credit card with the program online? This would be necessary if you have an NA PC account.
edit: just a reminder. this is a continued parse from the previous dDex50 point listed a couple posts ago. so ignore that one completely.

[Masamune]

If you mean the mob intimidating you... Hmm. I don't think I actually show that. I can look into finding a spot to put that if you're interested.

I was wearing Blood Feet, and since those mobs have DragonKiller trait, i was getting intimidations at a decent rate i wondered: "lol i might get enough samples for my feet intimidation rate, and by extension the whole DragonKillerTrait, and furthermore could even assume all other killer traits work same way?"

In addition, I can gear my sam to be able to get pretty comprehensive coverage of anything in the +20 to +36 dDex range, and up to +40 with a little finagling/borrowed gear, while still keeping str at 60. I'd be willing to help out with the testing if there's any particular points in that range that you want checked.

Your are welcome :D
but you gotta make sure of the Hpemde's level you are gonna whack on (use HiDef /checks).(to know its AGI)
next is knowing which str max gonna make you hit for 0dmg on a given Hpemde's level: use table @ post http://www.bluegartr.com/forum/3204151-post244.html
and last thing i guess yu knew already: have a lot of time (ie let yur char whack afk during night^^)
EDIT: if you want to know @ which sample size to stop your current parse, you can still ask us here or use that link http://www.dimensionresearch.com/res...conf_prop.html using what KP currently displays (wished is error<0.5%).

Regarding which dDEX points needed, like Shamaya hinted and considering your DEX capabilities, the following ones in priority:
- dDEX=27/28
- dDEX=34/35
EDIT: that s directed also for Pchan, since you said yu can go up to +29dDEX.

the other areas are either less important (high linearity asumptions) or out of reach (dDEX44+).

Hope that helps you and /cheer !

EDIT: orz freaking POL isnot accepting the 3 last digits code of my new CC, even though i used it irl since 2 weeks...

[Motenten]

Sam/Blm, 68+34=102 Dex vs lvl 70 Ul'hpemde (67 agi), dDex = 35.

Melee Damage
Player            Melee Dmg   Melee %   Hit/Miss   M.Acc %  M.Low/Hi    M.Avg  #Crit  C.Low/Hi   C.Avg     Crit%
Motenten                  0    0.00 % 15601/2735   85.08 %       0/0     0.00   2010       0/0    0.00   12.88 %

12.88% crit rate with a confidence interval of 0.53%.

[Byrthnoth]

Well, that throws out Floor(dDex/7). I bet dDex=36 is 14% though.

[pchan]

shamaya found 13% @ ddex 39 so it seems the 13% crit range is longer than expected while the 14% one is short.

[Masamune]

Thank you deeply Mot !

Yur point basically "connects" with Shamaya's @ dDEX39=13% too, meaning we can maybe safely assume a straight line from the 12%->13% jump (should occur around dDEX27-28 ), untill dDEX39.

Below an overview of our current results, including latest Pchan Shamaya and Motenten data.
Green curve is a model fit attempt,
Red is upper bound of confidence intervals,
Orange is lower bound.
http://www.bluegartr.com/forum/attac...1&d=1257513277

Also, i'm not that sure where poeple claimed crits caps @ dDEX=50, because when i was testing pDIFs on ronf mobs, all my parsed crits added together gave ~25.10% +- 1.1%. Or should i discredit those parses ?

Other notice we can make when eyeballing the graph is when yu trace a line going through Shamaya's points @ dDEX39 40 41 and 42, it goes directly connect with dDEX49 and 50. But the JP data show way higher values in between so... need confirmations.

EDIT: updated results

[Motenten]

Got a typo in your label there. It says dDex = Player's STR - Monster's VIT.

I figure the jump to 13 is somewhere around 28-30. I'll run d29 tonight.

[Byrthnoth]

Oops, for some reason I read his mule's crit rate (14.38 ) for that point and was kind of worried about it. 13.00 is much prettier.

I know there was old lore that Crit might continue to increase at ridiculous dDex values, so perhaps you're seeing that with your Level 0 tests Masa. Or there could be some mostly negligible level correction function that makes 1% difference over 70 levels.

[Masamune]

Got a typo in your label there. It says dDex = Player's STR - Monster's VIT.

ewww can you be more specific, i can't find the typo :s

Oops, for some reason I read his mule's crit rate (14.38 ) for that point and was kind of worried about it. 13.00 is much prettier.

wtf you talking about o.O mule of who? and 13% for which point ?

Also, just double-checked my lvl0 parses, it's actually 6k ronf mobs killed all together, not 16k XD my bad with typos...

[Gethsemane]

Nevermind.

[Gethsemane]

Got a typo in your label there. It says dDex = Player's STR - Monster's VIT.

I figure the jump to 13 is somewhere around 28-30. I'll run d29 tonight.

I assume he meant he is seeing it as: dDex = Player's DEX - Monster's AGI. dDex being the amount of Dexterity you are adding? Not sure, but it is known that Agility does in fact affect Criticals landing. Gah double post. Can a Moderator delete my previous if you see this? Thank you.

[Deadgye]

The mistake is in the picture graph you posted.

[Shamaya]

That Masa. What a noob!

http://img.photobucket.com/albums/v7...Rate49dDex.png
dDex49 ---^
I'm thinking it might be more important to continue this parse to reach an adequate (the "accepted") level of uncertainty. Funny when a 20k sample isn't enough. But looking at this, it's a bit screwy maybe. Compare this to the 50dDex sample. Compare ~40 to 42dDex. See what I mean? No matter, I'll just continue and see how it looks. It's just not looking as nice and linear as I wanted, given how large these samples are.

[Masamune]

omg please forgive me.... what a noob i am :S i'll fix and add Sham's point (great job Sham !).

EDIT: ok fixed thank you for reports guys, i removed also dDEX46&47 data from the JP taru, they had more than 2.5% error and relatively low sampling, but still kept his dDEX43&45 data since they are still "near" the linear fit for that region.

[pchan]

I'm doing dDEX=20 right now, and it puts me at


dDEX=20 : 972/7905=12.3% +/- 0.72% => 12% since it can't be 13 % due to the tests for [21->25] giving 12%... will let it run a few more hours.

The 3 rd jump is probably at dDEX=19 ? Earlier than expected.


edit :
Final date for dDEX=20 :

dDEX=20 : 1493/11992=12.45% +/- 0.59%

[Byrthnoth]

So, currently we're looking at:
dDex = 0?-6; Crit rate = +0%
dDex = 7-13; Crit rate = +1%
dDex = 14-19; Crit rate = +2%
dDex = 20-29; Crit rate = +3%
dDex = 30-39; Crit rate = +4%
dDex = 40-50; Crit rate goes between +5%? and +15%

(?)s where we aren't sure of the end of a grouping. We haven't looked at dDex=40 yet to confirm it's 14%, but it probably is.

I somewhat suspect funkiness in the mid-30s. That 13% range is just so huge. It'd be cool to see what dDex=38 is. Our Lolibri graphs hinted at a potential decrease at 35+.

PS. Does it seem likely that higher dDex values don't floor their crit rate? It seems like a lot of the high level tests only contain one integer in their 95% confidence interval and it can be quite a distance away from the mean (look at 49 or 42 for instance).

[Masamune]

Ok Pchan updated your point, good job. but now that i recovered my account, i'll still do my planned dDEX20 data, maybe it will confirm your result more closely to 12% ?

I'll go dDEX=0 after.

[Motenten]

Computer shut down (or decided to go into sleep mode) sometime last night, so lost a lot of the time I thought I was testing. Got things running again, and reached a 'reasonable' level of data.

Lvl 72 Ul'hpemde (neutral def at 360 att), 67 agi; 68+28=96 dex, dDex = 29

Melee Damage
Player            Melee Dmg   Melee %   Hit/Miss   M.Acc %  M.Low/Hi    M.Avg  #Crit  C.Low/Hi   C.Avg     Crit%
Motenten                  0    0.00 % 12110/1977   85.97 %       0/0     0.00   1481       0/0    0.00   12.23 %

12.23% crit rate, .58% confidence interval.

Spot checking during the day I saw it varying between roughly 11.75% and 12.25%, so overall pretty certain this is part of the 12% tier.

[Masamune]

Excellent report Mot !

o.O lol Mot gonna get the most interesting part of this testing (and maybe Sham too):

yur dDEX29 just killed another supposed jump from dDEX/7 model...

Byrthnoth were right we can assume some funkiness in range dDEX25-40 lol, /cheer Motenten !


Regarding my dDEX20 parse this morning, my result is :
1393/11994=11.61%

Updated my post above http://www.bluegartr.com/forum/3214066-post286.html with curves and a new dDEX/10 model for dDEX15-40 region, and dDEX20 29 data.
I updated wiki discussion page also, please check if i did any typo.

Critical points needed now are :
dDEX30
dDEX47
dDEX52

Less important but additional points could be useful @ :
dDEX17
dDEX27
dDEX32
dDEX0 (i'll do this one next night)

THEN DONE

[pchan]

dDEX=19 : 1197/10833=11.05% +/- 0.59%