Ranger Firing Range - Testing

Eorzea Time
 
 
 
Language: JP EN FR DE
users online
Forum » FFXI » Jobs » Ranger » Ranger Firing Range - Testing
Ranger Firing Range - Testing
 Leviathan.Comeatmebro
Offline
Server: Leviathan
Game: FFXI
user: Rairin
Posts: 5811
By Leviathan.Comeatmebro 2017-02-15 16:24:34
Link | Quote | Reply
 
Oddly, aureole seems to have a portion of it's delay which does not prevent melee actions but does prevent starting another shot. When using aureole in the rapid shot test, I got significantly more melee rounds in, killing my uragnite and forcing me to restart the test twice.

Original Data(70 snapshot, varying delays, no rapid shot, ~250 samples each)

Rapid Shot Data(70 snapshot, varying delays, 32 rapid shot, exactly 250 samples each)

Relic Bow Data - 200 shot sample on tests without rapid shot, 1000 shot sample on tests with

All my test data zipped up and organized:
https://www.dropbox.com/s/zoef9239wedoypg/RangeData.zip?dl=0
[+]
 Lakshmi.Byrth
VIP
Offline
Server: Lakshmi
Game: FFXI
user: Byrthnoth
Posts: 5491
By Lakshmi.Byrth 2017-02-15 20:58:49
Link | Quote | Reply
 
0RS slope: 2.530038
32RS slope: 2.315272

.68*(2.530038) + .32*x = 2.315272
[2.315272-(.68*2.530038)]/.32 = x
x = 1.858894
1.858894/2.530038 = .73473

So the value of RS even at capped snapshot is right in the 0.71825~.75 range we were expecting.


However, Rapid Shot did not appear to affect Aureole, so perhaps it only works on weapons with ammo.


This means it is likely Velocity Shot that's more potent than we expect (for some reason) at low delays.


Edit: For the Relic Bow set, it's clear you have some non-procs in every category. Did you have 90 Rapid shot + 33 from traits, or was it 90 rapid shot including 33 from traits?
 Leviathan.Comeatmebro
Offline
Server: Leviathan
Game: FFXI
user: Rairin
Posts: 5811
By Leviathan.Comeatmebro 2017-02-15 23:44:20
Link | Quote | Reply
 
90 known
14 kustawis
10 pursuers beret
12 arcadian body
15 zerde gloves
19 pursuer legs
10 pursuer feet
5 yemaya belt
5 merits

rng99 so did have max trait as well
Offline
Posts: 158
By Chriscoffey 2017-02-16 02:26:58
Link | Quote | Reply
 
Lakshmi.Byrth said: »
0RS slope: 2.530038
32RS slope: 2.315272

.68*(2.530038) + .32*x = 2.315272
[2.315272-(.68*2.530038)]/.32 = x
x = 1.858894
1.858894/2.530038 = .73473

So the value of RS even at capped snapshot is right in the 0.71825~.75 range we were expecting.


However, Rapid Shot did not appear to affect Aureole, so perhaps it only works on weapons with ammo.


This means it is likely Velocity Shot that's more potent than we expect (for some reason) at low delays.


Edit: For the Relic Bow set, it's clear you have some non-procs in every category. Did you have 90 Rapid shot + 33 from traits, or was it 90 rapid shot including 33 from traits?
I fixed this just so no one mistakes it or wonders where the .65 comes from.
 Lakshmi.Byrth
VIP
Offline
Server: Lakshmi
Game: FFXI
user: Byrthnoth
Posts: 5491
By Lakshmi.Byrth 2017-02-16 14:34:32
Link | Quote | Reply
 
Okay, so a few observations from the Relic Bow tests:
1) Longer delays (like Relic Bow with 0 snapshot) are more variable - Probably because more packets have happened between the first shot and the followup shot and packet interval jitter pushes it around.

2) ~1% of shots fired with 100% Rapid Shot are as slow or slower than those fired without it. - This could be a rate cap, or it could be a property of Rapid Shot. For instance, if Rapid Shot is reducing delay by 0 to 50% and only picking whole percentages, 2% of shots fired would have the same delay as previously.

3) In the Relic Bow 0 Snapshot test, the peaks of the distribution are still ~every 400ms and if you take 400ms bins the occupancy of each bin is ~the same from min to max. - As far as we can tell sitting at the client, the distribution Rapid Shot is pulled from is flat. If we can't see periodicity beyond the normal packet intervals here, we're never going to see it and it also doesn't matter to players.

4) If you subtract 1800 from all numbers and only look at procs in the rapid shot samples, the average reduction observed in the R90S0 sample was 27%, while R90S40 was 28%. - This is right on track for the JP model.



At this point, I'm pretty satisfied with our model of Rapid Shot. Given the range of average delay reductions observed, I'd say it averages -27% delay on shots where it procs. I'd say Rapid Shot is solved.

Velocity Shot is the only thing that hasn't been tested in isolation in this round.
 Lakshmi.Byrth
VIP
Offline
Server: Lakshmi
Game: FFXI
user: Byrthnoth
Posts: 5491
By Lakshmi.Byrth 2017-02-16 14:42:55
Link | Quote | Reply
 
Oh, one more thing:
5) The maximum "proc" (below the minimum observed in the matching Rapid Shot 0 sample) was a 2% delay reduction for both pairs of samples.

So perhaps it's a flat 2-50% delay reduction. (26%)


I'll note that there's no reason to prefer the JP model unless they're sourcing it directly from SE. We've way exceeded the level of precision that was even possible in FFXI of days past.
 Fenrir.Snaps
Offline
Server: Fenrir
Game: FFXI
user: Mojopojo
Posts: 1139
By Fenrir.Snaps 2017-02-16 16:11:43
Link | Quote | Reply
 
The reason to prefer the JP model is that it matches the design of other parts of the game and it matches the data we have. If another model matches the data better though we should use that.

Maybe I'll do some VS measurements when I get home tonight (0%, 25%, 34%). I am anticipating that the results will match Verda's model and that the over-performance we see is due to truncation of some kind that is magnified when compounded delay reduction sources are used.
Offline
Posts: 158
By Chriscoffey 2017-02-16 18:36:36
Link | Quote | Reply
 
I believe in the description link under overkill byrth tested and said he believed it halved delay reduction. I was wondering if any of you will be testing this further with full sets or not to verify the validation of everything so far in ranged delay reduction with these gear/abilities combined at lower delay? If so that would be phenomenal for the ranger community. Thanks ahead of time if so.
 Lakshmi.Byrth
VIP
Offline
Server: Lakshmi
Game: FFXI
user: Byrthnoth
Posts: 5491
By Lakshmi.Byrth 2017-02-17 03:42:16
Link | Quote | Reply
 
Fenrir.Snaps said: »
The reason to prefer the JP model is that it matches the design of other parts of the game and it matches the data we have. If another model matches the data better though we should use that.

The JP model (the one that analogizes it to Conserve MP) doesn't really fit the Relic Bow data. The relic bow data is pretty evenly distributed over 10 packet intervals (vs. the 7 delays predicted in the Conserve MP analogy model).