How to Dragoon Per Second: A DPS Paradigm

[Hustensaft]

Shrug! I don't know what the sim does.

Maybe the sim is right and the modelled rotation is wrong?

I don't even get why DoT clipping is so incredible bad - yes, skillspeed causes you to loose DoT ticks, but for the most part it also counters every lost DoT tick via the damage increase (we're looking at -2-3% damage lost for each dot application for each tick). If you go for simplicity as the Tick loss is pretty much neglectable - you're comparing a GCD gain to a potency gain.

If you take the rotation in the OP: 2884.2 potency across 10 GCDs (25 sec) or 288 per GCD.
You now take your GCD down to 2.2727 (== 11 GCDs in 25 sec), you add 288 damage for ~610 SS.

Converting that in DET: 610/7290 = 0.083 = 8.3% increase. (should be less - 610 SS would convert to less DET due to itembudget)

So a 10% DPS increase from SS translates to a DPS increase of less than 8% with DET. Does seem right to me that the sim has SS that strong.

[SunnyHirose]

The sim ends when the last event occurs after the duration cutoff - the "true duration" is then recorded and the dps calculated. There is less of a hard cutoff of all or nothing.

That's better than nothing, but it's still a discrete cutoff with more than one bias (the random DoT tick timing and whenever the last skill is used) that you're trying to model a continuous concept with, and a very sensitive one at that. Without some sort of proration or a sampling of skill speed values (some of them fractional), I'm afraid it's just gonna be screwy.

As for a realistic assessment of skill speed... I'm gonna make a judgment call and say it's not super different from how it was when I was raiding on MNK (for my alt). Just because it has the appearance of the worst dummy stat doesn't mean it's bad to have a lot. In certain pinches, it can make your life easier by the simple virtue of reaching your combo sooner (but the scaling this time ughhh). Actually, if I had to say, BotD maintenance is tougher on the whole than GL3 maintenance was in 2.x.

[pandabearcat]

Well I am still thinking there is something at work here.

Reasons why SKS should be valued lower:

1. DoT clipping. While this no longer decreases the value of DoTs (because of DoT SKS scaling) it means that if you end up clipping a dot, further SKS won't actually increase the damage of the DoT, lowering its dpct (relative).
2. CD usage. If a lot of DRG's high damage abilities are CDs rather than rotational abilities, then higher SKS simply means more usage of "bad" abilities, rather than increasing dps more efficiently. This doesn't make SKS bad, just lowers its effectiveness.
3. TP usage. The sim currently does not care about TP, but obviously in real scenarios you will. I am unsure about how "tight" DRG's TP consumption is, but it may be a concern.

Reasons why SKS would be valued higher than normal:

1. Fitting another GCD into a dps cd. For obvious reasons, since DRG are very cooldown focused, this means higher dps increase.
2. Higher Geirskogul usage, because of higher usages of 4th combo. This may very well be the reason that it can be better to have SKS. Also a reason why there may be more SKS plateaus.

So really I don't know. I certainly think SKS should be valued at higher than "nothing" past plateaus. I just don't know why.

This is one of the reasons why I made the sim - a lot of us have preconceptions on how stats or openers or whatever should behave. By distancing myself from this and making an impartial sim we might come to surprising conclusions.

Furthermore, to paraphrase the simcraft creators, I -want- people to be upset at the sim. I want them to say "you're stupid, there is no way SKS is this good" because it means people will scrutinize the output and either find an error, or learn something. And that makes a better sim, better stat weights and better rotations.

[Dervy]

But it won't be 288 potency per. Once you go beyond 2.4GCD, your chance of clipping Phlebotomize comes into play and Chaos Thrusts is even greater. The overall average potency per GCD decreases when your Skillspeed increases.

I can't pull out any math to back my argument unfortunately, but Skillspeed increasing DoT damage doesn't mitigate the DoT clipping once bit.

Also, when I threw together a growth curve for Skillspeed on our PPS, it wasn't linear at all.

[pandabearcat]

Well, just run the sim with varying degrees of SS and see what happens!

I trust in the priority rotation, though the breakpoints of "when should I IDC" may need tweaking.

[Dervy]

Panda/Sunny/Hustensaft if you PM me on Reddit or YouTube I'll give you a download link to my model so you can double check a few things if you'd like and see if there's any suggestions you could make.

And I just read your other post Panda. I agree that everything should be scrutinized. That's why I regularly ask people to check over my work cause I know I'm prone to make a mistake somewhere and learn from it. (Ty Husten)

We're only human after all! And if it does indeed shown that Skillspeed is better... then, well well well.

[enil]

Played around with the sample rotation and tried to get it as close as possible to the current optimal rotation.

Ending up with 16 CT/FT, 16 4th, 10 GK in the default sim.duration=210 parse. (could probably get better GK usage (buff optimisations etc.)
End up with an AVG dps of 990.74~
stat weights of
WD 1.0
STR 0.1352
CHR 0.0182
DET 0.0168
SKS 0.0069


I've noticed a couple of things after running a diag though.
All off-gcd buffs seem to be delaying all GCDs regardless double or single (double can delay next gcd to 3.0s+. single gcd seem to put your next gcd at about 2.8) As a result our average GCD becomes 2.532~ which is a major hit. (http://pastebin.com/z8iLHMmk)
Phleb still has it's old duration of 18 seconds instead of 24. (which was screwing with the rotation logic - ended up removing the conditionals for pheb/HT and putting them into the IDC/thrust combos).

[pandabearcat]

Yea the sim uses very unforgiving timings - I set GCD skill delay to 1 second and I believe oGCD delays are also 1 second, so any attempt to fit 2 (esp during opener) is going to leave a lot of delays.

If we have more data on how the delays actually function (esp jump delays - those were set to be very long iirc, from 2.5 days) then we can make the sim more accurate.

I will try to release a quick release w/o any revamp (since that introduced some bugs) that just has the jump fix so that you guys can do more accurate tests (iirc jump fix results in ~20-25 dps increase).

However you won't be able to "on the fly" tweak skill delays until I finish the skill system revamp which will likely take till next week.

I am glad you guys have played around with the sim and are writing better rotations.

I'm glad you guys found the reason behind the sks thing - if it was not clipping dots initially then getting to the point where it no longer needs to will end up being a huge breakpoint.

I was tempted to create another experimental setting "Incremental Stat" where it increments a stat from X to Y and gives you dps rankings every 5 stat or so. Should be easy to see jumps with that.

But the main plan is so that other people can write that logic for me - the UI should be simple to write your own code for once I get the documentation finished.

[enil]

If we have more data on how the delays actually function (esp jump delays - those were set to be very long iirc, from 2.5 days) then we can make the sim more accurate.

Problem is delays are very ping/latency dependant, someone who averages 150-200ms (me) will experience something significantly different from someone running 50-100ms.

Best to get someone with as low as ping as possible to get the base ability wind up/wind down and possibly implement some kind of latency variable (is this something we care about? - or we are just after pure theory).

[Dervy]

I guess we could base all of our work on Theory, then people with high latency can add a "latency variable" to panda's Sim and get measurements that way.

As long as the fundamentals of the Sim are concrete, players can just set their own variables for their own personal situation to get more personal weighting and adjust their own rotations accordingly.