Why is everyone anxious about a PLD Rework?

[Mikey_R]

You do realize that your party composition is a fairly significant confounding factor in any aDPS comparison, right? It's not just influenced by buff alignment, but also the number of buffs you have access to (minus very specific buffs that can be used for artificial single target padding). That's why it's specifically intended for self-comparison. This is why we switched to looking at rDPS in the first place.

Yes, which is why, for general balance, averages are used. Also, party composition plays a role in rDPS calculations as well, so....

Also, rDPS was never the main measure for DPS until Dancer was made available and even then it is only because so much of Dancer's damage is based on how much it provides to someone else, a constant 5% damage buff is significant after all.

No, neither nDPS or aDPS is appropriate for job balancing, and they were never intended to compare performances across different jobs in different compositions in the first place. If you wanted to do this sort of a comparison, you would need to specifically determine the amount of additional damage a player is doing under raid buffs, and then normalize it relative to the number of two minute buffs active. If you can determine the actual per buff damage contribution and equate it to rDPS, then you can say how significant this effect is.

In this case, we can compare between different groups if we take an average, as that reduces the impact of specific party compositions. You can then use that data to compare. So, since we have data for rDPS and aDPS for all tanks, and since no tanks have a raid buff, we can take the aDPS of each tank and take away their rDPS to get an average of how much each job gains from raid buffs:

DRK: 7374.70 - 6915.93 = 458.77
GNB: 7343.68 - 6966.77 = 376.91
WAR: 7214.75 - 6870.19 = 344.56
PLD: 7153.74 - 6890.23 = 263.51

Ordered for your convenience. This is how much each tank gains from raid buffs, based on the upper quartile. PLD just does not utilise raid buffs very well. This is shown when comparing aDPS but is not shown when comparing rDPS as PLD's raid contribution there is split between the rest of the party.

Again, rDPS is not the end all be all of damage as it only tells part of a picture. Just in the same way aDPS/nDPS only tells part of the picture or even just the raw damage numbers only tell part of the picture. Unfortunately, since people don't care about the underlying maths or what the data actually shows, and just want a single number to compare, rDPS has come out as the king of damage (thanks to Dancer) when it only tells part of a story.

[Lyth]

...

The aDPS values are not normalized relative to the number of two-minute buffs in each individual run. You'd need to go back and look at it in a per run basis from scratch.

[Mikey_R]

The aDPS values are not normalized relative to the number of two-minute buffs in each individual run. You'd need to go back and look at it in a per run basis from scratch.

I got the values from FFlogs over all current savage raids over the last 2 months, that is between 28,000 parses for PLD and 48,000 parses for DRK, being the 2 extremes on parses uploaded, GNB and WAR are between them. The mean of each job is then calculated. For what we are trying to achieve here, this is more than enough to average out any difference in party composition or whether they got another 2 minute raid buff in or not. If I were to go back and evaluate each and every individual parse and subsequently find the mean, I would find the exact same results.

However, despite all this, it wouldn't change the fact that rDPS is still not the end all be all for how to balance a job. Again, it just doesn't show the whole story and can be misleading, which is why other DPS metrics exist. No one metric is perfect, noone is claiming that they are, however, it also means you cannot base balance around one specific metric.

[Lyth]

...

You've completely missed the point.

When you look at rDPS, you're looking at your own damage plus the additional benefit that your teammates happen to generate under your 16% uptime buff. That second part is really the only variable present. When you look at aDPS, your damage is influenced by the number of those 16% uptime buffs present. That's a major confounding variable. In order to eliminate it, you have to normalize your damage contribution such that you're looking at a 'per buff' gain. Again, aDPS isn't telling you what you think it's telling you, and really only has meaning in the context of two runs with the same group.

The parameter of interest is 'raid clear time', which is directly correlated to 'total raid dps'. The only individual parameter that we can currently analyze in this manner for balance purposes is rDPS, because it shows you individual contributions. If you want to refine that model further, you need to actually quantify what the dps benefit is for having that additional burst under buffs. Which has to be done on a per buff basis. No parameter that we have currently does that.

[Mikey_R]

You've completely missed the point.

When you look at rDPS, you're looking at your own damage plus the additional benefit that your teammates happen to generate under your 16% uptime buff. That second part is really the only variable present. When you look at aDPS, your damage is influenced by the number of those 16% uptime buffs present. That's a major confounding variable. In order to eliminate it, you have to normalize your damage contribution such that you're looking at a 'per buff' gain. Again, aDPS isn't telling you what you think it's telling you, and really only has meaning in the context of two runs with the same group.

The parameter of interest is 'raid clear time', which is directly correlated to 'total raid dps'. The only individual parameter that we can currently analyze in this manner for balance purposes is rDPS, because it shows you individual contributions. If you want to refine that model further, you need to actually quantify what the dps benefit is for having that additional burst under buffs. Which has to be done on a per buff basis. No parameter that we have currently does that.

This isn't about total raid damage, this is about how well PLD takes advantage of raid buffs.

This has been said already, but I will say it again. If you were to judge PLD solely on their rDPS, then, in order to maximise it, the PLD would have to do a 'sub optimal' rotation. If we were instead to look at a PLD's aDPS, by doing the optimal thing they will increase it and because they are connected, it will increase the rDPS of the party as a whole, however, the PLD's personal rDPS will show it to be lower.

This is the whole point. rDPS is NOT the whole picture. You have to use the other metrics to dig deeper to see what is actually going on.

[Shurrikhan]

(Initial reply obsolete; near simultaneous posting.)

You've completely missed the point.

No, you just seem to clearly not be reading fully reading it.

First, there are parses enough that we have reliable sample across compositions, effectively normalizing it the matter to again give a range with its various quartiles just as we would if we went through these various logs and created a spreadsheet accordingly.

Second, you can see this for yourself on any individual parse, or history of parses through a single static, right on fflogs, just by mousing over the rDPS cell; it provides a complete breakdown.

Take, for instance, the highest PLD parse in the game for Agdistis. That PLD would seem to have blown his 98th percentile DRK co-tank out of the water, dealing 4.3% more raw DPS and 6.1% more rDPS than the DRK.

Now, we cannot draw a fair comparison as to overall performance possible between jobs from that single log --due to confounding variables in gear and skill-- but we can look at their difference in rDPS granted to others relative to their own. The PLD granted 199.4 rDPS to his party through raid buff exploitation. The DRK, despite falling 2 percentiles lower, granted 318.0 rDPS.

In short, despite the lower percentile the DRK also brought to the table 118.6 more rDPS to his party through buff exploitation than did the PLD, all of which was not shown whatsoever in his individual parse despite increasing the party's total DPS. If we were to plug that back in, their rDPS would, be only 4.4% apart, not 6.1%.

Again, those who better exploit their raid buffs bring additional value to the party not accounted for under individual rDPS. Here is one of those cases where "rDPS parity" actually wouldn't be enough for real parity, so long as one job provides less rDPS to its buffers than another would.

Let's step back and look at the reason for this discussion in the first place. The tanks are currently balanced when you look at rDPS. But the underlying claim here is that certain tanks ought to do more rDPS than others to offset the fact that less of their damage contribution occurs under buffs, which then impacts the rDPS of the buff providers. In order to evaluate that claim, you need to be able to quantify exactly what that rDPS offset should be, right? Otherwise this is all just feelycraft.

Precisely that. As for an appropriate rDPS offset, there are a few factors worth considering:

1. Would the community be able not to go apeshit the moment one job has an rDPS lead after being promised parity, since it wouldn't perfectly match with their current at-a-glance, single-chart-only view of balance as seen through individual rDPS?
   (Else, we'll need to have some degree of imbalance still just because the still-balanced choices will be considered underpowered, and across the larger playerbase... the purpose of balance is more on the perception of one's ability to play whatever they like -- within the obvious limits of affinity/differences in specific gear, etc.)
   
   
2. How much should we accept/allow that chasing 100th percentile parses will require meta comps?
   
   
3. Let's pretend we could come up with some sort of rDPS-other hybrid that fairly tracks buff exploitation? How much credit would that metric give its exploiter?¹

¹ In the end, being able to view the real contribution a particular job gives to a party with a single metric is ideal; it's just that current rDPS metrics aren't able to meet that purpose. If we could design something that would, improving on rDPS's intent, provide a real at-a-glance indicator of how much a job brings to a party, perfect. If not, though, we're going to have to accept that figuring out that contribution won't be one-and-done.

At a rough guess, that metric might have players retain half the effect of raid buffs received, effectively just averaging their current rDPS and nDPS. I don't know. I can't promise you a perfect solution off the cuff; if I could, it'd probably already be in use.