Patch 7.5 Simulation Results

Comparative simulation data for FFXIV raid compositions for patch 7.5 at BIS

View on GitHub

Patch 7.5 Simulation Results (BIS)

This repo contains generated results for target dummy rotations. We used 12:30 rotations in a full party, single-target scenario to generate expected values for every hit and derive a true expected dps for each comp. This value is not impacted by crit-rng (like logs are), so it represents a more precise measurement of the rotation.

Additionally, for each comp we ran 10000 simulations to build up statistical metrics like variance to help visualize the lower and upper bounds of each comp. These 10000 simulations are also used to handle jobs with random rotations; we have solvers for the MNK/BRD/DNC random elements (RDM we treated as static), which should normalize the effects of rng on these jobs as well.

Disclaimer

You should not use the results of these tools to discriminate against any jobs in a public setting.

Contents and Methodology

We experimented with simulating jobs in 8-player comp situations to assess the strength of particular jobs in a target dummy situation against their peers. To limit variables, we started with three baseline comps:

From these comps, we focus on specific roles by testing all unique permutations without duplicates. Each comp is labeled with the 1-2 jobs it is specifically testing. All permutations are simulated against each other and plotted in the same graph for easy side-by-side comparison.

Below are the links to all generated results, but I will highlight important ones later in the doc.

Comp Link Comment
(Tanks)-AST-SCH-MNK-NIN-BRD-PCT results folder  
(Tanks)-SGE-WHM-SAM-VPR-MCH-BLM results folder  
DRK-GNB-(Healers)-MNK-NIN-BRD-PCT results folder  
PLD-WAR-(Healers)-SAM-VPR-MCH-BLM results folder  
DRK-GNB-WHM-SCH-(Melees)-BRD-PCT results folder AST excluded to remove carding impact.
PLD-WAR-SGE-WHM-(Melees)-MCH-BLM results folder  
DRK-GNB-AST-SCH-SAM-NIN-(Ranged)-PCT results folder  
PLD-WAR-SGE-WHM-SAM-VPR-(Ranged)-BLM results folder  
DRK-GNB-WHM-SCH-MNK-NIN-BRD-(Caster) results folder  
DRK-GNB-AST-SCH-MNK-NIN-BRD-(Caster) results folder Ranged cards were used on the caster.
PLD-WAR-SGE-WHM-SAM-VPR-MCH-(Caster) results folder  

Rotations

All rotations are hand-sheeted 12:30 rotations, using pots in the opener, 6 minute, and 12 minute windows. All rotations attempted to align with a ~6s buff opener. Regarding GCD tiers, we picked a commonly recommended tier for each job.

As mentioned before, MNK/BRD/DNC have some random elements in the simulation:

DNC Solver Output

MNK and DNC also scan other rotations in the comp to simulate the chakra/esprit generation from the party.

We acknowledge there might be user error in the hand-sheeted rotations and rng-solvers. We invite people to call out blatant errors in rotations if the mistake is easily verifiable. If you have a disagreement with the specific optimizations in the sheet, please provide completed, 12:30 rotations via xiv-in-the-shell export with gearsets for us to review and resimulate. If the outcome is significantly different (100s of dps), we will post revised results.

Important Simulations

There are many generated outputs and plenty of them might not be very insightful, so I’ll outline the most important ones.

Files and Outputs

Outcomes

These are initial reactions based on the comp_dps graphs generated by our sim. You can also draw similar conclusions by looking at the comp_dps_over_time version if you only care about expected value.

Name Comment Link
Buff Comp Dps (Tanks) DRK comps are the strongest, but now WAR is the best partner for it, overtaking GNB. Link
No Buff Comp Dps (Tanks) WAR and GNB seem relatively good here, with DRK suffering in a low buff environment. Link
Buff Comp Dps (Healers) Buff-using healers seem very strong in this experiment, AST/SCH gaining over 2000 DPS against WHM/SGE. Link
No Buff Comp Dps (Healers) SGE/SCH seems to actually be a narrow winner here. The SGE buffs are massive, so this makes sense. Link
Buff Comp Dps (Melees) The new buffs put RPR/NIN as the new top duo, while also reducing the overall spread. Link
No Buff Comp Dps (Melees) RPR/MNK is the winner, with RPR still being the strongest here. Link
Buff Comp Dps (Ranged) BRD is the strongest in this experiment, even with using SAM for DNC Partner. MCH is down about 3000 DPS. Link
No Buff Comp Dps (Ranged) BRD is still a clear winner, but by a smaller margin (~1200 DPS). DNC is very slightly above MCH. Link
AST Buff Comp Dps (Caster) PCT seems to win by a narrow margin. All casters (including SMN!!!) are within a ~1000 DPS spread. Link
Buff Comp Dps (Caster) All casters are in a very tight spread, with RDM being a slight favorite. Link
No Buff Comp Dps (Caster) BLM takes a dominant lead in this scenario, cementing a niche in low-buff comps. Link

7.4 Retrospective & 7.5 Predictions

For 7.4 (link), I focused more on general strength of jobs relative to their roles. This time around, I would like to review some of the 7.4 outcomes and speculate a little on what might be the most interesting jobs in the new ultimate.

The jobs I would like to talk about the most are RPR and RDM. These jobs have both been overshadowed in peak output by other jobs like MNK/NIN/PCT, but the buffs to these jobs have put them in a comparable spot while also giving both of these jobs a niche: resource pooling and multi-target cleave.

Top speedkill logs this tier utilized both: M9S was a fight defined by how teams navigated the tower add phase. The top 2 times took a “non-meta” job to solve the adds phase with on-demand cleave.

The idea here is simple: both RDM and RPR supply on-demand burst and cleave, which can be very helpful when every other job might be lacking free cleave from their CDs. This lets everyone else stay focused on the boss and minimizes boss damage losses.

M10S saw the top team bring out a double caster comp, replacing a melee with RDM. RDM is shockingly powerful with cleave, enough so that it was worth dropping a whole other melee for.

Regarding the other melees, I can also provide some context here since I was working with parryprog for their theorycraft: they checked the other melees and found them all to be roughly comparable (NIN/MNK/DRG/RPR) in net outcomes for their goal. So, NIN was taken due to time constraints since it was a comfort pick.

M12SP2 had a Korean team do some work with carryover RPR, using the P1 fight to bring a full gauge. This is a known strength of RPR, so it’s nice to see a team invest the effort in executing runs like this.

Moving on to 7.5, we see some well reasoned buffs to some underperformers (idk why RPR was buffed and not MCH though). However, the new ultimate can always pose a challenge: RPR/RDM can struggle when it can’t generate gauge, and ultimates always mix in downtimes that allow other jobs like NIN to wait out their CDs while losing less relative to more gauge dependent jobs. PCT is the most extreme example of a job that can get a lot of work done during downtime (via painting), so our new standouts have the odds stacked against them.

Gauge-dependence can be an asset though: if your group can find ways to allow these jobs to pool gauge, you can find a massive amount of extra damage. Not every phase in ultimate is necessarily a tight check, so a clever group might be able to make this work. We also might see two-target or add phases as well, which can allow RDM/RPR to shine. RPR in particular is unique because it generates extra gauge when something dies under its Death’s Design, allowing you to cheat out a lot of extra damage.

Disclaimers

Bias

These simulations bias towards buff-stacking comps. While I do believe the strongest comp will heavily utilize raid-buffs, the reality for most people in most situations is that people will fail to play around raid-buffs well. This failure to perform optimally is not something we try to emulate with these simulations, so it’s hard to know exactly how these results translate to a real world scenario. As such, I wouldn’t overthink the results here regarding specific underperformers and I encourage you to consider the context of your own situation when evaluating this data.

Mistakes

To reiterate:

We acknowledge there might be user error in the hand-sheeted rotations and rng-solvers. We invite people to call out blatant errors in rotations if the mistake is easily verifiable. If you have a disagreement with the specific optimizations in the sheet, please provide completed, 12:30 rotations via xiv-in-the-shell export with gearsets for us to review and resimulate. If the outcome is significantly different (100s of dps), we will post revised results.

Additionally, we accept there might be bugs in the underlying sim code and will be on constant lookout for them. We do verify our expected damage against gearset-annotated logs to make sure we accurately simulate the correct ranges. We also verified against the generated rotations in xivgear.app for line-by-line expected value (we consistently were within 1-2 damage per line).

All that said, I personally am not an expert on every job and there are certainly… decisions… you might find in some sheets. I also coded the solvers for RNG jobs, so there’s definitely the possibility of mistakes.

Discriminatory Use

You should not use the results of these tools to discriminate against any jobs in a public setting.

These results come from an idealistic scenario and are biased towards jobs that play into a raid-buff meta. Some of the jobs that are under-performing in these simulations fare better in more real-world situations since they are less dependent on their team and can have more consistent output. The goal of this assessment is to provide a new perspective on job performance that is not feasible to see via submitted logs.

Credits

P.S

If you are interested in speeds or spreadsheet theorycraft, please reach out to me on discord (apollo.van.waddleburg). We are slowly rolling out many of the tools we used to generate this report to interested parties and would love to develop a community around damage simulation and speed kills.