Sentinel Spreadsheet data validation (running tally)

[Underfyre]

10 hours ago, Bopper said:

If you're seeing your in game performance to be better than arcanatime, that would be worthy of a new thread that investigates that claim. You'll probably want to do long duration tests and look for what your performance truly looks like. If you are seeing something different than what other users see, then you may want to get other users to provide you with their own performance results to see if there is a discrepancy.

 

I mean I am, and let's face it I'm going to change to sheet, but doing long term tests and splicing videos is a different meal for a different day.

 

Looking back at WoWCraft, everyone that was a seriously raider back in the TBC days had a cast bar that accounted for your latency and you used a /stopcast macro in your spells to cut your casts off when you passed that point in your cast. Now if I were making a DPS sheet back then, there's literally no way I could account for the latency of everyone that could use the sheet. I guess I could maybe put a cell to reference an average latency, but probably not. I'd just use the base casts times. I'm gunna say same thing applies here. Arcanatime is a relic of the days of dial-up and lag. It's obviously not a hard-and-fast rule, looking at 3 abilities was enough to show that.

[Bopper]

On 6/17/2021 at 6:54 PM, underfyre said:

Zapping Bolt

Cast time: 1

arcanatime: 1.19

Anim before hit: .73

 

Cast time: 1.033

Anim before hit: .75

 

This illustration seems to show the most info so I want to refer to this. Can you explain what all the image is showing?

 

I see you have it lined up at 46s, so I'll assume that is the start of an animation. Then I see at 47.033s you have it marked... is the end of the animation? What's going on at the next block, the 47.35s mark?

[Underfyre]

In that chunk is the time for activating the ability to the next ability being activated. The 1.033 is the time in seconds that I have selected in the video. I just have the video sliced and separated from the rest of the video to make it easier to select, at the 47.35 mark there is the just the rest of the video.

 

I'm reading the original Arcanaville post right now to see how I should approach timing things.

 

Ok. So here's what I'm getting. Your first attack will have Arcanatime, but, so long as you're queueing attacks, they will not. Which is obviously what I'm seeing. The attack time's I'm posted are attacks buried in the chain so they're going off with minimal delay.

 

Options are:

1) Base cast times

2) Arcanatime for the first attack, but nothing else.

 

 

[Caulderone]

Could the recent decoupling of animation time and rooting time have had an effect on this?

 

Also, the servers of today are much, much better, so maybe the timing of "server ticks" could be much smoother/faster/better/tighter, too?

[Underfyre]

Just a quick test using an attack chain of 1 - 4 - 20 - 6

 

Base cast time: 1 + 1 + 1.5 + 2.17 = 5.67

Observed time 1: 6.483

Observed time 2: 6.250

Observed time 3: 6.233

 

Doing "dirty" arcanatime and just adding .132 to each cast gets 6.2s, doing long arcanatime gets 6.47.

 

So back to square one, glad I made a separate sheet with arcanatime removed.

Edited June 19, 2021 by underfyre

[Underfyre]

Looking at the last video segment I have loaded breaks down to:

 

Charged Bolts: 1.217s

Zapping Bolt: 1.250s

Havoc Punch : 1.500s

Tesla Cage: 2.267s

 

Charged Bolts and Zapping Bolt are both 1s casts, 1.19 casts under arcanatime. Havoc Punch is taking its exact cast time, and Tesla Cage is somewhere between its 2.17s cast and its 2.38s arcanatime.

 

For reference, my timing is started when the ability is started and ends when the next ability is started. For the end of the rotation, it's when the ability goes on cooldown.

 

Using /netgraph shows my ping between 190 and 200.

Edited June 19, 2021 by underfyre

[Bopper]

42 minutes ago, Caulderone said:

Could the recent decoupling of animation time and rooting time have had an effect on this?

 

Also, the servers of today are much, much better, so maybe the timing of "server ticks" could be much smoother/faster/better/tighter, too?

Unlikely. Number Six did a test recently (2 months ago) and printed out from the mapserver the exact time at which each activation occurred, down to the microsecond (iirc). They actually saw worse performance than what arcanatime reports, where instead of 0.132s between ticks it's more like 0.138-0.139s. Granted, this was a single test on their environment on the one map they tested on. 

 

Edit: there's also extra nuance to arcanatime worth mentioning. If a frame lands on a combat tick, there is no rounding up to the next combat tick. Not sire if this is the case, but it might be why you're seeing Havoc Punch not having much surplus in its arcanatime. It might be saving 0.132 (or 0.138?) seconds because of this

Edited June 19, 2021 by Bopper

[Underfyre]

1 hour ago, Bopper said:

Edit: there's also extra nuance to arcanatime worth mentioning. If a frame lands on a combat tick, there is no rounding up to the next combat tick. Not sire if this is the case, but it might be why you're seeing Havoc Punch not having much surplus in its arcanatime. It might be saving 0.132 (or 0.138?) seconds because of this

 

I dunno, but I'm seeing betwixt 0 delay and a quarter second. I'm thinking I can damage the pylon a little and wait for the regen tick and start the test then, but am I really going magically land on the 0.132 second server tick? Prolly not.

[Underfyre]

So follow up, I did that. Not only did it "queue" the ability Charged Bolts for 0.233 seconds, but it also still took 1.283 to execute. Starting to wonder what will happen if I switch the order of the attacks. So far Havoc Punch is either going off after either 1.5s, or 1.75s. But I've been bust today so I only cut up and measured 4 segments.

[Underfyre]

Checking out changed sequences to see how Havoc Punch reacts.

Base cast time:  5.67

 

20 - 6 - 1 -4

Observed: 6.483

Havoc Punch: 1.767s

Tesla Cage: 2.233

Charged Bolts: 1.250s

Zapping Bolt: 1.233s

 

6 - 20 - 4 - 1

Observed: 6.517s

Tesla Cage: 2.517s

Havoc Punch: 1.533s

Zapping Bolt: 1.233s

Charged Bolts: 1.233s

 

1 - 20 - 4 - 6 - This one had a pause between Zapping Bolt and Tesla Cage. It was queued, but Bolt went on cooldown and Cage didn't activate for 0.283s.

Observed: 6.250s

Charged Bolts: 1.267s

Havoc Punch: 1.733s

Zapping Bolt: 0.983s

server pause?: 0.283 (1.267 combined)

Tesla Cage: 1.983s

[Caulderone]

Wow.  That is a rather large variance in both HP and TC.

[Underfyre]

Well in that last run either Zapping Bolt is casting under time, or Tesla Cage is. The total time should be 3.17 for the two of them, and it comes out as 3.249 with that weird pause that happened, but without that pause both cast under time.

[Bill Z Bubba]

On 6/19/2021 at 1:04 PM, Bopper said:

They actually saw worse performance than what arcanatime reports, where instead of 0.132s between ticks it's more like 0.138-0.139s. Granted, this was a single test on their environment on the one map they tested on. 

 

I don't know why my brain is fixating on this but are there any explanations as to why this might be?

[Underfyre]

Alright, so I started fiddling with the arcanatime function and changing the ClockInterval part of it because I wanted to see what interval I needed to get the numbers I'm seeing. I was changing it starting at 0.112 and went up to 0.167 and noticed that the RoundUp function is making a wave pattern and resetting the numbers. Also I never got a number that matched any of mine.

 

On the left are some of the observed times for Tesla Cage, 2.17 cast. Yellow highlights are where the cast time "resets" because of the RoundUp function. The resets happen in a predictable pattern, 1 digit more than it did the last time. anyway, I'm thinking of altering how I go about gauging how I measure DPS. I'm thinking I'm going to doing it quick rotational bursts based off of cast time and just manually adding up the damage done in the combat logs, then comparing it to what is expected. It's going to take a bunch of these with all the variable involved, but with all the variations I'm getting just trying to measure the activation times it's just a shit show all around anyway.

 

[Underfyre]

Disregard. Wrong numbers used.

 

Edited October 2, 2021 by underfyre

[Underfyre]

So I've been doing some playing around just to check how some things react. I've Implemented a modified (for resists) Inverse Pylon time formula, then pulled DPS from inverse time. It's probably wrong to use it that way, but whatever. With all the extraneous casts to keep track of during a Pylon kill, especially over a Sentinel kill time, there's probably no way to get exact numbers. I digress.

 

Back to inverse time/dps. 

 

 

I fudged the Total time (added 8.2s) number until Inverse time equaled the Water/SR run.

 

 

I added a regen/HP cell for the Pylon. These cells have formulas to react to Degenerative and Diamagnetic.

 

Diamagnetic Radial at 40% from normal rotation.

 

Degenerative Core at 9.72%

 

So, unless I'm just way off base with my numbers and thought processes, it looks like with "all things being equal" Degen Core=Reactive Radial<Diamagnetic Radial. At least in terms of a Pylon run.

 

 

 

[Underfyre]

Diamagnetic Radial Runs:

406 : 222.27

348 : 238.02

409 : 221.58

 

Avg - 387.67 : 226.74

 

I forgot I used Hyrbid once these runs, then maintained using it once each run. Definitely fucked with numbers.

 

378 : 229.27

388 : 226.66

384 : 227.69

397 : 224.42

347 : 238.33

341 : 240.28

332 : 243.33

369 : 231.74

395 : 224.90

409 : 221.58

 

Avg - 374 : 230.36

Edited October 27, 2021 by underfyre

[Underfyre]

Degen Radial Runs:

319 : 248.03

348 : 238.02

353 : 236.46

 

Avg - 340.00 : 240.61

 

Same applies to these runs.

 

 

339 : 240.94

351 : 237.07

394 : 225.15

393 : 225.40

399 : 223.93

354 : 236.15

353 : 236.46

317 : 248.79

369 : 231.74

370 : 231.46

 

Avg - 363.9 : 233.20

Edited October 28, 2021 by underfyre

[Underfyre]

So I just had a "no fucking way" moment. Been fiddling with the inverse time formula, figuring out how to add in the time lost from casting extra abilities and what not. Obviously what I'm getting isn't what I'm seeing. But for gits and shiggles I decided to make a duplicate sheet and drop in manual times instead of the Arcanatime it does on its own. So normally I'm seeing between +0.233 seconds minimum, .0267 seconds, and +0.347 seconds maximum. I just averaged it out to 0.307 seconds.

 

The abilities that need to be tracked are Hasten, Ageless, Aim and Voltaic Sentinel for the purpose of the "no variables Elec/Bio."

 

Baseline it's looking at 214.29 DPS.

 

We turn that into 30667.15 / (214.29 * (1 + (-4.26%)) - 102.26) and get 298.08 seconds to kill the Pylon. But now we have to add in the lost cast times.

 

Unfortunately there's no easy way to do this without creating a circular reference that just fucks up everything you hold dear in life. Basically you take the total time you just got, and divide that by all the recast timers of the abilities you need to track. For that rotation and track the 4 abilities mentioned earlier it ends up being around 43.12 seconds.

 

You add 43.12 to the 298.08 and you get 341.20, and after pushing that through the standard pylon dps equation you get 240.21 dps. The exact average of the runs I did for the build.

 

I'm obviously jumping the gun here, and technically I should really be running the "lost time" equation a second time to account for the fact that the total time is now 43 second longer. But whatever, I just thought that was super wierd.

 

Edited October 25, 2021 by underfyre
Doing a second times makes the total 347.43 seconds, which drops the dps down to 238.2.

[Underfyre]

Anyway, yes. Gun, jumped. My issue isn't that the observed numbers and predicted numbers don't match up. That was never going to happen with the wide array of cast times you can have in just a 1 minute span. No, my issue is with how the inverse pylon time is working out.

 

The idea was to see how it played with Degen and Diamagnetic. I wanted to account for "unmeasured timings" from Hasten and the like so I could see how those Interfaces held up in an "equal" environment and the time would change when you changed the Interface. I have some a small batch of runs to get some observed times for those, but when I put the build data into the sheet? It greatly over values -Res versus -HP and -Regen versus what I'm seeing in game.

 

I've tried accounting for lost time in 2 ways now. The way I described above, and seeing how many "rotations" exist in the total time, then dividing the unmeasured time cell by the amount of rotations, then putting that amount in a new DPS. Oddly the two ways give different results, the first way being closer to the observed times.

 

So we'll say I have 20.90 seconds of unaccounted for time that needs to be put in the calculation. The predicted kill time is 256.53 seconds. now it's 277.44 seconds.

 

The other way is we have a rotation that takes 32.32 seconds to complete. So we divide 256.53 by 32.32 and get 7.94. We have 7.94 rotations that take 32.32 seconds to complete. So I also broke that 20.90  seconds into 7.94 chunks to get 2.63 seconds. Theoretically the rotation should be taking 34.95 seconds to complete. now it's 6450.44 damage/34.95 seconds. Clearly this isn't helpful when I've observed that I do 240.61dps and I'm adding time to make the already too low 199.61 dps now 184.57 dps.

 

That's because before I couldn't really account for HP/Regen debuffs. That's where the inverse time calculation comes in. I can account for them in there. But like I said, it's over valuing -res. Possibly because -res is double dipping? It's being accounted for the initial dps calcs, but then also being accounted for in the time calcs. I'm not really sure how to isolate it properly to get a correct measure. Maybe @Bopper can come in and save me from myself here.

 

 

[Underfyre]

Flat out removing resists from the time/dps calc does not do the trick. It brings Degen and Reactive closer, but Degen should be doing a good 4-5% more.

[Underfyre]

Manually removing resists from ability damage calcs and only having them in the time/dps calcs also does not do the trick.

[Underfyre]

I guess I need to expand my sample size to 10 per Interface or something. Feels like math is math. With all else being the same and the only variable (kind of) being the Interface, the math says Reactive should not beat out Degen.

Edited October 28, 2021 by underfyre

[DarknessEternal]

I'm following your journey.  Just letting you know you're not talking to yourself.

[Underfyre]

Okay, now I have what I would call a minor sample size for each Interface. I have a sheet that has the resistances per ability removed and they are instead passed through the inverse time function with the regen and HP total, in the event that resists were double dipping on effectiveness. It seems like they were? I kinda have manual times put in where I can control the amount of lag, since I have 3 different delays observed. I also use an average delay time. None of this matters particularly when I just arbitrarily apply different levels of lag to get what I want.

 

Anyway, we have observed dps of 233.20 for Degen, 231.99 for Reactive, and 230.36 for Diamag. All are fairly close together. In the sheet tho? Not so much. Degen and Reactive work out, but Diamagnetic? It doesn't intend to play nicely I guess.  It's lagging behind 8% of where it should be compared to the other two. Now sure what to do with that information.