Data Simulation
Contents#
(2024-04-23 UPDATE): With the 3.04 client update, that extra frame has been removed. Subtract 0.03 seconds from each time dated before this.
(2022-12-15 WARNING): With the 3.0 client update, the game appears to be adding an extra frame to reported times. You may experience 0.03s slower times than described below, but data rewards shouldn't be significantly affected by this. (Some tests here, read bottom-to-top.)
Introduction#
Training data is one of the most vital resources in Girls' Frontline, used to increase skill levels. It is primarily acquired through Data Mode, in Combat Simulation - a timegated resource. There, you must kill a dummy with a custom-built echelon in as little time as possible. The faster the time, the higher the reward.
Here, I'll go over the best formations for each level, and their expected data yields.
Setup#
Equipment
The dummy has 10000/20000/40000 HP for basic/intermediate/advanced training, and 0 armor/evasion across the board. It doesn't attack, so your only goal is to deal as much damage as possible. The best equipment to use is as follows:
In most cases, prefer SPEQs over these equipment pieces. Notable exceptions:
- Defender prefers buckshot over her SPEQ
- AP SPEQs that offer <10 FP are worse than Sniper ammo
- Ammo boxes aren't needed for the fastest times, but are important insurance for slower teams. It sucks to be just barely short on damage and be forced to wait for your MGs to reload. M1918's SPEQ there has a damage penalty and is thus worse for speed.
- Many MG scope SPEQs are better than an RMR, but worse than a VFL vs 0 evasion.
- ARs receiving crit rate tiles (e.g., from M4A1), likely prefer a VFL+HV ammo+Focus chip loadout.
It's possible for generic SGs to do more damage with slugs, but you shouldn't be using those for datasim anyways. Interesting SG skills generally don't work with slugs.
Formation
Don't use normal echelon formations in datasim. You have two priorities:
- Optimize tile coverage and composition for damage output. Basic formations are 3-4 DPS dolls (ARs/MGs/RFs) plus 1-2 HGs.
- Have your DPS as close as possible to the dummy to reduce bullet travel time:
It's sort of like coalition drill formations, but with better doll selection. For example:
Having your ARs in the very front would be even better, but Mk23's tiles are likely to save more than 2 frames. P22 (would be moved to position 3) and Contender can both work, although as neither should get to activate their skills, Contender is generally preferred for her crit tiles. More formation examples further down.
Dolls
These battles, hopefully, end very quickly. Many dolls won't even get to activate their skills. Dolls with good base stats and fast-activating/passive skills are ideal. Even if your team is slow, the dummy only has 1 link - don't use grenadiers like 416 or Sopmod.
Some dolls want unusual skill settings:
- While Kord typically wants to be on manual, she should be set to auto for datasim. This immediately puts her into assault (+FP) mode rather than piercing (-FP).
- RPK-203 should also be set to autoskill, to immediately go into burst mode.
Basic Training#
Released with the Fixed Point boxes, Rhino allows for record times in all sims. Due to the post-3.0 frame delay, her basic time (which would be 0.13s before) only tied the pre-3.0 0.17s rifle time. Now, it can get 0.13. As it's random what multiplier Rhino will start with, expect 1-10 tries to get this time. Try shuffling echelon slots & tile positions (keeping tile buffs optimized and DPS dolls as close as possible) to get a different RNG seed. After getting your leaderboard time, it's best to switch to some other more consistent echelon - if Rhino rolls low, you'll often get below max data. Oath requirements have not been tested. You may want one on Rhino to reduce RNG.
The basic dummy can be oneshot with rifles, with the fastest legitimate time being 0.17s. Typical comp is (Jill or Contender)/(Contender or Python)/M1895 mod/HS.50/(Mosin-Nagant or high-FP RF: DSR, IWS, etc). Time is 0.17s if your rifles are enough, 0.20s if it's left barely alive and a HG finishes the target - 219 data either way.
While slightly slower, various Defender comps also easily get 219 data:
Python/Nagant revolver are an excellent combo, and using them plus 3 MGs with solid upfront damage gets you a safe 218:
An older, but still reasonable, echelon takes 4 MGs with first-volley skills and 1 HG (Contender/Python/Jill/P22/K5, ideally).
Intermediate Training#
Slow Shock brings a new Intermediate meta, with the above team being 0.17s-capable (post-3.04). This is the fastest legitimate time, and yields 99 data. This needs good Rhino RNG, so I'd recommend switching to a more stable defender comp after getting your leaderboard time. Requires oath on Rhino/Type56R.
Pre-Slow Shock, the best team leaned on Rhino plus your typical datasim rifles. Oath requirements have not been tested. You may want one on Rhino/HS.50/Type56R to reduce RNG.
Released alongside Mirror Stage, Type56R's mod raised the intermediate bar to near-basic levels. There's some RNG involved, and it's possible to end up with a ~1.33s time (and 98 data) if the initial burst isn't quite enough. If you don't have some of these dolls oathed, I'd recommend some of the following Defender comps for a safer 99.
0.60s is possible with HS.50, but it needs oaths and is generally not worth the effort.
Python/Nagant/3MG safely gets 98 intermediate, as do 1HG4MG echelons:
Python/Nagant echelons even work when using nonsensical MGs.
Advanced Training#
An astonishingly fast 0.87s time can be achieved with TPS (as leader), Rhino, Defender, M6 ASW, and Px4 Storm mod. Some other doll compositions may also work. Formation image pending.
1.33s was the previous top time, enabled by Kar mod (post 3.04). The team:
This comp may also output 1.37s/1.40s/1.60/1.87s with worse Rhino/damage variance luck. It seems to be stable for 54 data, even if Rhino rolls poorly. Not much to do besides adjusting the formation and maybe feeding a CR or two if you're falling short of the leaderboard time. Switching/removing X-Exos from the HGs could also help adjust RNG. An oath is required for Kar, strongly recommended for Rhino, and optional on De Lisle/Python but marginally helpful for consistency. All skills on auto, equipment:
The previous Rhino comp:
Oath requirements have not been tested. You may want one on Rhino/RPK-203 to reduce RNG. For regular data farming, this is not recommended as suboptimal rolls on Rhino can cause you to miss max data.
Some previous (and more consistent) comps enabled by Kolibri from Mirror Stage:
1.70s is also possible with RPK-203/HS.50/PKP/M1895 mod/Kolibri - see this and this. It seems to require fully oathed DPS dolls, though.
Even with sniper ammo, getting below 2 seconds without oathing is a bit RNG. 2.17-2.20 seconds is fairly likely, so you'll have to experiment with the formation. Kord is likely to do better than M1918 mod (for a no-oath example).
The current fastest legitimate time on EN is 1.33s. 54 advanced data is technically achievable, needing clear times below ~2.0s, but needs top-tier picks to reach. 53 is much simpler, as most MG/HG echelons will get you there.
Reusing intermediate teams works:
Some MG/HG "gunboats":
Notes#
- If I missed any fun/interesting/fast/effective compositions, feel free to shoot me a message and I'll probably add it in.
- These echelons and formations are exclusively for data simulations. Do not use them anywhere else.
- Some cheated times may be removed from the in-game leaderboards every half-decade or so. In other words, there are plenty of cheated times up on the leaderboard. I would not take a team from there unless it is also shown here or has a number (~5+) of high-level players (~lv150+) using it.
-
If you use GFAlarm, you can test echelons by adding ids
10004,10005,10006to target saving. The customized target is positioned farther back and will report different times. - Some of the fastest leaderboard times may need oaths, but it's not necessary or even recommended for getting the max or near max amounts of data.
- RNG is somewhat fixed in data sims - try moving dolls around in the formation (if possible) or in echelon order to shake things up if you're just short of a threshold. You may also be able to influence it by equipping exos (in otherwise empty HG slots) with slightly different EVA buffs. It does not change over time, so if you find a setup that works, it should work indefinitely.
- Due to RNG (damage variance, crits), you may not get exactly the same times as in the images above, even if using the same dolls. Most of the 3/4MG echelons did not use sniper ammo in those screenshots to be conservative; the Defender/RF/top-tier ones did.
- See this page for evaluations of data packages in the store, in case this isn't enough data. Since 2.09, you're typically best off buying extra energy once a day and waiting for a 2x simulation yield event, though.
"Fun" with Frames#
If you like leaderboards, every fraction of a second matters. Let's cover a few frame-level details you usually don't need to worry about.
Startup time
When comparing basic data times to the bullet travel times, you may notice they are one frame slower than you'd expect. This implies there is a general 1-frame delay at the start. We can also observe this in normal encounters, to convince ourselves it's not a bug with the endcard time reporting:
With 72 ROF (20f), TKB starts an attack at 1.40s (42f). Her previous attack must have happened at 22f, and her first one at 2f - one past the first frame. Screenshot taken against id1833, as DJMax loved shoving SF right in your face (eliminating walk times).
Customized Target frame counts
For ease of testing certain ideas, let's get some frame references for the Customized Target. Times assume you start out with a doll in the front column, so there's no walk-up or uncertainty in the grid position. Keep in mind that tests against "Save target" use the data sampling ids/times listed towards the very top of the page, and tests against "Customized Target" go by these timings.
We can also see a 1-frame "startup" here: Mosin-Nagant plus some HGs/fairy can easily oneshot the 5000 HP dummy, and across each position the recorded time is one frame past those measured above. (For example, 0.30s or 9 frames from position 1)
Race conditions?
Not exactly related to datasims, but I thought of this while testing and had to know the answer. If you have two rifles that are individually strong enough to oneshot the dummy and are placed equally far, who gets the credit?
As you may have been able to predict, it's skill bar order. Each configuration was run ~5 times with the same results in case it was random.
MG ROF
MGs are typically said to fire with an effective ROF of 150 (10 frames per attack, 3 attacks/second). But there are some claimed exceptions, like the "old big 4" MGs. Let's test the all relevant ones, plus a few more:
| MG | Frame delay | Est. startup | Notes |
|---|---|---|---|
| M2HB | 11 frames | 3 frames | |
| PK | 11 frames | 3 frames | |
| Gr MG5 | 11 frames | 3 frames | |
| PKP | 11 frames | 4 frames | |
| Lewis | 10 frames | 4 frames | |
| Kord | 10-11 frames | 5-6 frames | Autoskill |
| RPK-203 | 10 frames | 4 frames | Burst mode |
| RPK-203 | 11 frames | 3 frames | Continuous mode |
| LWMMG mod | 11-12 frames | 3-4 frames | |
| M1918 mod | 10 frames | 3 frames |
There's some weirdness going on with Kord & LWMMG - see here for my specific observations. Different bullet travel speeds could explain the longer startup times, but a brief check with RPK-203 indicates MGs have the same travel times as those tested above.
NTW
If you've ever wondered why NTW mod doesn't show up in basic data leaderboards: her mod messes up her first shot, slowing it by 12 frames.
Against the standard data dummy, placing NTW as close as possible would result in her first hit coming in at 0.53s (16f, 1 startup + 3 travel + 12 delay).
Old basic data strats involved a mod3 NTW (to get her SPEQ) and a separate lv100 pre-mod NTW to use in the simulation. HS.50, Sniper Ammo, and so on have thankfully made such strategies obsolete.
HS.50
HS.50 takes an extra frame to fire her shot. This makes placing her on 6, and your other rifle on 3/9, vital for basic data ranking.
This also means, even if you didn't need the other rifle, you'd be stuck at 0.17s instead of the theoretical max of 0.13s. Type56R and KSVK mod have the same delay.
Defender
SG bullet travel time is slower than other dolls:
Even compared to that, Defender takes a bit longer. Her first shot comes in no sooner than 0.27s (8f). The 5-frame travel time against a data dummy puts her at a 3-frame startup time.
She takes 81f to reload, so her second hit would come in at 2.97s.
Python
Python's passive doesn't distribute its first stack until 0.47s (14f), and Kolibri's active is similarly delayed (despite being listed as a "0 second ICD"):
And many more...
The dolls covered here are picked because of their relevance to data sims. There are surely others with similar weird/undocumented timings and behaviors. As always, for best results, test things out. For details on this level, a screen recorder & video player with frame advance is your best friend.
More like this
General commentary and fairy information
On real performance and models of it