Arknights: Endfield Inferred Probabilities
Version: v3.11.1
The published rates
Endfield publishes more than the other four games do (not just the base rate and a pity limit, but the actual curve, pull by pull, in the game's own text). So this is the one page here where we're just reading a number out (everywhere else on this site we're working one out).
Headhunting, published rates
| Rarity | Base rate |
|---|---|
| Six star | 0.800% |
| Five star | 8.000% |
| Four star | 91.200% |
Those sum to exactly 100.0, so there's no rounding slack anywhere. On the Zenless page we get to wonder whether their four star line is quietly absorbing a rounding residue from the other two - here there's nothing to absorb.
The scale is different too. A top rarity operator is a six star here and the tier below is a five star, so if you've come from any of the other games your instinct is one off.
The six star curve
The six star rate is a flat 0.8% for the first 65 pulls. From the 66th it goes up by 5 percentage points every pull, and Gryphline says so directly: 5.8% on the 66th, 10.8% on the 67th, and so forth.
Then it stops early. Five points a pull would get to certainty on the 85th, but the guarantee arrives on the 80th, and by the 79th the curve has only climbed to 70.8%. So the ramp gets cut off with nearly thirty points still to go, and the 80th pull covers the rest.
The six star rate, pull by pull
| Pull | Chance |
|---|---|
| 1 to 65 | 0.800% |
| 66 | 5.800% |
| 67 | 10.800% |
| 70 | 25.800% |
| 75 | 50.800% |
| 79 | 70.800% |
| 80 | 100.000% |
We model that shape exactly. Drawing a straight line up to the 80th instead - which is what our other four games do, since none of them publishes a step - would be off by about three quarters of a percent on the long run rate. Small, but it's wrong in the direction that tells you to save fewer pulls than you need.
The five star line
Five stars are a flat 8% with one guaranteed inside 10 pulls, and we model a six star as clearing the five star counter, because the text says a five star "or above" every 10. That reading is ours and it isn't settled, and the deep dive has the argument.
The 120 pull guarantee
On the limited banner there is a 50/50 on the six star, and losing it guarantees the next one is the rate-up operator. This is a common mechanic. However the additional mechanic is if you haven't got the rate-up operator by the 120th pull on that banner, the 120th pull is that operator.
It's banner scoped, it doesn't carry to the next one, and it resets when the banner ends. It also matters more than it sounds like it should, since about 22.6% of players actually reach it (the deep dive works out why).
There's no Epitomized Path here and no fate points, because this does the same job more simply. And every 240th pull on a banner hands you a duplicate of the rate-up operator, which we don't model.
There is no rate-up five star
The limited banner rates up the six star operator and nothing else. GGanalysis (the best public library for this sort of thing, and one we cite happily elsewhere) models a 50% rate-up five star as well, and as far as we can tell that's just wrong - the banner data has no field for one and none of Gryphline's own notices mentions one.
The pool you are drawing from
The non rate-up six star pool is the five standard operators plus the rate-up operators from the previous two limited banners. So the pool rolls forward, and you really can lose the 50/50 to somebody who was the headline act two banners ago.
Your six star pity also carries between banners of the same type, which isn't true of the 120 guarantee sitting right next to it. Two counters, two different rules, and it's an easy pair to mix up.
If you came from Genshin
Genshin against Endfield, mechanic by mechanic
| Mechanic | Genshin Impact | Arknights: Endfield |
|---|---|---|
| Top rarity | Five star | Six star |
| Base rate | 0.600% | 0.800% |
| Soft pity | From wish 74, not published | From pull 66, published outright |
| Hard pity | 90 | 80 |
| Featured rate | 50%, plus Capturing Radiance | 50%, nothing on top |
| A guarantee on top of the 50/50 | None on the character banner | 120 pulls, once per banner |
| Epitomized Path | 1 Fate Point on the weapon banner | None, and none needed |
| Second rarity | 5.100% base | 8.000% base |
| Does the top rarity clear the second counter | No | Yes, we model it that way |
| A pull costs | 160 Primogems | 500 Oroberyl |
What this does not cover
In addition to what we don't know about the official underlying mechanics (we do our best to guess at the model) we also aren't modeling some interactions at all because they're too intricate or because I'm too lazy.
Star Rail's Custom Warp isn't modelled, since designating which limited five stars replace standard ones in the losing half of the 50/50 makes the pool a per-player setting rather than a property of the banner, and the engine has no notion of a pool the player edits. Zenless has the same idea under two different names.
Zenless Bangboo aren't modelled. They're a third item class alongside agents and W-Engines, and while the engine can express a third class now, no Bangboo channel is built.
Zenless Exclusive Rescreening isn't modelled either, because three channels run at once with their own featured agents while the count toward the guarantee is shared across all three, and we have no way to share one pity counter between banners in our calculator.
The permanent banners are left out too, which is Stellar Warp in Star Rail and Star-Studded Cast in Zenless. They're the ones that never rotate, so nobody is really planning a budget around them.
What we do not model here
Arsenal Exchange, which is where weapons come from, is a separate system and we don't touch it. Our calculator is operators only, which is also why this game has one item type on it where the others have two.
Bond Quota is the per banner five star, and it's a shop and not a pull - duplicates earn you a currency and you spend it. How many pulls that takes depends on what you already own, which makes it a good question and not one we answer yet.
The free pulls are left out too. Urgent Recruitment gives you a ten pull at 30 pulls in that touches none of your counters, and the Headhunting Dossier gives you one at 60 that you spend on the next banner. Neither one is in the numbers here.
The four star operators don't appear anywhere on this site either, and that one bothers us - there are five of them, they have names, and they're 84.7% of everything you pull. Our engine has two rarity tiers and a nameless remainder, and here the remainder isn't nameless.
Banner data
[1] Arknights: Endfield Wiki (CC BY-SA)
Everything here comes from a community wiki reproducing the in-game rules panel. No first party page states these rates (Gryphline's own notices just say to check in game), so treat this as one remove from the source, unlike the three HoYoverse games where we read an endpoint they publish.