Honor Duel Shop Probability
Every question about the shop, from "will my hero show up?" to "how many badges will this cost me?", comes down to a few short formulas. This page walks through them with the odds Toto measured over 17,437 recorded rerolls. The shop calculator does the same sums for the shop in front of you.
Eight cards, eight sources
A shop shows eight cards. Each card is made by its own source, and a source is a short list of products with a chance for each. One source may always make a Random Pack, while another makes a hero card most of the time and a piece of equipment the rest.
- The eight sources draw on their own. What one makes does not change the others.
- A source is not a place on the screen. Its number is only a label.
- The lists change with the shop level, which rises with the round.
The cards on sale are hero cards with one copy or two, packs of 3 heroes picked from every hero or from one faction, equipment, Artifact XP, Trial Cards and the Random Legendary Hero. A two-copy card holds two copies of the same hero. A reroll costs 3 badges.
The full list for every source is under "See the odds" on the shop calculator.
Rounds, badges and shop level
Every round brings badges, and the shop level rises with the round. The level decides what the sources can make, and the badges decide how many rerolls and cards a round can pay for.
| Round | Badges gained | Shop level | Event |
|---|---|---|---|
| 1 | 60 / 70 | Lv. 1 | |
| 2 | +40 | Lv. 2 | |
| 3 | +40 | Lv. 2 | |
| 4 | +60 | Lv. 3 | Fitz's Trial |
| 5 | +60 | Lv. 3 | |
| 6 | +90 | Lv. 4 | |
| 7 | +90 | Lv. 4 | |
| 8 | +90 | Lv. 4 | Fitz's Trial |
| 9+ | +110 | Lv. 5 |
Will a product show up?
Several sources can make the same product. It is easier to work out the chance that all of them miss, and take that away from 1. Write p for a source's chance of making the product:
P(shows) = 1 − (1 − p1) × (1 − p2) × … × (1 − p8)
Two sources at 50% each do not add up to a sure thing. Both miss a quarter of the time, so the product shows in 75% of shops. This is each product's price in badges and how often it shows in a fresh shop:
| Product | Price | Lv. 1 | Lv. 2 | Lv. 3 | Lv. 4 | Lv. 5 |
|---|---|---|---|---|---|---|
| Hero | 15 | – | 100% | 100% | 100% | 100% |
| Hero ×2 | 30 | – | – | – | 27.7% | 100% |
| Random Legendary Hero | 35 | – | – | – | 15.7% | 18.8% |
| Random Pack | 18 | 100% | 100% | 24.9% | – | – |
| Lightbearer Pack | 33 | 43.8% | 25.0% | 25.0% | 7.6% | – |
| Mauler Pack | 33 | 43.8% | 25.0% | 25.0% | 7.6% | – |
| Wilder Pack | 33 | 43.8% | 25.0% | 25.0% | 7.6% | – |
| Graveborn Pack | 33 | 43.8% | 25.0% | 25.0% | 7.6% | – |
| Rare Equipment | 10 | 100% | 100% | 50.7% | – | – |
| Elite Equipment | 20 | – | 19.7% | 100% | 100% | 66.4% |
| Epic Equipment | 40 | – | – | – | 51.7% | 100% |
| Epic Trial Card | 6 | 39.0% | 35.1% | – | – | – |
| Legendary Trial Card | 12 | – | – | 41.3% | 48.0% | 15.8% |
| Mythic Trial Card | 22 | – | – | – | 4.9% | 15.8% |
| Supreme Trial Card | 45 | – | – | – | 0.77% | 4.1% |
| Artifact XP | 15 | 100% | 100% | 13.7% | 18.8% | 19.1% |
Will my hero show up?
A hero card names one of the shop's 32 heroes. The game does not show how it picks, so every hero is taken as equally likely. A source that makes a hero card with chance p then makes a card of your hero with chance p ÷ 32, and the same formula gives the chance q that a shop shows it:
q = 1 − (1 − p1 ÷ 32) × … × (1 − p8 ÷ 32)
From shop level 4 a hero can also show on a two-copy card. The table counts the two cards apart, and then together.
| Shop level | Hero | Hero ×2 | One hero shows | Rerolls on average | Badges on rerolls |
|---|---|---|---|---|---|
| 1 | – | – | – | – | – |
| 2 | 8.1% | – | 8.1% | 12.3 | 37 |
| 3 | 11.9% | – | 11.9% | 8.4 | 25 |
| 4 | 11.9% | 0.93% | 12.7% | 7.9 | 24 |
| 5 | 8.8% | 6.1% | 14.4% | 6.9 | 21 |
How many rerolls will it take?
Each reroll is a fresh try with the same chance q. The chance of seeing the card at least once in r rerolls, and the number of rerolls it takes on average, are:
P(within r rerolls) = 1 − (1 − q)r
average rerolls = 1 ÷ q
A reroll costs 3 badges, so the average cost of finding a card and buying it is:
average badges = 3 ÷ q + price
At shop level 5, one hero shows in 14.4% of shops. That is 6.9 rerolls on average, or about 21 badges on rerolls. The price depends on the card: in 39.0% of those shops the hero shows on two-copy cards only, at 30 badges in place of 15. The average price is then about 21 badges, and one copy costs about 42 badges in all.
An average is not a promise. After 7 rerolls there is still a 33.7% chance that the card has not shown once. For a small q this share is always close to 37%.
What a pack gives
A pack picks 3 heroes from its pool, and it may pick the same hero twice. With k wanted heroes in a pool of n, the chance that at least one of d picks is wanted is:
P(at least one) = 1 − (1 − k ÷ n)d
A Random Pack picks from all 32 heroes. A faction pack picks from the 8 of its faction, so the wanted heroes have to belong to it. A Random Legendary Hero picks one hero and gives 3 copies of it.
A pool is the shop's heroes less the ones you already hold at Mythic, which the shop and its packs no longer give. Every hero left in a pool is taken as equally likely.
| Wanted heroes | Random Pack | Faction Pack |
|---|---|---|
| 1 | 9.1% | 33.0% |
| 2 | 17.6% | 57.8% |
| 3 | 25.6% | 75.6% |
Dividing a pack's price by its chance gives the average badges spent before it pays off. For one wanted hero that is about 198 badges through Random Packs at 18 each, and about 100 through faction packs at 33. Selling the heroes you do not need gives 2 badges back for each, which lowers both figures a little.
What a lock changes
A locked card stays through the next reroll, and so does its source: that source makes nothing new while its card is locked. In the formulas above, the product then runs over the unlocked sources only.
- Locking a card from a source that never makes what you want costs nothing. Your chance on the next reroll is the same.
- Locking a card from a source that could have made it lowers your chance, because that source sits the reroll out.
- A bought card leaves an empty place until the next reroll. It is not replaced on the spot.
- Unlocking a card lets its source draw again on the next reroll.
Two sources can make the same product, so a card does not always say which source made it. The cards on show together help: at shop level 1, one source always makes a Random Pack and another makes either that pack or Rare Equipment, so two Random Packs on show came from those two sources, one each. Locking one of them keeps its own source out of the next reroll, not a source of your choosing. The calculator weighs each possible source by how well it fits the other cards on show.
What a purchase costs
- A purchase has to be paid in full before it is opened. Badges from selling what a pack gives come afterwards and cannot help pay for that pack.
- A hero card sells for 2 badges. On the calculator you name the heroes you would sell, and only those are sold when a pack gives them. A pack returns at most 6 badges, and so does a Random Legendary Hero, whose 3 copies sell together.
- When a pack misses, the badges from its sales are added to what is left before the next move is chosen. Each thing a pack can give is followed on its own. An average refund is never counted in advance.
- A bought card is gone until a paid reroll draws its place again, so one card cannot be bought twice.
Take two Random Packs on show at 18 badges each, and no sales. With 35 badges you can buy one, and with 36 both. Rerolling first costs 3, so buying two packs from the next shop, if it shows them, takes 39. With a single pack on show, 36 badges do not buy two: the first leaves 18, short of the 21 for a reroll and another pack.
Best chance and fewest badges
The calculator answers two different questions for each target, and each answer names the move to make first. Both rest on one rule: the value of a move is the value of each thing that can follow it, weighted by its chance.
Reading the answers
- Each answer is for one target on its own. Copies of different heroes are never added together. "Any wanted hero" and "Any target" mean finishing any one of them.
- A hero starts at one more copy wanted, and you can raise it. A one-copy card counts as 1, a two-copy card as 2 and a Random Legendary Hero as 3, always of one hero.
- A card marked as bought is not taken off the copies wanted. Enter the copies you still need after what you have really bought.
- Best chance is the chance of holding the target by the end of this round. It is not the chance of only seeing it in the next shop.
- "Buying only what is on show" is the chance with no reroll at all: the best set of purchases among the cards entered, several packs included when the badges allow.
Best chance
Best chance asks how likely you are to finish this round with the badges in hand.
V(s) = maxa Σo P(o | s, a) × V(s′)
The state s holds the copies still needed, the badges, the cards on show and what is known about their sources. a is a move you can pay for, o is one thing that can come of it, and s′ is the state after it. A finished goal is worth 1, and a state with no move left to pay for is worth 0. Each result of a purchase or a reroll is followed to its own best next move and weighed by its chance. Adding those up is the law of total probability.
This is why the sure card is not always the first buy. Buying a one-copy card now can be worse than keeping the badges for a two-copy card after a reroll.
Fewest badges
Fewest badges asks what the goal costs on average if you could keep rerolling.
E(s) = mina [ c(a) + Σo P(o | s, a) × E(s′) ]
E(s) is the average number of badges spent from the state s until the goal. Here the state holds the copies still needed, the cards on show and what is known about the sources of locked cards, and leaves out the badges in hand. c(a) is what a move costs: 3 badges for a reroll, the price for a purchase. Badges from selling a pack's heroes are taken off result by result. Once the goal is reached, E is 0.
A reroll with no locks can bring back the very same state, so E(s) stands on both sides of its own equation. The answer is the set of values for which every state's equation holds at once. With no locks and one card to find, which shows with chance q and costs p, it shrinks to the formula above:
E = 3 ÷ q + p
A Random Legendary Hero names your hero once in 32 times, so one hit costs 35 × 32 = 1,120 badges on average. Fewest badges does not pick it for a single hero.
The two answers can disagree. Buying a single card now may be the best chance this round, while saving the badges for a later shop costs less on average. And an average is only that: with bad luck the real spending can be much higher.
When to stop for the round
Honor Duel runs over several rounds, and the badges you keep are spent in later shops. Best chance looks only at this round. Fewest badges looks at the whole way to the goal: with no limit on badges every plan gets there in the end, so the plan to prefer is the one that spends the fewest on average.
That limitless count is only a way to compare choices. In play you have the badges in hand, so when the goal can no longer be finished this round, stopping beats spending on. Each round's first shop draws the unlocked cards again for free, so stopping shows you a new shop without paying for a reroll.
The calculator says to stop when its cheapest move cannot be paid for, when that move leaves no chance of finishing this round, or when the shop level rises next round and waiting costs less by more than a reroll.
Whether to lock or buy a card before stopping is judged by the same equation. A lock keeps a good card for the next round but keeps its source from drawing again. Buying frees the source, but the badges are spent even if a better card shows later. When the shop level rises next round, the sources make other things, so a lock or a purchase has to be judged against that level's lists.
With plenty of badges, best chance and fewest badges usually name the same move. When badges are tight they can part: if a pack of 3 heroes is the only way left to finish two copies this round, best chance buys it, while fewest badges leaves the costly pack and stops, to look for hero cards in later rounds.
Rules and observations
A rule is a comparison that holds in every case. When one choice is never worse than another, the other need not be looked at. An observation is a comparison with conditions, reached by putting average badges or chances side by side. It guides the decision while its conditions hold and can turn around when they do not, so it does not replace the calculation.
Rule 1. If you will buy a locked card anyway, buy it now
A locked card costs the same now or later, and a bought card's place is drawn again on the next reroll, which is one more chance. A lock is better than buying only if you might later give the card up and finish for less. What a lock is worth is that option to walk away, and whether it outweighs the extra draw is for the calculation to settle.
Rule 2. If you will buy both a pack and a sure card from one shop, open the pack first
Within one shop the order of purchases changes no price, and the cards stay until the next reroll. Opening the uncertain pack first lets you decide the rest after seeing what it gave. If the pack gives the hero, the sure card need not be bought. Opening the pack first is never worse.
Observation 1. If buying cannot overshoot, buy instead of locking
Hero cards cost 15 badges a copy, with one copy or with two. The usual use of a lock is to avoid an overshoot. With two copies still needed, buying a one-copy card means a later two-copy card pays for three copies, while with the one-copy card locked you can buy the two-copy card and drop the lock.
An overshoot happens only when an odd number of copies is left and a two-copy card fills it. Buying a two-copy card never changes whether the number left is odd or even, and buying a one-copy card when the number left is odd makes it even, which removes the risk. In those cases every way of finishing without the card still finishes, at no higher cost, with the card bought, so there is no reason to give it up.
If a product that gives several copies at once can show, such as the Random Legendary Hero with its 3, copies no longer cost the same each and this observation does not hold.
Take shop level 4, 60 badges, all 32 heroes in the pool, three copies needed and a two-copy card on show. Buying cannot overshoot, and it gives a 46.8% chance of finishing this round. Locking gives 47.6% when the card came from a source that cannot make a Random Legendary Hero, since the badges kept can still pay for one, and 46.0% when it came from the source that can, since that source makes nothing new while the card is locked. With 75 badges buying leads either way.
Observation 2. Needing two copies, lock a one-copy card when badges are tight and buy it when they are plenty
This is where the overshoot matters most. Take B badges and a one-copy card on show. With the card locked, finishing costs 30 badges either way: another one-copy card plus the locked one, or a two-copy card with the lock dropped. That leaves (B − 30) ÷ 3 rerolls.
With the card bought at once, the same number of rerolls is left when the last copy comes on a one-copy card, but (B − 45) ÷ 3, which is 5 fewer, when it comes on a two-copy card. In return the bought card's place is drawn again on every reroll. With few badges those 5 rerolls are a large share of all you have and locking is better. With many, the goal is usually reached before they matter and the extra draw is worth more.
Toto's calculator puts the boundary at about 60 badges at shop level 5 with all 32 heroes in the pool: locking is better below it and buying above, and near it the two differ by less than 1 percentage point. At the levels below, where two-copy cards are rare, buying is better even with few badges. This compares the chance of finishing this round. It does not count the worth of holding the bought card when the round ends short.
With two copies needed, a one-copy card on show and all 32 heroes in the pool, Toto's calculator gives these chances of finishing this round:
| Shop level | Badges | Lock | Buy |
|---|---|---|---|
| 5 | 45 | 46.4% | 36.9% |
| 5 | 60 | 71.2% | 71.0% |
| 5 | 75 | 86.3% | 86.7% |
| 4 | 45 | 46.1% | 46.8% |
The other cards on show move these a little.
Observation 3. If one copy is needed and the hero's two-copy card shows, buy it
For fewest badges, the other choice is to lock the two-copy card (30 badges) and reroll until the 15-badge card shows. If a reroll shows that card with chance p, it takes about 1 ÷ p rerolls, so waiting costs more on average whenever:
15 + 3 ÷ p ≥ 30, that is, p ≤ 20.0%
Every reroll meets the same choice, so trying a few rerolls first cannot beat the better of the two. By observation 4, packs do not change the comparison. At the shop levels where two-copy cards show, p is between 8.8% and 11.9% with all 32 heroes in the pool, so buying at once is better. p passes 20.0% only in a smaller pool, of 13 to 18 heroes or fewer depending on the shop level, for example once many heroes are at Mythic. Rerolling is then better. For best chance, buying finishes the goal on the spot, so you buy.
Observation 4. For one hero, a one-copy card beats a Random Pack or a faction pack
A one-copy card gives a wanted copy for certain at 15 badges. A pack picks 3 heroes from its pool with repeats, so with n heroes in the pool, k of them wanted, it gives 3 × k ÷ n wanted copies on average. With c for the pack's price, a copy through the pack costs at least as much as the card whenever:
c × n ÷ (3 × k) ≥ 15, that is, k ÷ n ≤ c ÷ 45
With one wanted hero, a copy costs 6 × n badges through Random Packs and 11 × m through packs of a faction with m heroes. With 32 heroes in the pool and 8 in the faction that is 192 and 88 badges, far above the card's 15. A pack beats one-copy cards only when wanted heroes make up a large share of its pool: more than 40.0% of the pool for a Random Pack and more than 73.3% of the faction for a faction pack. Selling everything a pack gives returns at most 6 badges, so with one wanted hero the answer stays the same.
Observation 5. For one hero, a faction pack beats a Random Pack
By the pack formula, a pack gives the hero with chance 1 − (1 − 1 ÷ n)3. Buying the same pack until it hits costs its price divided by that chance. With m heroes in the hero's faction and n in the whole pool, the two packs compare as:
33 ÷ (1 − (1 − 1 ÷ m)3) against 18 ÷ (1 − (1 − 1 ÷ n)3)
With 8 heroes in the faction and 32 in the pool, the faction pack costs about 100 badges and the Random Pack about 198. For small chances the chance is close to 3 ÷ x, so the comparison is roughly 11 × m against 6 × n. The two meet where the whole pool is about 2 times the size of the hero's faction: about 16 heroes for a faction of 8.
While the 4 factions keep similar numbers, the faction pack is always better. If many heroes of the other factions have left the pool, for example at Mythic, and it falls under that size, the Random Pack is better. When any of several wanted heroes of different factions will do, the Random Pack hits more often, so compare the two packs directly.
A rule holds in every case and can always be used. Use an observation only when the conditions in its reasoning hold: the pool size, the number of wanted heroes, whether a product that gives several copies can show. Otherwise the decision has to be calculated.
Where the odds come from
Nobody outside the game sees the sources. Toto recorded shops reroll by reroll and worked the lists out from them. The method is maximum likelihood estimation: choose the chances under which the recorded shops are most likely.
If each card named its source, the estimate would be a simple count. Over N draws from one source, with product k drawn nk times:
pk ≈ nk ÷ N
The sources are hidden, so dividing a product's total by eight times the rerolls gives only a pooled frequency, not each source's list. The records hold more than totals: which products show together, and how many of each one shop shows. For shops with no locks at one shop level, write c for a shop's count of each product and nc for the number of recorded shops with that count. The estimate is the set of lists θ that makes this sum largest:
Σc nc × log Pθ(c)
With the sources drawing on their own, the chance of a count c adds up every way the eight sources could have made it. A(c) is the set of those ways, and pj,z the chance that source j makes the product z it is handed in one of them:
Pθ(c) = Σz ∈ A(c) p1,z1 × p2,z2 × … × p8,z8
Every chance is 0 or more, and the chances of one source add up to 1.
- A kept card is not a fresh draw. A record counts only the newly drawn cards, read against the locks set just before the reroll and the sources those locked cards may hold. A lock added after the reroll belongs to the next one. Counting kept cards as fresh draws would skew the estimate.
- The range beside a measured figure is its 95% range. It comes from drawing the records again at random, with repeats, and repeating the estimate many times.
- Which products each source can make at all is worked out from the records too. The ranges hold that fixed and show only how far the chances may be off.
| Shop level | Rounds | Rerolls recorded |
|---|---|---|
| 1 | 1 | 988 |
| 2 | 2–3 | 2,047 |
| 3 | 4–5 | 2,395 |
| 4 | 6–8 | 4,545 |
| 5 | 9+ | 7,462 |
Four limits are worth knowing:
- A measured 100% is not proof that the game guarantees it.
- Different sets of lists can fit the same records about equally well, and the source numbers can be swapped around without changing anything.
- The records do not tell the 4 faction packs apart, so they are taken as equally likely.
- Heroes in a pool, and pieces of one equipment tier, are taken as equally likely too.