Palworld has no active-skill inheritance rate. Not a hidden one, not an unknown one — there is no property in the breeding system that concerns moves at all. What the game does store is a flag on each individual move saying whether it may be inherited, a per-species pool of egg moves, and exactly one cake that passes a parent's whole moveset down. Everything else you have read about breeding for attacks is built on nothing.
All 251 signature moves are flagged non-inheritable, so you cannot breed one onto another species. Each species has a pool of egg moves drawn from a set of 47. Parent level is not a factor anywhere. Special Cake is the only lever. And 46 of the 47 egg moves are just buyable as a skill fruit, which is almost always the better plan.
The one number nobody can give you
Ask how likely a child is to inherit a parent's attack and you will be told 50%, or "one move per generation", or "three from each parent", or "up to six with the right cake". None of those figures exists in the game.
Palworld keeps its breeding settings in one place, and the properties are all prefixed Combi_. The shipped server binary contains exactly ten such names:
| Property | Serialized value? | What it governs |
|---|---|---|
Combi_BossPalRate | Yes | Alpha chance on an egg |
Combi_PassiveInheritNum | Yes | How many passives carry over (1, 2, 3, 4) |
Combi_PassiveRandomAddNum | Yes | How many passives are added fresh |
Combi_TalentInheritNum | Yes | How many Potentials carry over |
Combi_MutationRate | No | Mutation chance |
Combi_MutationInitialRank | No | Mutation |
Combi_MutationMinTalent | No | Mutation |
Combi_MutationRandomCoefficient | No | Mutation |
Combi_MutationRankCoefficient | No | Mutation |
Combi_MutationRankDiffPenalty | No | Mutation |
Not one of the ten mentions active skills, moves or attacks. There is no Combi_WazaInheritNum, not even as an unused name. So a 50% move-inheritance chance has nowhere in the game to live, and neither does a per-generation cap.
Four of those ten properties carry actual values in the Blueprint defaults, and six are names only. That split is the important part: we are reading the same asset that does hand us the passive and Potential counts, and it simply has no field for moves. When we say a figure is absent, this is what absence looks like — see how we verify things.
What the game does store: a flag on every single move
Move definitions live in DT_WazaDataTable, one row per move, and every row carries a boolean named IgnoreRandomInherit. That is the only inheritance-related field anywhere in the move data, and it is a property of the move, not of the pairing.
| Count | Rows | Note |
|---|---|---|
| Move rows in the table | 384 | Every distinct move id in the game |
IgnoreRandomInherit = true | 281 | Flagged as non-inheritable |
Signature (Unique_*) rows | 251 | All 251 flagged. Zero exceptions |
| Flagged but not signature | 30 | Broken down below |
| Unflagged rows | 103 | Every one appears in a learnset or an egg pool |
So the half of the community claim that says a Pal's exclusive move can never be bred onto something else is exactly right, and it is right for all 251 of them without a single special case. That is an unusually clean result, and it is the only part of the active-skill folklore the data supports outright.
One honest limit on how we are reading that flag: our evidence is the property name plus its perfect correlation with the signature moves. We have not traced the code path that consumes it, so we describe it as "flagged non-inheritable" rather than claiming to have watched the game refuse.
Correction: Double Fang is itself a signature move
Double Fang gets named constantly as the exception that disproves the rule — a signature-looking move that shows up on Pals that should not have it. It is not an exception. The display name resolves to Unique_Garm_BiteV2 on Direhowl and to Unique_GuardianDog_BiteV2 on Yakumo, and the elemental versions — Double Fang (Electric), (Ice), (Ground), (Fire), (Dark) — are their own Unique_* rows on other species. Several different Pals each have their own signature bite that happens to share a name. All of them are flagged.
There are real non-signature exceptions, though, and they are worth naming because they mean the rule is not simply "signature moves only". Of the 30 flagged rows that are not signature moves:
- Seven are ordinary Pal moves you can meet in game: Bee Quiet, Creeping Bubbles, Predator Blast, Predator Mark, Predator Surge, Psycho Gravity and Star Mine. Each appears in at least one species' level-up learnset, and each is flagged non-inheritable.
- Nine carry
DisabledData = true— cut or unused rows, several with placeholder name entries. They are not obtainable, so the flag on them changes nothing. - Fourteen are internal: the player's own punch and roll,
Weapon_Use,WorkAttack, partner-skill projectiles and funnel bullets. Most have no display-name row at all.
The egg-move pool: 47 moves, 9 to 45 per species
A separate table, DT_WazaMasterTamago — tamago is Japanese for egg — lists which moves a species can be born knowing. It is two columns wide, species and move, and nothing else:
| Property of the table | Value |
|---|---|
| Rows | 7,111 |
| Distinct species keys | 282 |
| Moves per species | 9 to 45 |
| Distinct moves across the whole table | 47 |
| Signature moves in it | 0 |
Moves flagged IgnoreRandomInherit in it | 0 |
| Columns for level, parent or probability | None |
That the pool contains zero signature moves and zero flagged moves is a good sign — the two tables agree with each other, which is what you want before trusting either. Yakumo has the largest pool at 45 moves; the smallest, at 9, belong to early-game species such as Vixy, Caprity and Mozzarina.
The complete set of 47, by their in-game names:
Air Blade, Apocalypse, Aqua Burst, Aqua Surge, Beam Slicer, Charge Cannon, Circle Vine, Comet Strike, Crosswind, Curtain Splash, Dark Laser, Diamond Rain, Dragon Breath, Dragon Cannon, Dragon Meteor, Flame Funnel, Flame Wall, Freeze Wall, Holy Burst, Hydro Laser, Hydro Slicer, Iceberg, Ignis Breath, Ignis Rage, Lightning Bolt, Lock-on Laser, Multicutter, Nightmare Ball, Pal Blast, Poison Fog, Power Bomb, Reflect Leaf, Rockburst, Rocky Impact, Sand Tornado, Sand Twister, Solar Blast, Spark Blast, Spirit Flame, Splash, Thunder Rain, Thunderstorm, Torrential Blast, Tri-Lightning, TriSpark, Volcanic Rain, Wind Burst.
Every one of the 282 species keys in that table uses the internal BOSS_-prefixed name. We therefore cannot prove from the files alone that this is the table the game consults for a bred egg rather than one it consults when spawning an Alpha. What supports the egg reading is the table's own name and a function name in the server binary, GetMasterableWaza_Tamago. What we do not have is the call site. We present it as the egg-move pool because that is the best-supported reading, and we are telling you the join we could not close.
Two further coverage facts, in case you go looking yourself: 274 of the 282 keys correspond to a species in our roster of 289, and 15 roster species have no key in the table at all — ten of them variant forms such as Jormuntide Ignis and Mammorest Cryst, plus Penking, Kingpaca, Panthalus, Needoll and Astralym. Whether that means "no egg moves" or "keyed somewhere we have not extracted" is not something the table settles.
Internal move ids and displayed names diverge badly. Flamethrower is Ignis Breath in game, HydroPump is Hydro Laser, Thunderbolt is Lightning Bolt, BlizzardLance is Iceberg, ShadowBall is Nightmare Ball. Any list that quotes the raw ids is a straight dump of the table; the names above are resolved through the game's own text table.
Correction: keeping your breeders at level 1 cannot work
The advice is to hold breeding parents at level 1 so they know fewer moves, on the theory that a smaller parent moveset concentrates the child's roll. It is expensive advice — it means never using your best parents — and there are three independent reasons it fails.
1. Nothing in the breeding path reads a parent's level or moveset
The egg-move table is keyed by species with no level column, no parent column and no probability column. The level-up learnset, DT_WazaMasterLevel, is three columns wide — species, move, level — and carries no probability field either. There is no term for a parent's current moves to enter, so a level-1 parent and a level-50 parent present the same information to the system: their species.
2. The two pools do not even overlap
Anubis makes the cleanest demonstration in the game. Its egg pool and its level-up learnset share not one move:
| Anubis egg pool (16) | Anubis learns by levelling (9) |
|---|---|
| Hydro Laser, Dark Laser, Iceberg, Pal Blast, Aqua Burst, Dragon Breath, Ignis Breath, Lightning Bolt, Solar Blast, Dragon Cannon, Spark Blast, Lock-on Laser, Ignis Rage, Tri-Lightning, Dragon Meteor, Nightmare Ball | Stone Blast (Lv 1), Power Bomb (7), Sand Tornado (15), Spinning Roundhouse (22), Forceful Charge (30), Ground Smash (40), Rock Lance (50), Sand Twister (55), Rocky Impact (70) |
Levelling an Anubis cannot dilute a pool that never contained its level-up moves in the first place. Three of the nine on the right are signature moves and flagged non-inheritable regardless, so they were never candidates either.
3. Under the one cake that does read parent moves, the advice inverts
Special Cake sets bInheritAllActiveSkills, which by its name passes the parents' active skills to the child. That is the only place in the data where a parent's actual moveset can matter — and a level-1 parent has learned almost nothing yet. If you are using that cake, the level-1 breeder is the worst parent you could pick.
Special Cake is the only active-skill lever, and the item text hides it
Cake effects live in DA_BreedingItemEffectData as a seven-field struct per cake. Four cakes have an entry, and exactly one of the four sets the active-skill flag:
| Cake | bInheritAllActiveSkills | PassiveInheritCountOverride |
|---|---|---|
| Mushroom Cake | false | 0 |
| Vegetable Cake | false | 0 |
| Extravagant Vegetable Cake | false | 0 |
| Special Cake | true | 4 |
Now read what the game tells you about it. The shipped description for that cake is "More likely inherit multiple passive skills from their parents." Passives only. The asset sets both fields; the item text mentions one. That is the whole reason the community consensus is that this cake is a passives cake — the description under-describes what the data does, and nobody has any reason to doubt an in-game tooltip.
You will see "Special Cake lets a child inherit up to six active skills". bInheritAllActiveSkills is a true/false flag with no number beside it, and no other field in the seven-field struct counts moves. Where the six comes from we cannot say, but it is not from this asset. We also have not verified in game what "all active skills" resolves to when the parents' combined lists exceed the slots a child has — the flag is confirmed, its resolution is not. Why we stop there.
Correction: the breed-only move list is one move, not eleven
A list of roughly eleven "moves you can only get by breeding" circulates widely. Every move on it is learnable by levelling up on at least one species, so as stated it is wrong. What is true of those moves is narrower and more useful: they have no skill fruit.
Cross-referencing the 47 egg-pool moves against the game's SkillCard_ item rows — 93 of them exist — gives a clean answer. 46 of the 47 egg moves have a skill fruit. Exactly one does not: Poison Fog.
| Poison Fog | Detail |
|---|---|
| In the egg pool of | Beakon (14-move pool) and Hangyu Cryst (16-move pool) |
| Skill fruit | None — the only one of the 47 with no SkillCard_ item |
| Learnable by levelling on | Nyafia, Caprity Noct, Prunelia, Cinnamoth, Lyleen, Tarantriss, Venusa, Vaelet — eight species, and Lyleen's entry appears only under its tower-boss key |
| Learnable by Beakon or Hangyu Cryst | No. Neither appears in the level-up table for it |
So the honest version of the claim is: there is one move in Palworld that a specific Pal can only be born with. Everything else you might want on a Pal, you can hand it.
Known versus equipped moves
The distinction players draw between the moves a Pal knows and the ones it has equipped is real — the save data keeps two separate fields, SaveParameterValue_MasteredWaza and SaveParameterValue_EquipWaza. What we cannot show is that breeding cares. No Combi_ property reads either field, so the popular advice to unequip a parent's unwanted moves before breeding has no support in the data. It costs nothing to do, and we have no evidence it does anything.
What to actually do
- Buy or farm the fruit. 46 of the 47 egg moves exist as a skill fruit item. A fruit is deterministic, immediate, and does not need a cake, a farm or an incubator.
- Breed for a move only for Poison Fog, and only on Beakon or Hangyu Cryst. That is the entire list.
- Stop paying for the level-1 trick. Use your best parents. Their level is not read anywhere in the breeding path, and under Special Cake a well-levelled parent is strictly better.
- Do not plan a chain around signature moves. All 251 are flagged. No cake, no pairing and no parent level changes that.
- Spend your breeding effort on what breeding actually decides: the species, which is fully deterministic, and the passives. Plan the species route with the breeding calculator and the breeding guide, and the passives with the passive skills guide. The cake side is in the cakes and Breeding Farm guide.
Verified against the extracted move tables on 17 August 2026: 384 move rows, 281 flagged, 251 signature rows all flagged; 7,111 egg-pool rows across 282 species keys and 47 distinct moves; Anubis's 16 egg moves and 9 level-up moves with zero overlap; 46 of 47 egg moves matched to a skill-fruit item row. Move names resolved through the game's own text table rather than quoted as internal ids.