Permadeath is the idea that, when your character dies, they cannot be brought back to life. Simple, right? Well, no. There are many flavors of permanence which feel very differently. This article aims to categorize and clarify them.
Permaconsequence by Player Choice
While the concept is most commonly referred to as “permadeath”, can we really say “permadeath” when the game is about a robot, or when the character canonically does not die, but escapes (as in HyperRogue) or is captured (as in Chess)? Thus, we will be mainly talking about permaconsequence, which means that the consequences of player’s actions are final, and we cannot simply return to an earlier state of the game. We start with some flavors of permaconsequence, classified by how much choice does the player have in playing this way.
Competitive flavor. Before video games, we had games played against other humans over the board. And the players themselves (or tournament judges) were responsible for enforcing the rules. It is quite natural that permaconsequence is necessary: if the players could just reverse their moves, nobody would ever lose. You could still ask the other player to let you undo your bad moves, but usually, it is not in their interest to allow that — so do not expect them to comply.
Examples: Chess has a specific permaconsequence rule, called touch-move rule: if a player has touched a piece, they must move or capture it (unless they announce they move it for some other reason).
Anti-example: Dominion online has an option to undo your moves, if other players agree. Most players consider it a good style to usually agree (at least when no new information was revealed in the meantime) — it is more fun to play quickly and sometimes click in the wrong place and undo that, than to play carefully.
Arcade-style flavor. First video games took inspiration from the above. However, the computer became the party responsible for enforcing the rules, and that included permanent consequence. Arcade games are typically short.
Examples: Pacman, Mario, Asteroids, Tetris, Spelunky.
Self-imposed flavor. While arcade games were typically short, we later moved to longer video games, which could not be completed in one sitting. Thus, the save game feature was introduced. The history of this feature is surprisingly hard to find; many sources claim that Legend of Zelda (1987) was the first game featuring it, but early computer role-playing games had it in 70’s. It was not hard to realize that this feature could be used to avoid permanent consequence by save-scumming. Game developers realized this too, and assumed that the players would likely save-scum in their game design, and balance for that. Hardcore players might still want to play with permanent consequence.
Commonsense flavor. This is very similar to self-imposed in that the game itself neither discourages nor encourages savescumming, however, if you think about the design of the game, it is clear that permanent consequence is intended. This is common in randomized strategy games, where you can decide to attack enemies, and an attack may randomly fail. Savescumming would allow the player to always win every attack, clearly ignoring an essential part of how the game was designed. So players who love the game will self-impose permanent consequence.
Examples: XCOM, Civilization, Castle of the Winds.
Anti-Example: Sokoban, Baba is You. These games are also about planning and making decisions, however, there is no randomness. Just like in strategy games, it would be possible for players to try out the plan in their head, and only execute it when they are sure that it works; but most players think that it is better to just do it on the board and possibly undo.
Optional flavor. In these games, the player decides before starting the game whether they will play with permanent consequence. If yes, the game will force it, for example, by deleting the save file when loaded. There are subflavors depending on which option is the default and how hard it is to change (from an option displayed prominently in the new game screen, through a option in the options menu, to an option labelled as cheat).
Examples: Diablo II, Hollow Knight, Dungeons of Dredmor, NetHack, Caves of Qud.
Gatekeeper flavor. Here, there is no option: permanent consequences are forced by the game, for example, by quitting on save and deleting the save on load. This is similar to arcade-style permanent consequence, but the difference is that the game is long enough to need a save feature… which usually means that it is also long enough to make it frustrating. This flavor is rare, because most people do not like the frustration of what feels like wasting a few hours. Even if you feel it is cheating, who cares if they are cheating in a single player game. One reason to use it is gatekeeping (the developers only wants hardcore players to play their game). People hearing about a game featuring permadeath may assume it to be gatekeeper flavor and thus not play them, but usually, such games are better described by some other flavor, including the ones below. It is also worth to note that, unless the game is played in some online system, the player can avoid gatekeeping by manually making backups of their save files (note that this is yet another difference from arcade-style). Popular gatekeeper games tend to be very good (if the players are willing to be frustrated, there must be something they love about the game…), so it is better to cheat like that than to not play the game at all.
Examples: Noita, FTL (Faster Than Light)
A bit of anti-example: Rogue, a game designed when most games were arcade-style (1980). Playtesters found it long enough to request a save feature; this save feature was added, but the authors did not like how it was used to save-scum, so they decided to enforce permadeath. This can be seen as competitive flavor —Rogue was originally played on multi-user systems, and the players were competing for high scores.
Pay-or-die flavor. An interesting variant where a game has two versions: a free version which is gatekeeping, and a paid version where permanent consequence is optional. This is done to appeal both to hardcore roguelike fans (who tend to be fine with gatekeeper permadeath but expect the game to be free, like most of the best roguelikes are) and more casual players (who expect games to be paid).
Examples: Ancient Domains of Mystery, Tales of Maj’Eyal
Permaconsequence by Severity
Here, we will categorize permaconsequence by severity.
No permaconsequence. The game lets the player save the game at any moment.
Example: Battle for Wesnoth lets the player save even in the middle of enemies’ turn.
Checkpoint. In this variant, saving the game is possible only at checkpoints, which are chosen by the game designer. This is generally not seen as a form of permaconsequence, but in some cases, if we separate a fragment of the game between two consecutive checkpoints (or, especially, after the last checkpoint), it may be long and challenging enough to basically count as an arcade-style permaconsequence game.
Example: Cave Story: in the end, we get a sequence of difficult boss fights without checkpoints in between. If we go for the best ending, we get a sequence of even harder boss fights after that. I think I have first learned of Cave Story in the roguelike communities, where this was seen as interesting for roguelike players. Similarly, Legerdemain and recently released Moonring are checkpoint-based roguelikes. The main dungeons of Moonring are 7-level deep and without checkpoints, making them basically roguelikes with full permaconsequence. People already recommend Moonring to new roguelike players.
Start again. This is what is usually meant by permaconsequence —player can start a new game and try again.
One try. In rare cases, the game is supposed to be played only once, so you cannot try again. This is rarely enforced, since most permaconsequence games aim for replayability. An experienced roguelike player may win a new roguelike on their first try, which might count as one-try commonsense.
Now, we will concentrate now on how permadeath fits in all of this.
Revival. So far, non-permanent consequence meant that the player would be allowed to restore an earlier state of the game. There is also another form of non-permadeath, where a dead character is simply brought back to life. Technically, this is different from permaconsequence —for example, if the character has tried to save themselves by using some kind of limited resource, that resource will be permanently gone, with no way to restore it. While technically this is a rather big difference, from the point of view of the player, revival mechanics are similar to ones based on restoring. Flavors such as optional or commonsense can be applied to (non-)revival as well.
Examples: DCSS (optional revival), Angband (optional revival), Diablo multiplayer (commonsense non-revival).
Permafail. In some, but not all, permaconsequence games, a likely consequence of player’s mistake is losing the game right away. Let us call this stronger version of permaconsequence permafail. While permadeath mechanics are sometimes permafail, sometimes, they are not.
Example: Rogue, apparently named so because you play a single character, opposed to most strategy games and classic RPGs. An usual effect of using a bad strategy is this character dying, and the player immediately loses the game in this case.
Anti-example: Hydra Slayer, HyperRogue in Orb Strategy Mode, Slay the Spire. In these games, tactical mistakes are not likely to lead to immediate loss, but rather, cause loss of health or some other (non-regenerating) resource. A player loses by making lots of such mistakes, or by using a suboptimal long-term strategy.
Anti-example: in Darkest Dungeon, permadeath is featured, in the sense that your characters die permanently. However, this is just a loss of resource, you can recruit new ones. While permaconsequence is still there, in roguelike communities, I have seen an opinion that this game does not feel like a roguelite (nor like a roguelike).
Upwards Metaprogression. In some games, characters die permanently, but you get resources which can be used to make further characters stronger. Usually, the more successful the character was, the more resources are available for upgrade. You will need these upgrades to progress in the game, at least for all but the best players. Thus, this kind of permadeath should not really be seen as a form of permaconsequence, but a part of the intended game progression.
Example: Rogue Legacy, Rogue Light,Hades. In Rogue Light, it seems impossible to defeat the final boss with the first character.
So far, we have been focusing on the “challenge” aspect of permadeath and permanent consequence. But it is worth to mention other aspects which are not directly related to challenge.
Strategic fail risk management. Consider a game which is a sequence of combat situations; for every combat, the player can either choose to fight (possibly losing the whole game) or to drink a potion (a limited resource, drink too many and they won’t have it when needed).
The player can roughly tell how risky combat is. This simple model looks random, but it already has strategy! Here is a graph of the probability of winning the game when player’s strategy is “if combat is safer than x, fight” (the exact rules of the game used in this simulation). Good players win 99% games, but if you play too safe or too overconfident, you lose.
So here we have a strategic element emerging as a consequence of permafail, resource management, and randomness (but not all games having these features would feature strategic fail risk management). Interestingly, this element is also related to computational complexity. Of course in actual games (like DCSS which inspired this) there is more strategy (computing the risk more precisely, applying good tactics so the risk is smaller, etc.) For players who like less challenge, such a game can still be interesting when played without permadeath, because you can still try to play well (e.g., good tactics). However, they lose an interesting aspect of the game.
Roleplaying. Another aspect worth mentioning is roleplaying. Role playing is about immersion, being a character in the world. If that character can die, or lose something permanently, the feeling of immersion is stronger. Roguelikes did evolve from role-playing games for a reason. I know a proverb “a deminer makes a mistake only once in their life”. This would make Minesweeper a canonical randomized permadeath game.
As we have seen, while “permadeath” is a commonly used feature, it is used to refer to a wide range of different mechanics, which feel very differently. Permadeath is also often given as a defining feature of roguelikes, although there are reasons why it should not be considered one. In literal roguelikes, permadeath is usually non-arcade-style permanent consequence, and usually optional (either by restore or revival), while in games with roguelike elements, forced (arcade-style) or metaprogression tends to be more common.