# Gamecraft: the vocabulary of making games

462 terms. Each has a plain definition, what it is confused with, a real example, and a sentence you could say to a designer or an agent. Playable versions are at https://gamecraft.dqnotes.com.

## 1. What a game is

The first words: verbs, rules, goals, loops, and the agreement a player makes when they start to play.

### Verb
Also: Player action, What the player can do.
What the player can do, named as a doing word: run, jump, shoot, trade, talk. The list of verbs is the quickest honest description of any game.
Two games with the same art and story but different verbs are different games. If you cannot list the verbs, you do not yet know what you are making, and nor does anyone you ask to build it.
Not mechanic: A verb is what the player does. A mechanic is the whole rule behind it, including what the game does back.
Seen in: Mario began as Jumpman in Donkey Kong, a character named after his verb. VVVVVV takes the jump away altogether: besides walking, your only verb is flipping gravity, and the whole game grows out of that.
Say it: "This game has three verbs: run, jump and grapple. Anything that needs a fourth button gets cut, or becomes a new use of those three."
https://gamecraft.dqnotes.com/game/verb

### Rule
Also: Game rule.
A statement of what is allowed and what happens next: a bishop moves diagonally; touching a ghost costs a life. Rules stay the same however you play.
The fun is in the rules, and one rule can hold up a whole game. Change it and a tense game can turn into a screensaver, with not one pixel of the art different.
Not goal: A rule says what can happen. A goal says what you want to happen.
Seen in: Snake rests on one rule: eating makes you longer. Your own tail, the thing you have to avoid, is made out of your success.
Say it: "Add one rule: each apple makes the snake three squares longer. Nothing else changes. That one rule makes the game harder with no new content."
https://gamecraft.dqnotes.com/game/rule

### Goal
Also: Objective.
What the player is trying to make happen. A goal turns every action into a good one or a bad one.
Without a goal, nothing you do is better than anything else, and players drift off. A clear goal gives the player a question to ask of each move: does this help?
Not reward: A goal is what you aim for. A reward is what you are given. A game can have a clear goal and no reward at all besides reaching it.
Seen in: Pong's cabinet gave one instruction: "Avoid missing ball for high score." Six words, and you know what a good move is.
Say it: "The player does not know what they are doing here. Put one goal on screen: reach the flag at the top of the tower. Everything else in this level is on the way there."
https://gamecraft.dqnotes.com/game/goal

### Win and lose states
Also: End conditions, Game over, Fail state.
The conditions that end the game, for better or worse: the flag is reached, the health runs out. Some games have one, both or neither.
They decide what kind of question the game is asking. Only a lose state asks how long you can last, only a win state asks how soon you will get there, and both together give every moment something at stake.
Not goal: A goal is what you are trying to do. A win state is the exact check the game runs to say that you did it.
Seen in: Classic Tetris has a lose state and no real win: the stack rises until it reaches the top. Animal Crossing has neither, and you simply live there.
Say it: "Win is reaching the exit. Lose is the timer hitting zero. Falling down a pit is neither: it just costs you ten seconds."
https://gamecraft.dqnotes.com/game/win-and-lose-states

### Game state
Also: State, World state.
Everything the game needs to remember at this moment: where each thing is, how fast it is going, whose turn it is, who is alive. All else can be rebuilt from it.
Saving, loading, replays, undo and online play all depend on knowing exactly what the state is. A value left out of it becomes a bug that only shows up after a reload.
Not save file: The state is everything in play right now. A save file is the part of it you wrote down, and it can leave out what can safely be rebuilt.
Seen in: A chess position is a full game state you can see at a glance, plus a little you cannot: whose turn it is, and whether castling is still allowed.
Say it: "Keep the whole game state in one object: positions, speeds, health and timers. Saving is writing that object out, loading is reading it back, and particles are not part of it."
https://gamecraft.dqnotes.com/game/game-state

### Mechanic
Also: Game mechanic.
A verb together with the rules that answer it: what you do, and what the game does back. Double jump is a mechanic: the verb is jump, and the rule is that you may do it once more in the air.
It is the smallest piece of a design you can add, remove or argue about. Saying which mechanic you mean stops vague talk about "more gameplay".
Not verb: A verb is the action. Two games can share the verb jump and have very different mechanics, because the rules behind it differ.
Seen in: Portal's central mechanic is two linked holes, along with the rule that your speed comes through with you. The verb is only "shoot", and the rule is what makes it a puzzle.
Say it: "Keep the jump button, and change the mechanic: touching a wall counts as ground, and jumping off it sends you the other way at 300 px/s."
https://gamecraft.dqnotes.com/game/mechanic

### System
Also: Game system, Interlocking mechanics, Systemic design.
A set of mechanics that feed into each other, so that what one does changes what another can do. Fire burns grass, grass grows back, and wind moves fire.
Mechanics that connect produce situations nobody wrote by hand. A level has only what you built, while a system keeps surprising the people who made it.
Not mechanic: A mechanic does one thing. A system is what several do to each other, and the interesting part is the wiring between them.
Seen in: The Legend of Zelda: Breath of the Wild runs fire, wind, water and electricity as rules that apply to everything: grass catches light, the heat lifts your glider, and metal draws lightning.
Say it: "Do not script the burning bridge. Make wood flammable, fire spread to what it touches and rain put it out, and let the bridge burn because of those rules."
https://gamecraft.dqnotes.com/game/system

### Core loop
Also: Gameplay loop, Compulsion loop.
The short cycle of actions a player repeats the most, where finishing one round sets up the next. Fight, get loot, buy gear, fight something bigger.
This is most of what a player does for most of their hours. If the loop is not good by the tenth lap, no amount of content will rescue it.
Not game loop: The core loop is what the player repeats, over seconds or minutes. The game loop is the code that runs once a frame.
Seen in: Jaime Griesemer of Bungie described Halo as "30 seconds of fun" repeated over and over in different places, with different weapons and enemies. Monster Hunter runs on a longer one: hunt, carve, craft, then hunt something bigger.
Say it: "Core loop is dig, sell, upgrade, dig deeper, about ninety seconds a lap. Selling must always pay for something that makes the next lap better."
https://gamecraft.dqnotes.com/game/core-loop

### Feedback
Also: Player feedback, Game response.
Any way the game tells you what just happened and how you did: a sound, a number, a flash, a word. It closes the gap between doing something and knowing what it did.
People learn by correcting, and they can only correct what they are told about. Without feedback, a player cannot tell a bad plan from bad luck, and they stop trying to improve.
Not juice: Feedback is information: did it work, and how well? Juice is how richly that information is dressed. A plain "12 ms early" is feedback with no juice at all.
Seen in: Dance Dance Revolution judges each step in words, from Perfect to Boo, the moment you make it, so you always know how close you were and which way to adjust.
Say it: "When a shot misses, show where it landed and play a dull thud. Right now the player cannot tell whether they aimed badly or the weapon is random."
https://gamecraft.dqnotes.com/game/feedback

### Constraint
Also: Limitation, Design limit.
A limit you put on the player on purpose: little ammo, one button, a clock, a small map. It takes away an easy answer so the player has to find a better one.
Unlimited options produce one dominant habit and no thought. Limits are what make choices cost something, and a good limit is often the identity of a game.
Not difficulty: Difficulty asks you to do a thing better. A constraint changes which thing you should be doing.
Seen in: In Superhot, time moves only when you move. That one limit turns a shooter into something closer to a puzzle.
Say it: "Cap the player at two bullets, refilling one every 1.4 seconds. I want them waiting for enemies to line up, not holding down fire."
https://gamecraft.dqnotes.com/game/constraint

### Abstraction
Also: Simplification, Level of detail.
Leaving out the parts of a real thing that do not matter to your game, and keeping the ones that do. A chess knight is a whole cavalry charge reduced to one L-shaped move.
You cannot simulate everything, and the player cannot care about everything. What you leave out decides what the game is about and how much it asks of the player.
Not realism: Realism is how closely the game resembles the world. Abstraction is a choice about what to keep, and a very abstract game can still get the important part right.
Seen in: Football Manager can show a whole match as coloured dots on a pitch. It leaves out almost everything you would see at a stadium, and keeps what you are there to judge.
Say it: "Do not simulate each soldier. Treat each squad as one unit with one health number. Battles should resolve in seconds, and the choice is which squad to send, not how each one aims."
https://gamecraft.dqnotes.com/game/abstraction

### Affordance
Also: Perceived affordance, Visual language.
What an object lets you do, and how plainly it suggests that: a handle invites a pull, a ledge invites a climb. The psychologist James Gibson coined the word, and Don Norman brought it to design.
Players read a game from what things look like before reading any text. When the look and the use agree, nobody needs a tutorial. When they disagree, people push on walls for an hour.
Not signifier: Strictly, the affordance is what the object can do, and the signifier is the mark that tells you so. Game designers often say affordance for both.
Seen in: Mirror's Edge paints the things you can climb and use in bright red against a white city, so you can read your route at a sprint.
Say it: "Every climbable ledge gets the same worn yellow lip, and nothing else in the game uses that yellow. Players should know where they can go without being told."
https://gamecraft.dqnotes.com/game/affordance

### Interesting decision
Also: Meier's rule, A series of interesting decisions.
A choice with no obviously right answer, where you know enough to reason about it. A line credited to Sid Meier, from a 1989 talk, says a game is "a series of interesting decisions".
If one option is always best, it is not a choice. If you cannot tell the options apart, it is a coin toss. Between those two is where a player actually has to think.
Not meaningful choice: Interesting is about the moment of choosing: is it hard to decide? Meaningful is about what follows: did it change anything?
Seen in: Sid Meier returned to the idea in his own GDC talk in 2012, using his Civilization games, where a turn might set you weighing a city now against a better site two moves away.
Say it: "This shop is not an interesting decision, because one item is always best. Make the strong weapon cost health, so that a player on low health should sometimes pick the weak one."
https://gamecraft.dqnotes.com/game/interesting-decision

### Meaningful choice
Also: Meaningful play, Real consequences.
A choice whose result you can see and which changes what comes later. Katie Salen and Eric Zimmerman's Rules of Play asks for both: the outcome must be discernible and integrated into the larger game.
A choice nothing follows from is a choice players soon stop caring about, and if the result is hidden it may as well not exist. Both waste the work that went into the choice.
Not illusion of choice: An illusion of choice feels meaningful the first time and leads back to the same place. It is cheap to build, and players notice on a second playthrough.
Seen in: In Mass Effect, a mission on Virmire makes you save one of two squad members. The other is gone for the rest of the series, and it shapes who is beside you from then on.
Say it: "If siding with the guild changes only one line of dialogue, cut the choice. Either the guild closes its shops to you afterwards, or we do not ask."
https://gamecraft.dqnotes.com/game/meaningful-choice

### Progression
Also: Player progression, Gating, Unlocks.
The way a game changes as you play, giving you more to do and more of it within reach: new abilities, new places, stronger numbers.
It gives the player a reason to think beyond the next minute. A locked door you have already seen is a promise, and it keeps pulling you back through the game.
Not difficulty curve: Progression is what you gain. A difficulty curve is what the game demands. Good games raise both together, so the gain does not make you bored and the demand does not make you quit.
Seen in: Super Metroid is laid out around this: each new power, such as the Morph Ball, turns a gap you walked past hours ago into a way forward.
Say it: "Put a locked door in the very first room, in plain view. The key arrives in hour two, and I want the player to remember that door the second they get it."
https://gamecraft.dqnotes.com/game/progression

### MDA framework
Also: Mechanics, dynamics, aesthetics, MDA.
A way of seeing a game in three layers: mechanics are the rules, dynamics are what those rules do when played, and aesthetics are how it feels. It comes from a 2004 paper by Robin Hunicke, Marc LeBlanc and Robert Zubek.
Designers can only touch the rules and players only feel the top layer, with dynamics between them. It explains why a small rule change can transform a game, and it replaces the word "fun" with eight more exact kinds: sensation, fantasy, narrative, challenge, fellowship, discovery, expression and submission.
Not aesthetics as art style: In MDA, aesthetics means the emotional response, such as challenge or fellowship. It has nothing to do with how the game looks.
Seen in: The paper's own example is Monopoly: rents make the rich richer, the gap widens, and soon only one or two players are really invested. Its suggested fixes are at the level of mechanics: subsidies for poor players, and taxes for rich ones.
Say it: "The feeling we want is a tense race right to the end. Dynamically that means the lead has to keep changing. So the mechanic to try is taxing the leader on each lap, about 20%."
https://gamecraft.dqnotes.com/game/mda

### Magic circle
Also: The circle of play.
The boundary between a game and ordinary life. Inside, rules and goals you would never accept elsewhere are real. Johan Huizinga used the phrase in Homo Ludens (1938), and Rules of Play made it a term of game design.
It explains why players accept being lied to, shot at and bankrupted, and why it hurts when the outside gets in, whether through a cheater, a tilted friend or an unskippable advert.
Not immersion: Immersion is a feeling of being absorbed in a world. The magic circle is an agreement about which rules apply, and you can be inside it during a game of cards with no world at all.
Seen in: At a poker table, lying about your hand is the game and nobody takes offence. Say the same thing about your wages to the same friend the next morning, and it would be a lie.
Say it: "Do not put a purchase pop-up in the middle of a match. Everything on this screen should belong to the game; save the store for the menu, outside the circle."
https://gamecraft.dqnotes.com/game/magic-circle

### Lusory attitude
Also: Accepting the rules.
Agreeing to a game's rules because they make the game possible, though there are easier ways to reach the goal. Bernard Suits defined playing a game as "the voluntary attempt to overcome unnecessary obstacles".
It says what cheating really takes away. A player who skips the obstacle reaches the goal and has not played, which is why auto-win codes get boring so quickly.
Not magic circle: The magic circle is the boundary around the game. The lusory attitude is the player's own willingness to step inside and stay there.
Seen in: A Nuzlocke run of Pokémon adds hard rules that the game does not enforce, such as treating a fainted Pokémon as dead. Nothing makes players keep to them, and that is the whole point.
Say it: "Do not punish players who ignore the speedrun route. Add a pure mode with no map markers, and let the people who want a harder game choose it themselves."
https://gamecraft.dqnotes.com/game/lusory-attitude

### Paidia and ludus
Also: Free play and structured play, Play and game.
Two ends of one scale, named by Roger Caillois. Paidia is free, made-up play, with no rules to speak of. Ludus is play bound by rules and goals, made harder on purpose.
It stops the argument about whether something "counts" as a game. Most good games sit somewhere along the scale, and knowing where you want yours saves you from adding scores to a thing that needed none.
Not toy versus game: Paidia and ludus describe how a person is playing. Toy and game describe the thing they are playing with. You can play freely with a chess set by building towers from the pieces.
Seen in: Minecraft holds both at once: Creative mode is close to pure paidia, with free blocks and no danger, and speedrunning its Survival mode to the Ender Dragon is about as ludic as play can be.
Say it: "Open with paidia. Give them five minutes of painting the town with no objectives. Only then introduce the first quest, and never lock the paintbrush away."
https://gamecraft.dqnotes.com/game/paidia-and-ludus

### Four kinds of play
Also: Agon, alea, mimicry, ilinx, Caillois's categories.
Roger Caillois's four appeals of play, from Man, Play and Games (1958): agon is contest, alea is chance, mimicry is make-believe, and ilinx is vertigo. Most games mix several.
It gives you words for what someone wants from a game. Adding contest to a game people came to for make-believe, or luck to a game they came to for contest, can hurt it badly without changing anything they could call a bug.
Not genre: A genre describes what a game is like. These describe what the player is enjoying, and a single genre can offer any of them.
Seen in: Street Fighter is almost pure agon, a roulette wheel is pure alea, The Sims is largely mimicry, and Wipeout at top speed is reaching for ilinx.
Say it: "Our audience comes for mimicry: they want to be a captain. So do not make the crew rankings public. That turns the game into agon, and the wrong people will show up."
https://gamecraft.dqnotes.com/game/four-kinds-of-play

### Toy versus game
Also: Software toy, Sandbox.
A toy is something fun to handle with no goal attached. A game adds a goal and some way of failing at it. Many good games are a good toy with a game built around it.
If the bare action is not pleasant with no score attached, goals and rewards are just paint. Testing the toy first is one of the cheapest experiments in making games.
Not paidia and ludus: Toy and game describe the object. Paidia and ludus describe how someone is playing with it.
Seen in: Will Wright said of SimCity, "I always wanted it to be much more open-ended, more of a toy." His publisher wanted a win or lose scenario, and to him it felt more like Lego.
Say it: "Before any levels exist, I want a grey room with just the grappling hook in it. If people do not play with that for ten minutes, a goal will not save it."
https://gamecraft.dqnotes.com/game/toy-versus-game

### Player fantasy
Also: Power fantasy, Class fantasy, Role fantasy.
Who the game lets you feel you are: a dragon, a master thief, someone who gave up a desk job to run a farm. It is a promise that the numbers, the verbs and the art all have to keep.
It settles design arguments that rules alone cannot. If the fantasy is a dragon, then a villager cannot take twenty hits, however well that balances on a spreadsheet.
Not story: The story is what happens. The fantasy is who you get to be while it happens, and it can be fully present in a game with no story.
Seen in: Stardew Valley opens with you at a grey corporate desk, opening a letter from your grandfather that leaves you his farm. The fantasy is stated in that first minute, and the rest of the game keeps it.
Say it: "The fantasy is a heist crew who are always one step ahead. So when an alarm goes off, it should feel like part of the plan. Do not make it a fail screen."
https://gamecraft.dqnotes.com/game/player-fantasy

### Design pillars
Also: Pillars, Core pillars.
A short list, usually three to five words or phrases, stating what a game must be. Every feature is checked against them, and whatever does not serve one gets cut.
Teams have far more good ideas than they have months. Pillars turn "I don't like it" into "which pillar is that for?", and they decide what gets cut before the deadline does.
Not features: Features are things the game has. Pillars are what those features must add up to. "Crafting" is a feature; "you built this place yourself" could be a pillar.
Seen in: Hugo Martin, who led the design of id Software's Doom games, has described writing down three keywords and using them as guideposts for story, gameplay, systems and visuals alike.
Say it: "Our three pillars are fast, brutal and readable. Fishing is lovely and it serves none of them, so it is cut. Let's not reopen this unless a pillar changes."
https://gamecraft.dqnotes.com/game/design-pillars

## 2. Feel

Why one jump feels like a promise kept and another feels like wading: the words for control, weight and impact.

### Game feel
Also: Feel, Kinesthetics.
The physical sensation of controlling something in a game: how directly it answers you, how much it seems to weigh, how its impacts land. Steve Swink named its parts in his 2008 book Game Feel.
It is the first thing a player judges and the last thing a design document describes, so it is usually what goes missing when a game is built from a written plan.
Not juice: Feel is the whole sensation of control. Juice is the layer of feedback on top. A game can be juicy and still feel like mud.
Seen in: Super Mario Bros. teaches itself in seconds through feel alone: the run builds up, and the faster you are running, the further the jump carries you.
Say it: "The rules are fine but the game feel is dead. Fix control first: shorten acceleration, add a fast fall, then we will look at effects."
https://gamecraft.dqnotes.com/feel/game-feel

### Acceleration
Also: Ramp-up, Tight controls (when short).
The time a character takes to go from standing to top speed, and its partner deceleration, the time to stop again. Short feels tight. Long feels heavy or slippery.
These two numbers decide how much a player trusts their own landings. Long stopping time reads as ice, whether or not you meant it.
Not top speed: Top speed is how fast you end up. Acceleration is how long getting there takes, and it is the number people mean when they call controls tight.
Seen in: Celeste reaches full run speed almost at once, because its platforms demand precision. The characters in Super Mario Bros. take noticeably longer, which is why running there feels like building momentum.
Say it: "Run acceleration should reach top speed in about four frames and stop in about four. I want it tight, not slippery, because the platforms are narrow."
https://gamecraft.dqnotes.com/feel/acceleration

### Coyote time
Also: Ledge forgiveness, Late jump.
The short moment after running off a ledge when a jump still works. It is named after Wile E. Coyote, who hangs in the air until he looks down.
Players press jump when they believe they are on the edge, and they are often a few frames past it. Without coyote time, the game punishes a press they made on time.
Not jump buffering: Coyote time forgives a press that is too late, after leaving the ground. Buffering forgives one that is too early, before reaching it.
Seen in: Celeste has it, and its creator Maddy Thorson listed it first in a 2020 thread on the game's hidden forgiveness.
Say it: "Give the jump 100 ms of coyote time so a press just past the ledge still counts. Do not show it to the player."
https://gamecraft.dqnotes.com/feel/coyote-time

### Jump buffering
Also: Input buffer, Early jump.
The game remembers a jump pressed just before landing and performs it on the first frame you touch the ground.
Nobody can hit a one-frame window on purpose, again and again. Without a buffer, chained jumps feel as if the button sometimes ignores you.
Not coyote time: Buffering forgives a press that came too early. Coyote time forgives one that came too late.
Seen in: Celeste buffers the jump, so pressing just before landing makes you jump on the exact frame you land.
Say it: "Buffer jump presses for 120 ms before landing, so a chain of jumps never drops an input."
https://gamecraft.dqnotes.com/feel/jump-buffering

### Variable jump height
Also: Jump cut, Hold to jump higher.
A tap makes a short hop, a hold makes a full jump. In practice, letting go early just increases gravity for the rest of the rise.
One button gets a whole range of expression, and a designer can place both low ceilings and high ledges without adding a control.
Not fixed jump: A fixed jump is decided the moment you press, as in Castlevania, and you commit to the arc. A variable jump stays under your control while you rise.
Seen in: Super Mario Bros. made it famous: hold A longer and Mario goes higher. Classic Castlevania does the opposite, on purpose, and its committed jump is much of its character.
Say it: "Make jump height variable: if the player releases during the rise, multiply gravity by 2.5 until the apex."
https://gamecraft.dqnotes.com/feel/variable-jump-height

### Fast fall
Also: Fall gravity, Asymmetric gravity.
Stronger gravity on the way down than on the way up. A real thrown ball has an even arc; a good video game jump usually does not.
An even arc looks correct and feels like floating. A quicker fall gives the same jump weight and shortens the wait to act again.
Not floaty: Floaty is the complaint and fast fall is usually the fix. Floaty comes from too long in the air, especially on the way down.
Seen in: Super Mario Bros. already pulled Mario down harder after the top of his jump than on the way up, and most tight platformers since have done the same.
Say it: "Jump feels floaty. Keep the height at 150 px, take 0.36 s to reach it, and multiply gravity by 1.7 on the way down."
https://gamecraft.dqnotes.com/feel/fast-fall

### Apex hang
Also: Hang time, Jump peak gravity.
Gravity eases for a moment at the top of a jump, so the character hangs there briefly. It buys time to aim the landing.
The peak is where players make their decisions. A little extra time there makes the player feel more skilful and costs nothing in level design.
Not air control: Apex hang gives you more time in the air at the peak. Air control decides how much you can steer during that time.
Seen in: Celeste halves gravity at the peak of a held jump, which its creator says gives you more time to adjust and also looks pleasant.
Say it: "Halve gravity while vertical speed is under 70 px/s and jump is held, so the peak gives players a beat to line up the landing."
https://gamecraft.dqnotes.com/feel/apex-hang

### Air control
Also: Aerial steering.
How much you can change direction while in the air. Physics says none. Most games give nearly all of it.
Air control is a design choice about commitment. High control forgives. Low control makes every jump a decision you must live with.
Not double jump: Air control steers the jump you already made. A double jump makes another one.
Seen in: Mario can steer freely in the air. The original Prince of Persia and Castlevania allow almost nothing once you have left the ground, so each jump is a commitment.
Say it: "Set air control to 85% of ground acceleration. Players should be able to correct a jump but not reverse it instantly."
https://gamecraft.dqnotes.com/feel/air-control

### Corner correction
Also: Corner nudge, Ledge snap.
When your head clips the very corner of a block during a jump, the game slides you sideways around it rather than stopping you dead.
A jump that dies because of two pixels feels like a bug even when it is geometrically correct. Players never see the correction, only that the game is fair.
Not hitbox: A smaller hitbox makes every contact more lenient. Corner correction changes only what happens when you clip an edge.
Seen in: Celeste does this for jumps: if you bonk a corner, it wiggles you around it.
Say it: "Add corner correction: if a rising player overlaps a ceiling by 8 px or less horizontally, push them clear and keep their upward speed."
https://gamecraft.dqnotes.com/feel/corner-correction

### Squash and stretch
A shape flattens on impact and pulls long at speed, and keeps its volume while doing it. It is the first of Disney's twelve principles of animation.
It says what a thing is made of, and how hard it landed, without a single word. A square that squashes has weight and character.
Not scaling: Scaling changes volume. Squash and stretch trades width for height, so the thing feels like one lump of the same material.
Seen in: Celeste squashes Madeline's sprite on every landing and stretches it on every jump.
Say it: "On landing, squash the character to 70% height and 130% width and spring it back in 150 ms. Keep its area constant."
https://gamecraft.dqnotes.com/feel/squash-and-stretch

### Anticipation
Also: Wind-up, Telegraph (on enemies).
A small movement the other way before a big one: a crouch before a jump, a pulled-back arm before a punch. Another of Disney's twelve principles.
It tells the eye where to look before anything happens. On player actions it has to be very short, because it delays the response. On enemies it can be long, and then it is a warning.
Not input latency: Anticipation is motion you choose to show. Latency is delay you failed to remove. On a player's jump, too much anticipation feels exactly like latency.
Seen in: Dark Souls bosses raise a weapon for a long, readable beat before it falls. Mario, by contrast, leaves the ground on the very next frame, because he has to answer at once.
Say it: "Give the boss's slam half a second of anticipation so it reads, but give the player's jump none: they must leave the ground the frame they press."
https://gamecraft.dqnotes.com/feel/anticipation

### Follow-through
Also: Overlapping action, Secondary motion.
Loose parts keep moving after the main body stops: hair, capes, a tail, a scarf. They arrive late and settle slowly.
The eye reads the lag as physics and mass, and it costs almost nothing, since a rope simulation is cheap. It is one of the quickest ways to make a stiff character seem alive.
Not squash and stretch: Squash and stretch changes the body's own shape. Follow-through is what the parts attached to it do afterwards.
Seen in: The scarf in Journey trails behind the traveller, and its length also shows how far you can fly. Few details in games do two jobs so neatly.
Say it: "Give the character a scarf on a rope simulation: pinned at the neck, dragged by run speed, settling for half a second after they stop."
https://gamecraft.dqnotes.com/feel/follow-through

### Particles
Also: Particle effects, VFX, Dust and sparks.
Many small, short-lived sprites thrown off by an event, each with its own speed, fade and gravity: dust on a landing, sparks on a hit.
They answer the question "did something just touch something?" at a glance. They also carry direction and force: a landing that throws dust further felt harder.
Not screen shake: Particles happen at the point of contact, in the world. Shake happens to the whole camera, so it affects everything at once.
Seen in: Hollow Knight throws sparks and soul on each nail hit, and Nuclear Throne fills the screen with shells and debris on every shot.
Say it: "Spawn ten dust particles at the feet on landing, spreading sideways, scaled by fall speed, each fading out in about 0.4 s."
https://gamecraft.dqnotes.com/feel/particles

### Hit stop
Also: Hit pause, Hitlag, Freeze frames.
The game freezes for a few frames at the moment of impact. Fighting games have used it for decades, and it is among the strongest tools for making a blow feel heavy.
A hit lasting one frame barely registers. Freezing that frame gives the eye time to take it in, and makes contact feel as if the blow met real resistance.
Not slow motion: Hit stop is a complete stop lasting a few hundredths of a second. Slow motion is a slowdown lasting long enough for you to act during it.
Seen in: Street Fighter stops both fighters on every connecting hit, and the Smash Bros. games lengthen the freeze with the strength of the hit.
Say it: "Add 85 ms of hit stop on a stomp: freeze the player and enemy, keep particles and camera moving. Scale it up for heavier hits."
https://gamecraft.dqnotes.com/feel/hit-stop

### Screen shake
Also: Camera shake, Trauma.
The camera jolts when something big happens. Good shake is driven by a "trauma" value that decays, with the offset following trauma squared, so small hits stay small.
It spreads a single impact across the whole screen. Overused, it just makes everything unreadable and can make people feel sick, so provide a setting to reduce it.
Not hit stop: Hit stop stops time. Shake moves the camera. They work together: freeze first, then shake.
Seen in: Vlambeer's Nuclear Throne shakes on almost every shot, and Jan Willem Nijman's talk The Art of Screenshake made the technique famous.
Say it: "Add camera shake on heavy hits with a trauma model: decaying trauma, offset equal to trauma squared, smooth noise instead of random jumps. Offer a slider to reduce it."
https://gamecraft.dqnotes.com/feel/screen-shake

### Juice
Also: Game juice, Polish, Game feel effects.
All the extra feedback that makes a small action feel big: sound, particles, shake, flashes, bounce. The term spread through the 2012 talk Juice It or Lose It by Martin Jonasson and Petri Purho.
The same rules can feel lifeless or irresistible. Juice is cheap compared with new content, and it is often what makes people want to press the button again.
Not game feel: Juice is output: what the game does back at you. Feel also includes input and physics. You cannot juice your way out of a bad jump.
Seen in: Peggle plays Ode to Joy in slow motion on the final peg. It changes nothing in the rules, and it is what everyone remembers.
Say it: "The rules work but it is dry. Juice the match: particles from each cleared tile, rising pitch on combos, a small shake on big clears, and a score that counts up instead of snapping."
https://gamecraft.dqnotes.com/feel/juice

### Camera lookahead
Also: Camera lead, Forward focus.
The camera leans toward where you are heading, so you see more of what is coming than what you have passed.
A camera centred on the player shows as much behind as ahead, which is half wasted. Fast games become unfair if danger can arrive before you can see it.
Not dead zone: Lookahead moves the camera ahead of the player on purpose. A dead zone keeps the camera still while the player moves within a box.
Seen in: Super Mario World shifts the view toward the direction Mario is facing, so more of the screen shows what is ahead of him.
Say it: "Lead the camera 200 px in the direction of travel, ease it over a second, and do not snap when the player turns around."
https://gamecraft.dqnotes.com/feel/camera-lookahead

### Camera dead zone
Also: Camera window, Camera trap.
A box around the middle of the screen in which the player can move without the camera moving at all. The camera only follows once they push against the edge of the box.
Small movements should not shift the entire world. A locked camera makes tiny corrections feel like an earthquake.
Not stick dead zone: Same words, different thing. A stick dead zone ignores small movements of an analogue stick. A camera dead zone ignores small movements of the player.
Seen in: Most 2D platformers use one, from Super Mario Bros. on. Itay Keren's essay Scroll Back catalogues dozens of variations.
Say it: "Use a camera window 160 px wide centred on screen. The player can wander inside it and the camera will not move."
https://gamecraft.dqnotes.com/feel/dead-zone

### Platform snapping
Also: Vertical camera lock, Ground anchoring.
The camera follows up and down only when the player lands on something, not during every jump.
If the camera rises with every jump, the ground bounces under the player and they lose their reference for judging the landing.
Not dead zone: A dead zone ignores movement inside a box. Platform snapping ignores vertical movement until a landing, whatever its size.
Seen in: Super Mario World does this: the view holds steady through a jump and moves up only once Mario lands on a higher platform.
Say it: "Vertically, keep the camera still during jumps. Move it only when the player lands on a new height, easing over a third of a second."
https://gamecraft.dqnotes.com/feel/platform-snapping

### Input latency
Also: Input lag, Button-to-pixel delay.
The time between pressing a button and seeing the result on screen. Every stage adds some: the controller, the game's frames, the display.
People cannot name it, but they feel it. Above a tenth of a second, a game starts to feel as if you are suggesting things rather than doing them.
Not network latency: Input latency is on your own machine. Network latency is between machines. Both add up in an online game, but they have different fixes.
Seen in: Rhythm games such as Guitar Hero include a calibration screen to measure and offset this delay, because timing the music depends on it.
Say it: "Keep button-to-pixel latency under 50 ms. The character starts moving on the same frame as the press, and no animation can delay it."
https://gamecraft.dqnotes.com/feel/input-latency

### Frame rate
Also: FPS, Frames per second.
How many new pictures the game shows each second. Higher looks smoother and also shortens the delay before your press can show up.
A steady lower rate often feels better than an uneven higher one. Players notice a stutter more than they notice a number.
Not refresh rate: Frame rate is what the game produces. Refresh rate is what the screen can show. A 60 Hz display cannot show you 120 FPS.
Seen in: Fighting games such as Street Fighter lock to 60 FPS and treat that as law: each move's timing is measured in those frames.
Say it: "Hold 60 FPS during combat, even if that means dropping resolution. Frame pacing matters more here than sharpness."
https://gamecraft.dqnotes.com/feel/frame-rate

### Hitbox
Also: Hurtbox, Collision box.
The invisible shape the game actually checks for contact. Strictly, a hitbox deals damage and a hurtbox receives it. Neither has to match the art.
Players blame themselves for deaths that seem fair and blame the game for those that do not. Making the player's hurtbox a bit smaller than the sprite is an easy way to get more of the first kind.
Not hurtbox: A hitbox is the part that hits you. A hurtbox is the part of you that can be hit. People say hitbox for both.
Seen in: In shoot 'em ups such as Touhou Project, the player's hit area is a tiny dot at the centre of the ship, which is what makes dodging through dense bullets possible.
Say it: "Shrink the player's hurtbox to 70% of the sprite, and make enemy hitboxes a little smaller than their art, so near misses count as misses."
https://gamecraft.dqnotes.com/feel/hitbox

### Invincibility frames
Also: I-frames, Mercy invincibility.
A brief window in which you cannot be hurt. After taking damage it stops one mistake killing you twice. On a dodge, it is what makes rolling through an attack possible.
Without them, one touch in a crowd can drain a whole health bar in a blink. Players read the flicker that signals them without being taught.
Not hit stop: Hit stop stops the action on impact. I-frames let the action carry on while you are safe.
Seen in: Dark Souls gives its roll i-frames, which is why rolling into an attack beats rolling away from it. Mega Man flashes after a hit and passes through enemies.
Say it: "After a hit, give the player 1.2 s of invincibility and blink the sprite at 15 Hz so it is clear they are safe."
https://gamecraft.dqnotes.com/feel/invincibility-frames

### Frame data
Also: Startup, active, recovery.
Any attack can be measured in three stretches: startup before it can hit, active frames when it can, and recovery before you can act again.
These three numbers define what a move is for. A fast start makes it safe to use at any time, a long active time makes it easy to land, and a long recovery is what your enemy punishes.
Not damage: Damage is what a move does if it lands. Frame data decides whether it lands, and what happens to you if it misses.
Seen in: Fighting game communities publish frame data for every move, and many modern games such as Street Fighter 6 show it live in training mode.
Say it: "The heavy attack should be 18 frames of startup, 4 active, 30 of recovery: strong but punishable if it misses."
https://gamecraft.dqnotes.com/feel/frame-data

### Animation cancelling
Also: Cancel, Interrupt.
Cutting one action short by starting another, skipping its recovery. Some games forbid it, some allow it, some build everything around it.
It is a choice between weight and responsiveness. No cancelling makes each attack a commitment. Free cancelling makes combat fast and expressive.
Not frame data: Frame data describes a move on its own. Cancelling is the rule about whether a later move can cut it off.
Seen in: Street Fighter II's combos came partly from cancelling moves into specials, which the designers found during testing and kept. Devil May Cry is built on it.
Say it: "Let dodge cancel any attack's recovery but not its startup. Players can get out of a mistake, but not after they commit to a swing."
https://gamecraft.dqnotes.com/feel/animation-cancelling

### Aim assist
Also: Magnetism, Reticle friction, Auto-aim.
The game quietly helps you aim. Usually friction, where the cursor slows near a target, and magnetism, where it pulls gently towards one.
A thumbstick is far less precise than a mouse. Without assistance, console shooters would mostly be a contest against the controller, not the enemy.
Not auto-lock: Auto-lock picks the target and points you at it. Aim assist only leans on your aim, so you still feel you did it.
Seen in: Halo built its console shooting around sticky aiming and a magnetic pull on bullets, and most console shooters have followed its lead.
Say it: "Add aim assist on gamepad only: slow the reticle by 40% over a target, and bend bullets 2 degrees towards it. Never pull the camera without input."
https://gamecraft.dqnotes.com/feel/aim-assist

### Stick dead zone
Also: Analogue dead zone.
A small circle in the middle of a thumbstick's range where movement is ignored, so that worn sticks that do not centre perfectly do not make your character creep.
Too small and the character drifts by itself. Too big and small movements disappear. Shooters are often judged by this one setting.
Not camera dead zone: Same words, different thing. This one ignores small stick movements. A camera dead zone ignores small player movements.
Seen in: Competitive shooters such as Call of Duty let you tune stick dead zones yourself, because players with worn controllers need them bigger.
Say it: "Use a radial dead zone of 12% with a smooth response curve after it, and let players change it in settings."
https://gamecraft.dqnotes.com/feel/stick-dead-zone

### Haptics
Also: Rumble, Vibration, Force feedback.
Feedback you feel through the controller: from one spinning motor to precise actuators that can suggest texture and resistance.
It adds a sense no screen can reach, and it arrives faster than thought. Like shake, it can turn into noise if every event has it.
Not screen shake: Shake moves what you see. Haptics move what you are holding.
Seen in: Astro's Playroom was built to show off the DualSense: you feel rain, sand and the different tension of the triggers.
Say it: "Use short, sharp haptics for hits and a low continuous rumble while the engine runs. Keep the strongest pulse for the boss's slam."
https://gamecraft.dqnotes.com/feel/haptics

## 3. Challenge

Difficulty, pacing and fairness: how a game stays hard without ever feeling cheap.

### Difficulty curve
Also: Challenge curve, Ramp.
How hard a game is over time, drawn as a line. A good one starts low and climbs at about the speed the player is getting better.
If the line rises faster than the player, they quit in frustration. If it rises slower, they quit in boredom. Almost every complaint about difficulty is really a complaint about the shape of this line at one point.
Not learning curve: A difficulty curve is how much the game demands. A learning curve is how fast the player gets better. The art is keeping one just above the other.
Seen in: Tetris draws its curve with a single number: as the level rises, pieces fall faster. Slay the Spire lets players bend theirs, with 20 Ascension levels that each add one more difficulty rule.
Say it: "World 2 is too flat. Enemies should reach full speed by level 2-3, not 2-8. Draw me the curve we have and the one we want."
https://gamecraft.dqnotes.com/challenge/difficulty-curve

### Difficulty spike
Also: Difficulty wall, Brick wall.
One point where a game suddenly gets much harder than what came before, and then often eases off again.
A spike is where players quit. Those who could have finished the game hit a stretch that nothing prepared them for, and there is no reason to come back. A deliberate spike needs warning and a good reason.
Not skill gate: A spike is any sudden jump, usually an accident. A skill gate is a deliberate check on one skill the game has already been teaching.
Seen in: Battletoads on the NES is the famous case: its third level, the Turbo Tunnel, asks for split-second memorised reactions on a speeder bike, and many players never saw what came after it.
Say it: "Level 6 is a spike: playtesters pass 1 to 5 in two tries each, then take twenty on 6. Move the laser section to level 9 and add a practice room before it."
https://gamecraft.dqnotes.com/challenge/difficulty-spike

### Learning curve
Also: Skill curve, Onboarding slope.
How quickly a player gets better at a game with practice. People say "steep" to mean hard to learn, though on a real chart a steep curve means learning fast.
Early progress is what keeps people playing. If a game takes hours to make any sense, most will not stay long enough to find out whether it is good.
Not difficulty curve: A learning curve is about the player. A difficulty curve is about the game. A hard game can still be learned quickly.
Seen in: Dwarf Fortress is known for one of the longest in games. Its community took "Losing is fun!" as a motto, because you will lose many times before you understand why.
Say it: "New players need one input in the first minute, two by minute five, and the full moveset by the end of the first hour. Don't show the crafting menu before then."
https://gamecraft.dqnotes.com/challenge/learning-curve

### Skill floor and ceiling
Also: Low floor, high ceiling, Easy to learn, hard to master.
The floor is the least skill you need to do anything at all. The ceiling is the most skill the game can reward. The two can be moved separately.
A high floor keeps beginners out. A low ceiling makes experts leave. Games with a long life tend to have a low floor and a high ceiling, and they do it by rewarding precision without demanding it.
Not difficulty: Difficulty is how hard a given moment is. Floor and ceiling are the range of skill that the game can tell apart.
Seen in: In Rocket League, anyone can drive into a ball and feel useful within minutes. At the top, players fly through the air on boost and steer the ball with their car, using the same few controls.
Say it: "Keep the floor low: any press in a 300 ms window lands the hit. Raise the ceiling: a press inside the middle 30 ms does three times the damage. Newcomers should never know that second number exists."
https://gamecraft.dqnotes.com/challenge/skill-floor-and-ceiling

### Teach, test, twist
Also: Kishōtenketsu in levels, Introduce, develop, twist.
A way of building a level around one idea: show it somewhere safe, ask for it where failure counts, then change it so that what you learned has to bend. Then move on.
Players learn from doing, without text, and the game never asks for something it has not shown. Each idea also gets used up and left behind before it grows stale.
Not tutorial: A tutorial stops the game to explain itself. Teach, test, twist does the teaching inside the level, so it never looks like teaching.
Seen in: Koichi Hayashida, director of Super Mario 3D Land, has described building levels in four steps, drawing on the four-part story form kishōtenketsu: introduce an idea, develop it, twist it, conclude. He said 3D Land is where he grew strict about it.
Say it: "Give the bounce pad three rooms. Room one: a pad in front of a wall, with no way to die. Room two: the same pad before a pit. Room three: two pads in a row, and you press in mid-air."
https://gamecraft.dqnotes.com/challenge/teach-test-twist

### Pacing
Also: Rhythm of play, Beats.
The order and length of what you ask of the player: fight, rest, explore, talk, fight again. It is about what comes after what, more than how hard any one part is.
Non-stop action wears people out and makes each fight feel the same. Long quiet stretches lose them. A strong moment gets its strength from what came just before it.
Not difficulty curve: A difficulty curve is how hard things are. Pacing is what you do, and when. An easy game can be badly paced, and a hard one beautifully.
Seen in: Doom (2016) alternates sealed arena fights with quieter corridors of exploring and pick-ups, so each new arena lands as an event.
Say it: "After each boss, give the player a safe room with a save point and no enemies for at least a minute. Then one short fight to warm up, and only then the next big setpiece."
https://gamecraft.dqnotes.com/challenge/pacing

### Sawtooth pacing
Also: Sawtooth difficulty curve, Peaks and valleys.
Difficulty that climbs, drops, then climbs higher, like the teeth of a saw. Each hard stretch ends with an easy one, and the overall trend keeps rising.
A straight ramp means ever-growing strain. A drop after each peak gives relief and a moment of feeling strong, and it is often the place to introduce the next new idea safely.
Not difficulty spike: A spike is one unplanned jump with no build-up. A sawtooth is a pattern of planned rises and planned rests.
Seen in: Where's My Water? works this way: each set of levels gets harder until you finish it, then the next set eases off before climbing again.
Say it: "Use a sawtooth across the world: each of the four levels harder than the last, then drop back to about 60% for the first level of the next world and introduce the new enemy there."
https://gamecraft.dqnotes.com/challenge/sawtooth-pacing

### Intensity curve
Also: Drama curve, Encounter pacing.
How intense the action feels from moment to moment, drawn as a line. Designers shape it with peaks and valleys, and some games measure it live and react.
Intensity is what players will remember as the shape of the game. A flat line feels dull even when it is dangerous, and a line that never falls leaves people numb.
Not difficulty curve: Difficulty is how hard it is to succeed. Intensity is how much is going on. A swarm of weak enemies can be intense and easy.
Seen in: Left 4 Dead's AI Director tracks each survivor's intensity. It builds pressure to a peak, holds it for a short time, then backs off into a real lull. Valve's Mike Booth called this adaptive dramatic pacing.
Say it: "Drive enemy spawns from a director: build until player intensity reaches 80%, hold for four seconds, then send nothing for at least ten seconds, whatever the level says."
https://gamecraft.dqnotes.com/challenge/intensity-curve

### Tension and release
Also: Build and release, Relief.
Letting worry build, then letting it go. A long held threat followed by safety hits harder than either alone, and the relief is often what players remember the most.
Tension needs an end to mean anything. If a horror game is always frightening, players get used to it within minutes and stop being afraid.
Not difficulty: Tension is how much you dread what might happen. A scene can be hugely tense with almost no real danger.
Seen in: The early Resident Evil games have save rooms: safe rooms with a typewriter and calm music. No enemies come in, and hearing that music after a long corridor is a physical relief.
Say it: "After the chase, cut the music completely and bring the lights up. Put the save point right there. Don't start the next threat until the player has left that room by choice."
https://gamecraft.dqnotes.com/challenge/tension-and-release

### Telegraphing
Also: Tell, Wind-up, Attack cue.
Showing what is about to happen before it does, in time for the player to respond: a raised arm, a glow, a sound. It turns an attack into a question the player can answer.
A blow you could have read is your fault, and you will try again. A blow that came from nowhere is the game's fault, and you may not. Reactions take about a quarter of a second, so a tell has to be longer than that.
Not anticipation: Anticipation is the animation principle behind it: the pull-back before a move. Telegraphing is its job in play: warning you, with enough time, which answer you need.
Seen in: In Punch-Out!! on the Wii, Glass Joe steps back before a jab and crouches before an uppercut. Each opponent is a set of tells to learn, and reading them is the whole game.
Say it: "Give the slam 600 ms of wind-up with a rising sound, and make it look different from the sweep. A player should know which one is coming just from the first frames."
https://gamecraft.dqnotes.com/challenge/telegraphing

### Fair failure
Also: Fair death, Earned failure.
Failing in a way the player accepts as their own doing: they saw it coming, they know what went wrong, and they know what to try next. Whether a death feels fair matters more than how often it happens.
Players blame themselves for deaths that seem fair and the game for those that do not. It is why very hard games can have devoted followings, while easy ones with cheap shots get refunds.
Not easy: Fair does not mean gentle. A game can kill you a thousand times and be fair each time, as long as each death has a cause you can name.
Seen in: Super Meat Boy kills you constantly and shows every attempt at once when you finish a level, a swarm of your own earlier selves. You can see that each death was yours.
Say it: "Nothing can kill the player unless it has been visible on screen for at least a second. Replace the off-screen spike drops with ones that shake first."
https://gamecraft.dqnotes.com/challenge/fair-failure

### Cost of failure
Also: Death penalty, Failure state cost, Setback.
What you lose when you fail: seconds, progress, items, or the whole run. It is a separate dial from how hard the game is.
It decides how the player feels about trying. Cheap failure encourages daring and lets a game be brutally hard. Expensive failure makes every step tense, but spends the player's time repeating things they can already do.
Not difficulty: Difficulty is how likely you are to fail. Cost is what failing takes from you. Super Meat Boy is high on the first and low on the second.
Seen in: Super Meat Boy puts you back at the start of a short level in about a second. In Dark Souls, you drop your souls where you die, and you lose them for good if you die again before getting back to them.
Say it: "Respawn instantly at the start of the room, with no reload screen and no walk back. The room can be much harder if a retry costs one second."
https://gamecraft.dqnotes.com/challenge/cost-of-failure

### Permadeath
Also: Permanent death, Hardcore mode, Ironman.
When a character dies, they are gone for good, with no reloading. It may end the whole run or only remove that one character.
It gives each choice real weight, and it changes which choices are correct. It also throws away hours of a player's time in a moment, so it usually needs short runs or a big reason to accept it.
Not game over: Game over lets you pick up from your last save. Permadeath means there is no save to go back to.
Seen in: Rogue (1980) ended your run for good at each death, and gave roguelikes their name. In XCOM, a soldier you have named and levelled for hours can die on one bad shot and stay dead, and Ironman mode takes away your save-scumming as well.
Say it: "Named soldiers can die for good, but losing one costs the player a soldier, never the campaign. And give an optional Ironman mode with one save slot for those who want it."
https://gamecraft.dqnotes.com/challenge/permadeath

### Risk and reward
Also: Push your luck, Risk-reward.
Giving the player a safe option and a dangerous one that pays better. The risk has to be real and the reward has to be worth it.
It makes the player do the choosing of how hard the game is, moment to moment. If the danger never pays, nobody takes it. If it always pays, it was never a risk.
Not randomness: Randomness is what happens to you. Risk and reward is a choice you make with the odds in view.
Seen in: Gears of War's active reload lets you tap during a reload: hit the sweet spot for a fast reload and stronger shots, or miss and jam the gun. Slay the Spire marks tougher elite fights on its map, and they pay out relics.
Say it: "Put the best loot on the side route past the sniper. The main path should be safe and give about half as much. I want players to look at their health before they choose."
https://gamecraft.dqnotes.com/challenge/risk-and-reward

### Time pressure
Also: Timer, Countdown, Clock.
Limiting how long a player has to act. It changes the kind of thinking a task takes, from careful working out to quick judgement.
It makes easy tasks tense without changing them, and stops players from stalling. Too much of it, though, punishes the slow and turns thinking into panic, so many games let you turn it off.
Not difficulty: Difficulty asks whether you can solve it. Time pressure asks whether you can solve it now.
Seen in: Super Mario Bros. has a timer that hardly ever kills anyone, but its music speeds up when under 100 is left, and the whole mood changes. Prince of Persia gives you 60 minutes to finish the whole game.
Say it: "Give each room 30 seconds. At 10 seconds, speed the music up and flash the clock, but do not punish until zero. I want them hurrying long before it is really a danger."
https://gamecraft.dqnotes.com/challenge/time-pressure

### Execution versus knowledge
Also: Skill versus knowledge, Dexterity versus puzzle.
Two kinds of challenge. Execution is about being able to do it: timing, precision, nerve. Knowledge is about knowing what to do: the trick, the route, the answer.
They fail in different ways. A knowledge challenge is over the moment you learn the answer, and a guide can spoil it. An execution challenge survives spoilers, but it gets stuck on hands that can't do it. Most difficulty complaints are really about which kind the player wanted.
Not difficulty: Difficulty says how hard something is. This says what sort of hard it is, which tells you whether practice or information would help.
Seen in: In Outer Wilds, you have your whole ship from the first minute, with no upgrades at all. Nothing carries over between time loops but what you have learned. Super Meat Boy is the other pole: you can see the route, and doing it is the problem.
Say it: "The final room should test execution only. Show the whole route up front with arrows, so that dying there is never about not knowing what to do."
https://gamecraft.dqnotes.com/challenge/execution-versus-knowledge

### Skill gate
Also: Skill check, Gatekeeper.
A deliberate point you cannot pass until you can do a certain thing. It stops you moving ahead without a skill that everything later will demand.
Without one, a player can carry a bad habit deep into a game and meet a brutal wall where nothing explains the problem. A gate moves that lesson to where it is cheap to learn.
Not difficulty spike: A spike is a sudden jump in difficulty. A skill gate tests one named skill the game has already taught you.
Seen in: Sekiro's early duel with Genichiro Ashina is hard to get through by dodging and hacking. It pushes you to learn to deflect, which the rest of the game is built on.
Say it: "Before the second boss, add a guard who takes no damage except after a parry. No text. If you get past him you can parry, and you will need it in the next room."
https://gamecraft.dqnotes.com/challenge/skill-gate

### Boss
Also: Boss fight, End-of-level boss.
A big set-piece enemy that ends a stretch of the game and tests what the stretch taught you. Usually it has phases: as its health falls, it changes how it fights.
It gives a section of the game a shape and a finish, and asks you to prove what you have learned. A boss that is only a tall health bar teaches nothing and just takes a long time.
Not tough enemy: A tough enemy is a harder version of what you have been fighting. A boss has its own rules, its own arena, and usually asks a new question.
Seen in: Super Mario Bros. ends every castle with Bowser, but the first seven are impostors in disguise. Only World 8-4 has the real one. Shadow of the Colossus is nothing but boss fights: each colossus is a climbing puzzle you solve on its body.
Say it: "Three phases: at 66% health, it adds the high swing the player learned in the last room. At 33% the tells get a third shorter. Below that, no new moves."
https://gamecraft.dqnotes.com/challenge/boss

### Difficulty modes
Also: Difficulty settings, Easy, normal, hard.
Versions of the same game tuned for different players, chosen by the player, usually at the start. They change numbers such as enemy health, damage and how many enemies there are.
Players differ too much for one setting to suit them all. The catch: new players choose before they know the game, and people can be too proud to pick easy, so naming and the option to change later both matter.
Not assist mode: Difficulty modes bundle everything into one choice. An assist mode lets you change one thing at a time, for your own reason.
Seen in: Doom (1993) sells its modes in words, from "I'm Too Young to Die" to "Nightmare!". Wolfenstein 3D's easiest mode, "Can I Play, Daddy?", makes fun of you for picking it, which is one way to get people to avoid the mode they need.
Say it: "Three modes: Story has half damage and no enemy reinforcements. Normal is the game as designed. Hard adds enemies but never shortens telegraphs, so it stays fair. Let players switch at any checkpoint."
https://gamecraft.dqnotes.com/challenge/difficulty-modes

### Assist mode
Also: Assist options, Accessibility settings.
A menu of separate settings that make specific things easier, each your own choice: slower game speed, extra jumps, no damage. You keep the rest of the game as it was designed.
People are stuck on a game for different reasons: one has trouble with reactions, another with their hands, another has only got an hour. One slider can't address them all, and treating the choice as normal keeps people playing rather than feeling shamed.
Not difficulty modes: A difficulty mode says this is who you are. An assist option says this is the one thing in your way.
Seen in: Celeste's Assist Mode lets you lower game speed to as little as 50% in steps of 10%, grant infinite stamina, add air dashes, and turn off death. You can switch it on and off at any time.
Say it: "Add separate toggles: game speed from 50% to 100%, one extra jump, and no fall damage. Don't lock achievements, and don't make a comment when someone turns them on."
https://gamecraft.dqnotes.com/challenge/assist-mode

### Invisible assistance
Also: Hidden mechanics, Player-favouring tricks, Generous design.
Help the game gives you without saying so: a hit area smaller than it looks, a health bar that drains more slowly near the end, a jump that works just after the ledge.
Near misses and last-second survivals are the best moments in a game, and they happen far more often when the odds are quietly bent. Players feel skilled, as long as they never catch you doing it.
Not assist mode: An assist mode is help you ask for and can see. Invisible assistance is on for everyone, and it only works if no one notices it.
Seen in: Developers shared these tricks in a thread started by designer Jennifer Scheurle in 2017. One she named herself: Assassin's Creed and Doom both count the last sliver of a health bar as more hit points than it looks, so you often feel you barely survived.
Say it: "Make the last 20% of the health bar take half damage, and shrink the player's hit area to 60% of the sprite. Never show either. Then test whether anyone can tell."
https://gamecraft.dqnotes.com/challenge/invisible-assistance

### Dynamic difficulty adjustment
Also: DDA, Adaptive difficulty.
The game watches how you are doing and changes its difficulty while you play, usually without telling you. Do badly and it eases off; do well and it pushes harder.
It can hold a far wider range of players in the band that is neither boring nor crushing. The risks: players who notice may feel patronised, and playing badly on purpose can become the best strategy.
Not rubber-banding: DDA adjusts the challenge to the player over time. Rubber-banding is a kind of it that specifically keeps racers or rivals close to each other.
Seen in: Naughty Dog built this into Crash Bandicoot, and Andy Gavin later wrote about it: fail repeatedly and you get extra masks and lives, and even slower boulders, with the aim of helping weak players without changing the game for strong ones. God Hand does it in the open, with a level meter on screen.
Say it: "Track deaths over the last ten minutes. After three deaths in a room, spawn 20% fewer enemies and drop one extra health pack. Raise it again slowly, and never tell the player."
https://gamecraft.dqnotes.com/challenge/dynamic-difficulty-adjustment

### Rubber-banding
Also: Rubber band AI, Catch-up mechanic.
Pulling competitors back together: those in front are held back and those behind are helped along, as if everyone were joined by elastic.
It keeps a race close and gives everyone hope to the very end, which suits parties. The cost is that being good feels punished, and skill counts for less when the leader is always within reach.
Not dynamic difficulty adjustment: DDA tunes the game to one player's skill. Rubber-banding reacts to the gap between competitors, in that race, right now.
Seen in: Mario Kart gives better items to racers further back, and the spiny shell, which first appeared in Mario Kart 64, flies straight at whoever is in first place.
Say it: "Give cars in last place up to 30% more top speed based on their gap to the leader, and cut it to nothing within 100 metres of the finish. Don't slow the leader down, because players notice that."
https://gamecraft.dqnotes.com/challenge/rubber-banding

### Grinding
Also: Farming, Level grinding.
Repeating easy tasks, not for their own sake, but to build up the experience or items that let you progress. The player has mastered the task and is just waiting for a number.
A little gives a sense of steady growth and lets a stuck player trade time for power. A lot means players are paying with their hours for something that is no longer a challenge, which they will start to resent.
Not practice: In practice, repetition makes you better. In a grind, it makes your character better, and you stay the same.
Seen in: Pokémon players know it from the long grass: battling wild Pokémon, again and again, until the team is strong enough for the next Gym Leader. Disgaea embraces it, with level caps in the thousands.
Say it: "Reaching the third boss should take at most two extra fights on the way. If playtesters stop to grind, the XP curve is too steep between levels 8 and 10."
https://gamecraft.dqnotes.com/challenge/grinding

### Mastery
Also: Skill mastery, Competence.
Getting really good at something, and being able to see that you have. The challenge stays still and you are the one who changes.
It is the most lasting reward a game can offer, and it costs nothing to supply. Games that can support years of improvement keep their players for years, provided improvement is something they can see.
Not progression: Progression gives your character more. Mastery gives you more. Take every upgrade away and your mastery is still there.
Seen in: In December 2023, a 13-year-old called Willis Gibson played the 1989 NES Tetris until it crashed at level 157, the first person known to do so. The game had stayed exactly the same for 34 years, and the players had kept getting better.
Say it: "Show the player a ghost of their best run from the second attempt on, and their time on every room. Don't add upgrades here: getting faster is the reward."
https://gamecraft.dqnotes.com/challenge/mastery

## 4. Space

Level design: how a place teaches, leads and surprises without saying a word.

### Critical path
Also: Golden path, Main path.
The shortest route through a level or game that still gets you to the end, touching only what is required. The golden path is its richer cousin: the route designers hope most players take, past the things worth seeing.
It tells you what every player is guaranteed to see, which is where the things they must understand have to go. Anything off it might never be seen by half your players.
Not golden path: Studios use the two loosely. Where they differ, the critical path is the least you can do, and the golden path is the route the designer intends. Speedrunners take the first; a first-time player should find the second.
Seen in: The warp zones in Super Mario Bros. shrink its critical path to eight levels of the thirty-two: 1-1, 1-2, 4-1, 4-2 and all of world 8. Everything else is optional, which is why speedruns take that route.
Say it: "Put the wall-jump tutorial on the critical path, in room three, because every player passes it. The hidden armoury can stay off it."
https://gamecraft.dqnotes.com/space/critical-path

### Landmark
Also: Orientation point, Navigational landmark.
Something big and unique that can be seen from far away, so players always know where they are relative to it. Lost people navigate by these in cities, and players do the same.
Without one, every street looks alike and a player has only the map screen to tell them where they are. With one, they build a picture of the place in their head.
Not weenie: A landmark tells you where you are. A weenie makes you want to go there. One building can do both, but a mountain you can never reach is only a landmark.
Seen in: The Citadel in Half-Life 2 towers over City 17 and stays in view from far across the city, so you always know which way the heart of the place lies.
Say it: "Everyone gets lost in the old town. Add a clock tower at the north end, tall enough to clear the rooftops from anywhere in the district."
https://gamecraft.dqnotes.com/space/landmark

### Weenie
Also: Wienie, Visual magnet.
A big, tempting sight placed at the end of a route, so people walk towards it without being told. Walt Disney's term: Sleeping Beauty Castle at the end of Main Street is the first and most famous one.
A sign says go this way and players resent it. A weenie makes going that way their own idea. The story usually told is that Walt named it after the sausage he used to get his dog to follow him.
Not signposting: Signposting tells you which way is correct. A weenie makes you curious enough to walk there. One is information and the other is desire.
Seen in: The mountain with the split peak in Journey is on the horizon from the first dunes, and that glimpse is all the instruction you get. Scott Rogers brought the word to game designers in a 2009 GDC talk on what games can learn from Disneyland.
Say it: "The player needs a reason to cross this valley. Put a lit tower at the far side, centred in the view as they leave the cave, and take the waypoint marker away."
https://gamecraft.dqnotes.com/space/weenie

### Breadcrumbing
Also: Breadcrumbs, Breadcrumb trail.
A trail of small rewards or items, each in sight of the last, that leads the player along a route. The player only ever decides to get the next one.
It guides without saying so, and it works in the dark or in a large space where a single goal would be out of sight. Space them further apart than the player can see and the trail simply stops being one.
Not signposting: Breadcrumbs pay you for walking the route, again and again. A signpost pays nothing and tells you once.
Seen in: Super Mario games lay lines of coins along a route, and arc them through the air to show the shape of a jump you might not have thought to try.
Say it: "Players miss the upper route. Run a line of six coins up the ramp, no more than two screens' width apart, and arc the last three over the gap."
https://gamecraft.dqnotes.com/space/breadcrumbing

### Signposting
Also: Yellow paint, Visual language, Affordance marking.
Marking what matters so that it stands out from everything else: a colour on every climbable ledge, a light over the correct door, one kind of crack on every wall that breaks. Players learn the code once and read it for the rest of the game.
Detailed art makes everything look equally important. Without a visual code, players search every surface by trying it, and that reads as being lost or stupid.
Not breadcrumbing: Signposting marks a thing so it can be found. Breadcrumbs lay a route you follow one step at a time.
Seen in: Mirror's Edge paints the way ahead in red with its Runner Vision, so you can read a route across rooftops while at full sprint. The remake of Resident Evil 4 set off an argument in 2023 over its yellow paint on crates and ledges.
Say it: "Paint every climbable ledge with the same yellow strip, 10 cm wide, and use that yellow on nothing else in the game."
https://gamecraft.dqnotes.com/space/signposting

### Funnelling
Also: Funneling, Player flow, Soft guidance.
Shaping a space so that people drift towards one place without noticing: walls that lean inward, a road that narrows, a floor that slopes down. They believe they chose the route.
A square room with a door in one wall makes players walk its edges looking for the way on. Angle the walls towards the door and most of them are through it before they have had to think.
Not choke point: Funnelling is about guiding movement. A choke point is about what happens when a fight meets a narrow place. A funnel can end at a choke point, and often does.
Seen in: Left 4 Dead runs its survivors down streets and through corridors with one-way drops, so a group stays together and keeps moving forward through a level that looks like an open city.
Say it: "Players stall in the courtyard. Angle both side walls in by about 30 degrees towards the gate, and light only the gate end."
https://gamecraft.dqnotes.com/space/funnelling

### Gating
Also: Progression gate, Hard gate, Soft gate.
Anything that stops the player going further until they have done something. A hard gate cannot be passed at all. A soft gate can be, but it is dangerous or expensive to try early.
Gates decide the order in which the game gets experienced, so abilities and story arrive in a sequence you can design for. Too many, or too obvious, and an open world feels like a corridor with scenery.
Not lock and key: Gating is the general idea of holding the player back. Lock and key is one way of doing it. A boss, a skill check or a strong enemy can be a gate with no key anywhere.
Seen in: In Metroid a low tunnel near the start of the game stops you until you collect the Morph Ball. Breath of the Wild keeps you on the Great Plateau until you earn the paraglider, then gates almost nothing else.
Say it: "Put a hard gate at the end of the first region, a cliff you need the grappling hook for, so everyone gets the same first hour. After that, use soft gates only: strong enemies that a good player can sneak past."
https://gamecraft.dqnotes.com/space/gating

### Lock and key
Also: Key and door, Ability gate.
A gate that one specific thing opens. The key may be an actual key, or an ability that works as one: a hookshot, a double jump, a bomb.
It turns a map into a puzzle about order, and shows the player a door now that will pay off later. The designer's job is to make sure no key ends up behind its own lock.
Not soft lock: Lock and key is a design tool you use on purpose. A soft lock is the failure you get when the player can no longer reach a key they need.
Seen in: Doom hides red, blue and yellow keys around each level and colours their doors to match. Metroid-style games use abilities the same way, so every new power reopens the map you have already walked.
Say it: "Show the player the red door in the first minute, then put its key at the far end of the east wing, so the second half of the level has a purpose."
https://gamecraft.dqnotes.com/space/lock-and-key

### Soft lock
Also: Softlock, Unwinnable state, Dead end.
The game is still running and you can still move, but you can no longer finish it, and it does not tell you. A hard lock is the frozen version, where nothing responds at all.
A player may lose hours before they find out, and all they can do is reload an old save. Few things break trust as completely. Anything that can be spent, destroyed or left behind has to be checked for it.
Not hard lock: A hard lock is a crash: the program has stopped. In a soft lock the program is fine and the player is the one who has been stranded.
Seen in: Sierra adventures such as the King's Quest series could strand you hours after you walked past an item you could no longer go back for. Ron Gilbert's 1989 essay Why Adventure Games Suck told designers never to require such an item.
Say it: "Small keys are consumed, so check the dungeon can't be stranded: count doors against keys in every order, and if a player can run out, put a spare in a chest that refills."
https://gamecraft.dqnotes.com/space/soft-lock

### Teaching through level design
Also: Show, don't tell, Silent tutorial, World 1-1.
Arranging the first spaces so that the player works out the rules by bumping into them, safely, with no text. The obstacle comes first, in a place where failing costs nothing, and the lesson arrives as an idea of their own.
People skip instructions and remember what they worked out for themselves. A level that teaches also respects the player, who gets to feel clever rather than instructed.
Not tutorial: A tutorial explains the rule and then asks you to do it. Teaching through the level asks you to do it, and you only find out afterwards that you learned a rule.
Seen in: Super Mario Bros. opens with empty ground, then a Goomba walking at you, then blocks above your head. Miyamoto and Tezuka have explained that World 1-1 was built so players could gradually and naturally understand what they were doing.
Say it: "Before the first pit, put a wall one jump high with no danger around it, so they can only get past by pressing jump. Then the pit is a test of something they already know."
https://gamecraft.dqnotes.com/space/teaching-through-level-design

### Kishotenketsu
Also: Kishōtenketsu, Four-step level design, Ki-sho-ten-ketsu.
A four-part structure from Chinese poetry and Japanese four-panel comics: introduce an idea, develop it, twist it, conclude. There is no conflict required. In a level, one mechanic gets those four beats and then the level ends.
It stops levels being a mechanic that simply gets harder until you are tired of it. The twist makes you think about something you thought you knew, and ending right after it means the idea never outstays its welcome.
Not difficulty ramp: A ramp asks the same question faster. Kishotenketsu changes the question at the third beat, and may even get easier there.
Seen in: Koichi Hayashida, director of Super Mario 3D Land, described this structure for its levels in a 2012 Gamasutra interview: learn a mechanic, use it in a harder scenario, something crazy makes you rethink it, then show mastery. He likened it to four-panel comics.
Say it: "Give the moving-platform level four beats: one platform over safe ground, then over a pit, then two that move on opposite beats so you have to wait instead of hurry, then one quick victory lap. Then end it."
https://gamecraft.dqnotes.com/space/kishotenketsu

### Player metrics
Also: Metrics, Character metrics, Level metrics.
The measurements of what the character can do: jump height, longest jump, run speed, how high a wall they can climb, how wide a door they fit through. Every level is built out of these numbers.
Decide them late or change them after levels exist and every gap in the game is wrong. Written down, they let you build a level before there is any art, and know it can be crossed.
Not analytics: Same word, different job. Player metrics here are the character's measurements. Analytics, also called metrics, are data about how people played.
Seen in: Super Mario Bros. was planned on graph paper in a grid of blocks, so a designer can count a gap in blocks and know whether a running jump will clear it.
Say it: "Freeze the metrics this week: jump 3 m high, 6 m long at a run, doors 2 m wide, cover 1.2 m tall. Every room we block out from now on uses these."
https://gamecraft.dqnotes.com/space/player-metrics

### Sightline
Also: Line of sight, LOS, Sight lane.
How far and where a player can see, and be seen, from a given spot. In a shooter, a sightline is also a line of fire: if I can see you, I can hit you.
A long clear line favours one weapon and one kind of player. A level with none feels cramped. Which sightlines exist is the most important decision in a multiplayer map, and players find any you did not mean to leave in.
Not draw distance: Draw distance is how far the engine bothers to render. A sightline is about what the walls allow. A wall ends a sightline no matter how far the engine could draw.
Seen in: On Dust II in Counter-Strike, a sniper can see straight down mid through the double doors, and crossing it is a known risk. Teams throw a smoke grenade there to cut the line.
Say it: "That corridor gives the sniper rifle 80 m of clear line right into spawn. Add a pillar a third of the way along, so nothing has more than 30 m."
https://gamecraft.dqnotes.com/space/sightline

### Choke point
Also: Chokepoint, Bottleneck.
A narrow place every attacker must pass through, so only a few of them can fight at once. Numbers stop counting: the side in the narrow place fights the front rank, not the army.
It makes a small force strong, which gives defenders something to hold and attackers a problem to solve. A map with no alternative to one becomes a stalemate that nobody enjoys.
Not funnelling: Funnelling steers people, usually without their noticing. A choke point decides a fight, and everyone involved knows it is there.
Seen in: StarCraft bases sit at the top of a narrow ramp, so Terran players wall it off with buildings early on, and a handful of units can hold the entrance against far more.
Say it: "Attackers lose every round at the bridge. It is a two-wide choke with no way round, so widen it to five or add a second route under it."
https://gamecraft.dqnotes.com/space/choke-point

### Cover
Also: Hard cover, Half cover, Cover system.
Anything that stops bullets and that you can stand behind. Its height and spacing decide the rhythm of a fight: pop out, shoot, get back, move to the next piece.
Without it, a shooter is a contest of who saw whom first. Cover gives the player a place to think and turns a fight into a series of decisions about where to stand next.
Not concealment: Concealment hides you and cover protects you. Tall grass hides you and stops nothing, which is a fine thing to make a player learn the hard way.
Seen in: Gears of War (2006) built its shooting around snapping into cover and sprinting between pieces of it, and a generation of third-person shooters copied the idea.
Say it: "Space the waist-high cover about 8 m apart, a two-second sprint, so moving up always means a real moment in the open."
https://gamecraft.dqnotes.com/space/cover

### Flanking route
Also: Flank, Side route, Third lane.
A second way into a fight that comes at the enemy from the side or behind. It exists so that a defender can never watch everything at once.
It breaks stalemates: strong positions stop being unbeatable, and they need watching. It also makes you look over your shoulder, which is much of what makes a shooter tense.
Not shortcut: A shortcut saves time for someone going the same way. A flanking route may take longer, because what it buys is the angle you arrive from.
Seen in: Dust II in Counter-Strike is three lanes joined at the middle: a team holding one bombsite can be hit from mid as well as from the long route, so you can't just stare down one lane.
Say it: "The hill is too strong: nothing reaches it except up the front. Add a side tunnel that comes out behind the top, longer by about ten seconds of running."
https://gamecraft.dqnotes.com/space/flanking-route

### Verticality
Also: High ground, Vertical level design.
Using height as part of the layout, so that where you stand up and down matters as much as where you stand across. Whoever is above sees more and has less exposed to them.
A flat map offers two dimensions of decisions and a tall one offers three. High ground gives you something to fight for, and a dangerous route up can pay for itself with what you can see from there.
Not multiple floors: Floors stacked on top of each other with no way to see or shoot between them are several flat levels. Verticality means the heights can affect each other.
Seen in: Batman: Arkham Asylum's predator rooms put gargoyles above armed guards. From a perch you can see them all and pick who to take next, and on the floor you would last seconds.
Say it: "Give the arena a balcony 4 m up, reached by one exposed staircase. It should see two thirds of the floor, and anyone up there can be seen from it too."
https://gamecraft.dqnotes.com/space/verticality

### Arena
Also: Combat space, Kill box, Lock-in.
A space built for one fight: big enough to move around in, usually with exits closed until it ends. The room itself sets the terms, so you can't solve it by backing into the last corridor.
Enemies can't pressure a player who has found a safe doorway. Close the doors and players have to use the whole room, which is where the fight you actually designed takes place.
Not corridor fight: In a corridor, enemies come from one direction, one at a time. In an arena they come from all sides, and your position changes all the time.
Seen in: Doom (2016) locks you into arenas until every demon is dead, and health and ammunition come from killing them, so standing still is the worst thing you can do.
Say it: "Players are camping the entrance and shooting demons as they come through. Seal that door when the first wave spawns, and put health at the far side of the room."
https://gamecraft.dqnotes.com/space/arena

### Encounter design
Also: Combat encounter, Enemy mix, Wave design.
Choosing which enemies, how many, where they start and when they arrive, for one fight in one space. The same six enemies can be trivial or terrifying depending on the mix.
Enemy types are words and an encounter is a sentence. Good mixes make one enemy's job the answer to your answer for another, so no single habit works for the whole fight.
Not difficulty: Difficulty is how hard it is. Encounter design is what kind of problem it is. Doubling everyone's health makes it harder and no more interesting.
Seen in: Jaime Griesemer called Halo about 30 seconds of fun repeated, and he has said the part often cut off is that those 30 seconds keep changing with new places, weapons and enemies. Doom Eternal's Hugo Martin describes demons as chess pieces.
Say it: "This fight is six rushers and it is boring: players just back up and shoot. Replace two with long-range shooters on the balcony, so that backing up walks them into fire."
https://gamecraft.dqnotes.com/space/encounter-design

### Backtracking
Also: Retreading, Going back the way you came.
Sending the player back through a place they have already been. Done well, the place has changed or they have: new powers open new doors. Done badly, it is the same corridor again with nothing to do.
It gets more use out of everything you built, and returning to a place you once found hard shows you how far you have come. The cost is walking, so it needs either something new on the way or a quick way through.
Not exploring: Exploring takes you somewhere new by your own choice. Backtracking takes you somewhere old, usually because the game requires it.
Seen in: Metroid Prime ends by sending you back across the world to hunt twelve Chozo Artifacts, which many players found the weakest part of a game that otherwise uses returning so well.
Say it: "The way back from the boss is 90 seconds of empty corridor. Add a lift back to the hub, or change the corridor: flood it, and put new enemies in it, so it is not the same walk."
https://gamecraft.dqnotes.com/space/backtracking

### Shortcut
Also: Looping level design, Loop back, One-way door.
A door, lift or ladder that links a far part of the level back to an earlier one, usually opened from the far side. Looping level design builds the whole map around these, so the level folds back on itself.
It turns the place you just fought through into a route you now own, and cuts the walk after every death. Finding one is a reward that costs the designer no new art at all.
Not fast travel: Fast travel skips the world through a menu. A shortcut is a piece of the world, and finding it changes how you picture the map.
Seen in: Dark Souls is famous for it: the lift at the Undead Parish drops you back down to Firelink Shrine, hours of winding route later, and suddenly the map in your head joins up.
Say it: "The run back to the boss takes three minutes. Add a door in the boss antechamber that only opens from that side and leads to the first bonfire."
https://gamecraft.dqnotes.com/space/shortcut

### Hub and spoke
Also: Hub world, Hub level, Overworld hub.
A central place with levels or regions branching off it, so that the player returns to the middle and picks where to go next.
If one part is too hard, the player can do another. It also gives them one familiar home, which is why the shape shows up in theme parks as well as games. The cost is walking back through the centre.
Not open world: A hub links separate places through one centre. An open world is one continuous place you can cross in any direction.
Seen in: Super Mario 64 uses Peach's Castle as a hub, with paintings leading to its courses. If a course is too hard you walk back and jump into another one. Disneyland is the same shape, with its lands around a central plaza.
Say it: "Players quit when the third level stops them. Change the line into a hub with four spokes, any three of which open the final door."
https://gamecraft.dqnotes.com/space/hub-and-spoke

### Density
Also: Negative space, Content density, Breathing room.
How much there is to do in a given stretch of ground, and how much empty ground lies between. The emptiness is negative space, and like the gap between notes in music, it is doing a job.
Too thin, and the player is walking and bored. Too thick, and nothing stands out, because every marker is competing with the next. Quiet stretches are what make the next discovery feel like one.
Not size: Size is how much ground there is. Density is how much happens on it. A small map can be sparse and a huge one can be cramped.
Seen in: Shadow of the Colossus has a vast land with almost nothing in it but its sixteen colossi, and the long empty rides between them are what makes each one feel enormous.
Say it: "The map has an icon every 20 seconds and players are numb to them. Cut them by half and leave a full minute of quiet riding before each major site."
https://gamecraft.dqnotes.com/space/density

### Set piece
Also: Setpiece, Scripted sequence, Spectacle moment.
A big, expensive moment built once for one place, often scripted, that the player will talk about afterwards: the building that falls down while you are in it.
It gives a long game peaks that stay in the memory, and a marketing team something to show. It is costly and works best once, because a second go reveals that it plays out the same way every time.
Not cutscene: In a cutscene you watch. In a set piece you are still in control while the world does something spectacular around you.
Seen in: Uncharted 2 has a chapter set on a moving train, fought along the carriages at speed, with the scenery changing around you the whole way.
Say it: "Save the budget for one set piece: at the end of chapter five the bridge collapses behind the player as they run across it. Nothing else in the chapter needs scripting."
https://gamecraft.dqnotes.com/space/set-piece

### Vista
Also: Reveal, Overlook, Compression and release.
A wide view placed deliberately, usually after a tight space, so the world seems to open up all at once. Tunnels and cave mouths do half of the work, because the eye has just been starved of distance.
It shows the player how big the world is and where they might go next, and it costs no gameplay. Without the tight space before it, the same view is just more scenery.
Not landmark: A landmark is a thing you can see from many places. A vista is one place you can see many things from.
Seen in: Breath of the Wild walks you out of a dark cave and onto a cliff edge, and the camera swings out over all of Hyrule. It is the first thing the game wants you to see, and most of what you see from there is somewhere you can go.
Say it: "Run the path through a narrow canyon for 30 seconds, turn a corner, and open onto the whole valley. Keep the camera steady and let them stop."
https://gamecraft.dqnotes.com/space/vista

### Secrets
Also: Hidden areas, Secret rooms, Easter eggs (when they are a joke).
Things placed off the main route for people who look for them: a wall that is not a wall, a ledge you must think to jump to. Their value is in being found, which needs a hint fair enough to notice.
They reward curiosity, and they make a player trust that looking closely is worthwhile, so they look closely at everything. They also give people something to tell each other about.
Not easter egg: A secret is part of the game, usually with a reward in it. An Easter egg is a joke or a signature, hidden for the fun of it, that gives you nothing to use.
Seen in: Doom counts secrets on the tally screen after every level, so you know you missed one and go back to press on walls. Wolfenstein 3D's push walls got players pressing on walls before that.
Say it: "Put one secret in each level, always behind a wall with a slightly off-colour texture. Show a count at the end of the level, so people start looking for that texture."
https://gamecraft.dqnotes.com/space/secrets

## 5. Systems

Economies, randomness, feedback loops and balance: the machinery beneath the surface.

### Resource
Also: Currency, Economy.
Anything a player has a limited amount of and can spend or use up to get something else: gold, ammunition, health, time on the clock, a turn. If it is scarce and you can choose how to spend it, it is a resource.
A choice only matters if it costs something. Nearly every interesting decision in a game comes down to which resource to spend, and with too much of everything there are no decisions left.
Not stat: A stat describes you and usually stays put, like strength. A resource goes down when you use it. Mana is a resource and intelligence is a stat.
Seen in: Resident Evil gives you so little ammunition that each bullet is a question. StarCraft runs on two resources, minerals and gas, and needing both is what shapes every build.
Say it: "Ammunition is not scarce enough to matter. Cut drops by 40%, and let players trade health items for bullets, so that shooting costs them something."
https://gamecraft.dqnotes.com/systems/resource

### Positive feedback loop
Also: Snowballing, Rich get richer, Reinforcing loop.
A loop where doing well makes you better at doing well: winning gets you the thing that helps you win. Positive here means it amplifies, not that it is good.
It gives a game a shape, because a small lead becomes a decision and the game ends instead of dragging on. Left unchecked, the winner is settled long before the last hour, which everyone else has to sit through anyway.
Not negative feedback loop: A positive loop widens the gap between players. A negative loop closes it. The words say nothing about which is pleasant.
Seen in: Monopoly: owning property pays rent, which buys more property. The game is often decided an hour before it finishes. In League of Legends, a kill gives gold, gold buys items, and items get more kills.
Say it: "Kills give too much gold, so the first fight decides the whole match. Cut the kill bounty by half and pay it to the whole team instead of the killer."
https://gamecraft.dqnotes.com/systems/positive-feedback-loop

### Negative feedback loop
Also: Rubber banding, Catch-up mechanic, Balancing loop.
A loop that pulls the leader back or lifts whoever is behind, so that the contest stays close. Negative here means it dampens, not that it is bad.
It keeps everyone in the game and the ending in doubt. Too strong, and being ahead means nothing, so playing well stops feeling as if it does anything.
Not positive feedback loop: A negative loop shrinks a lead. A positive loop grows it. Most good games use both, with one for each stage of the match.
Seen in: Mario Kart hands the best items to whoever is at the back, and since Mario Kart 64 a spiny blue shell hunts down the player in first place.
Say it: "Races are over by the second lap. Add rubber banding: give karts in last place a 5% speed boost, shrinking to nothing as they reach the middle of the pack."
https://gamecraft.dqnotes.com/systems/negative-feedback-loop

### Emergence
Also: Emergent gameplay, Emergent behaviour.
Behaviour that nobody wrote, arising from a few simple rules interacting. The game's designers can be as surprised by it as anyone.
It gets you far more play than you authored, and players take it personally because they found it themselves. It can't be fully tested and may go somewhere you didn't want, so you steer it by tuning rules, not by writing outcomes.
Not random: Randomness is unpredictable because a die says so. Emergence is unpredictable because the rules are too tangled to foresee, and it does the same thing every time from the same start.
Seen in: Dwarf Fortress players found cats dying in taverns: cats walked through spilled alcohol, licked their paws clean, and got drunk. No one designed it: separate rules for spills, grooming and getting drunk did it together.
Say it: "Stop scripting the fire. Give objects a burn rule and a heat value, and let it spread by itself. I want players to find uses for it that we didn't plan."
https://gamecraft.dqnotes.com/systems/emergence

### Systemic design
Also: Chemistry engine, Multiplicative gameplay, Immersive sim logic.
Building a game from general rules that apply to everything, rather than special cases for each puzzle. If fire burns grass, it burns all grass, whoever lit it, and any solution those rules allow is a real solution.
One new rule multiplies with every rule you already have, so content pays for itself, and players trust the world: if it looks as if it should work, it does. It costs you tight control over what happens.
Not emergence: Systemic design is how you build. Emergence is what you may get from it. You can also build systems that interact and produce nothing surprising.
Seen in: At GDC 2017, the Breath of the Wild team described a chemistry engine that works out what happens between fire, water, ice, wind and materials. Grass burns, the heat lifts your glider, and metal draws lightning, all from the same rules.
Say it: "Don't script the ice puzzle. Add one rule: cold freezes water into a walkable surface. Then it also works on rivers, enemy moats and rain, without any more work from us."
https://gamecraft.dqnotes.com/systems/systemic-design

### Faucets and sinks
Also: Sources and drains, Gold sink, Money sink.
A faucet creates a resource out of nothing, as when monsters drop gold. A sink destroys it for good, as when you pay an NPC for repairs. Money passing between players is neither: it just moves.
Every online game's economy is decided by these two numbers. If more comes in than goes out, money piles up forever and prices follow. Designers have to keep inventing things worth paying for, just to get the money out.
Not trading: Selling an item to another player moves money and creates none. Selling it to a shop creates money, which makes that shop a faucet, and it is easy to forget this.
Seen in: EVE Online publishes a Monthly Economic Report with its sinks and faucets listed, so players can see how much ISK came in from NPC bounties and how much left through taxes and fees.
Say it: "Daily quests add 400 gold a player and nothing removes it. Add a sink: repairs cost 5% of what the player holds, so it settles instead of climbing."
https://gamecraft.dqnotes.com/systems/faucets-and-sinks

### Inflation
Also: Mudflation, Runaway prices, Hyperinflation.
Prices rise because there is more money around, not because anything got better. In a game with no sinks it is almost guaranteed, since players keep producing money and rarely destroy it.
Veterans shrug at prices that lock a new player out for good, because the starting gold that once bought a sword now buys a loaf of bread. Eventually players stop using the official money.
Not power creep: Inflation is too much money chasing the same goods. Power creep is the goods themselves getting stronger. They often arrive together and they have different fixes.
Seen in: In Diablo II gold was so plentiful that players stopped using it, and traded in Stone of Jordan rings instead. Then item duplication flooded the servers with those rings as well, and they lost their value too.
Say it: "Sword prices have tripled in a month. Don't just raise the shop price: add a 5% tax on auction sales that removes the money, and a 2,000-gold mount for the people holding the most."
https://gamecraft.dqnotes.com/systems/inflation

### RNG
Also: Random number generator, PRNG, Luck.
The part of a game that makes up random numbers. It is really a formula that produces a long list that only looks random, starting from one number called a seed, so the same seed always gives the same list.
Randomness makes each attempt different and lets a weaker player win sometimes. A saved seed also lets you replay a run, test a bug, or share a world, and it stops players reloading until they get lucky.
Not procedural generation: RNG only hands out numbers. Procedural generation is what you build from them: maps, loot, names.
Seen in: XCOM: Enemy Unknown saves its random seed, so reloading and taking the same shot gives you the same miss. Enemy Within added a Save Scum option for people who want a fresh roll each time.
Say it: "Use a seeded generator and store the seed in the save, so reloading can't re-roll a shot. Log the seed with every bug report so we can replay what happened."
https://gamecraft.dqnotes.com/systems/rng

### Input and output randomness
Also: Pre-roll and post-roll, Input luck, Output luck.
Input randomness happens before you choose, and you plan around it: the cards you were dealt, a random map. Output randomness happens after you choose and can overturn it: you aim, and then a die says whether you hit.
Input randomness rewards skill, because a good player reads the situation better. Output randomness builds tension and makes losses feel unfair. Moving a die from after the choice to before it changes who wins, with no other change at all.
Not difficulty: Neither kind is harder as such. A game with no output luck can be brutal. The question is whether you can do anything about the luck.
Seen in: Into the Breach shows you what every enemy will do next turn, and your attacks don't miss, so a lost building is usually your own oversight. XCOM rolls after you commit, which is how a 95% shot comes to miss at the worst moment.
Say it: "Players hate missing a 90% shot. Move the luck before the choice: show which enemies are exposed this turn, and make every shot you can take actually hit."
https://gamecraft.dqnotes.com/systems/input-and-output-randomness

### Pseudo-random distribution
Also: PRD, Bad luck protection, Shuffle bag.
Randomness that bends towards its average: each miss raises the chance of the next hit, and each hit resets it. The long-run rate stays the same and the streaks mostly go away. A shuffle bag does a similar job by dealing out a fixed set.
True randomness is streaky, and people read streaks as broken. A 25% crit that fails twenty times looks like a bug, and it will happen to somebody. PRD removes those stories, along with the lucky streaks.
Not true random: With true random each roll ignores the last, so a 25% chance is always 25%. With PRD, a miss makes the next roll more likely. Over thousands of rolls both give about 25%.
Seen in: Dota 2 uses it for many chance-based effects. A 25% chance starts at about 8.5% and adds that much again after every miss, until it fires and starts over.
Say it: "Use a pseudo-random distribution on crits: 25% overall, but start at 8.5% and add 8.5% per miss, so that nobody can ever miss twelve in a row."
https://gamecraft.dqnotes.com/systems/pseudo-random-distribution

### Pity timer
Also: Pity system, Hard pity, Soft pity, Bad luck protection.
A guarantee that after so many failed rolls, the next one pays out. It cuts off the worst luck anyone can have, without changing the rate for everyone else very much.
At a 1% drop rate, some players will go hundreds of rolls with nothing, and they are the ones who quit or write the angry post. A pity timer sets a worst case you can state in public, and it matters more when real money is involved.
Not pseudo-random distribution: PRD nudges the odds up a little after every miss. A pity timer usually leaves the odds alone and then steps in once, at a fixed count. Some games use both.
Seen in: Genshin Impact lists a 0.6% base rate for a 5-star character and guarantees one by the 90th wish. Counting the guarantee, it states an overall rate of 1.6%.
Say it: "Legendary drop stays at 1%, but add a pity timer: guaranteed by roll 150. Show the counter, because being able to see it is half of why it works."
https://gamecraft.dqnotes.com/systems/pity-timer

### Weighted loot table
Also: Loot table, Drop table, Drop rates.
A list of what can drop, each with a weight. Add up the weights, roll a number in that range, and see whose slice it lands in. A weight of 1 in a table totalling 100 is a 1% chance.
It puts an item's rarity in one number a designer can change without touching code. Most of the mistakes come from forgetting that adding a new entry shrinks the share of everything else.
Not rarity: Rarity is what the player is told: a label and a colour. The weight is the real number behind it. The two drift apart if nobody is checking.
Seen in: Minecraft stores its chest and mob drops as loot tables in plain data files: pools of entries, each with a weight, so anyone can read and change them in a data pack.
Say it: "Set the boss loot table to weights 70 common, 25 rare, 4 epic and 1 legendary. Roll it 100,000 times in a test and check the shares before we ship."
https://gamecraft.dqnotes.com/systems/weighted-loot-table

### Opportunity cost
Also: Trade-off, Tempo cost, What you gave up.
What a choice really costs you is the best other thing you could have done with it. The gold you spent on a farm is a soldier you don't have when the raid comes.
It is what separates a real choice from a shopping list. If you can have both, or one is plainly free, there is nothing to think about. Well-designed games make you feel the option you turned down.
Not price: The price is what the tag says. The opportunity cost is what else that price would have bought, and it changes with the moment: gold is worth more one turn before an attack.
Seen in: StarCraft asks it over and over in the first minutes: another worker pays off later, but it is a marine you don't have if the enemy attacks early. A greedy build wins if you survive it.
Say it: "Upgrades are free choices right now, because everyone can afford all of them by turn ten. Make each one cost a full turn of production, so that upgrading means not building."
https://gamecraft.dqnotes.com/systems/opportunity-cost

### Dominant strategy
Also: Optimal strategy, No-brainer, Solved.
A choice that is at least as good as every alternative, whatever the opponent does. Once people find it, the rest of your options might as well not exist.
A game with one is no longer a game of choices. Players will find it however it looks, and they will use it even if it is dull, because they play to win, and they then blame you for the boredom.
Not strong option: A strong option wins often and still has something that beats it. A dominant strategy has nothing that beats it, so no other choice ever makes sense.
Seen in: Online Mario Kart Wii came to be ruled by bikes that drift on the inside of turns, above all Funky Kong on the Flame Runner, because they kept their speed through corners that karts could not.
Say it: "The sword beats both other weapons in every match-up we simulated, so it is a dominant strategy. Cut its damage by 15% so that the spear beats it again."
https://gamecraft.dqnotes.com/systems/dominant-strategy

### Intransitive balance
Also: Rock paper scissors, Counter system, Weapon triangle.
A loop of counters where A beats B, B beats C and C beats A, so there is no best choice, only a best choice against what the other side is doing.
It is the cheapest way to get balance: no number needs to be perfect, because everything already has an answer. It moves the skill to reading your opponent. Too strong and matches are settled at the pick screen.
Not fair: Everything being evenly matched is one kind of balance. Intransitive balance is the opposite: every match-up is lopsided, and the whole set evens out.
Seen in: The Fire Emblem series uses a weapon triangle: swords beat axes, axes beat lances, lances beat swords. Pokémon's types are a much bigger version of the same idea.
Say it: "Let archers beat infantry 2 to 1, infantry beat cavalry 2 to 1, and cavalry beat archers 2 to 1, and scout the enemy before each battle so the choice can be informed."
https://gamecraft.dqnotes.com/systems/intransitive-balance

### Power curve
Also: Linear growth, Exponential growth, Progression curve, Scaling.
How a player's strength rises over time, usually drawn against the same for enemies. Linear growth adds the same amount each step. Exponential growth multiplies by the same amount each step, so it starts slowly and then leaves everything behind.
The gap between the two lines is your difficulty, at every point in the game. Getting the shape wrong means enemies become either trivial or impossible, and no amount of tuning at level 40 can fix a curve that was wrong from level 1.
Not difficulty curve: The power curve measures strength. The difficulty curve is how hard it feels, and that also depends on the player's skill, which grows without appearing on any chart.
Seen in: Cookie Clicker raises each building's price by 15% with every one you buy, so prices double about every five. World of Warcraft had to shrink its numbers in Shadowlands, cutting the level cap from 120 to 60, after years of growth had made them absurd.
Say it: "Damage should grow 8% a level and enemy health 10%, so that fights get a little longer each level, and a player ten levels over can still lose to a group."
https://gamecraft.dqnotes.com/systems/power-curve

### Soft cap and hard cap
Also: Diminishing returns, Stat cap, Level cap.
A hard cap is a limit that can't be passed. A soft cap is a point past which each extra bit does less, so more is still allowed, and hardly worth buying.
Without limits, the best build pours everything into one stat, and at the extreme the stat breaks the game. Caps keep a good number from turning into an infinite one, and they push players to spread their points around.
Not level cap: A level cap limits how far your character can level. A stat cap limits one number. The same words get used for both, so say which you mean.
Seen in: Path of Exile limits elemental resistances to 75% by default, a hard cap, since 100% would mean no damage at all. Elden Ring's Vigor gives far less health per point after 40, and less again after 60, a soft cap.
Say it: "Crit chance needs a soft cap: full value to 40%, then half value to 60%, and a hard stop at 75%, so stacking it never makes other stats pointless."
https://gamecraft.dqnotes.com/systems/soft-and-hard-cap

### Buff and nerf
Also: Balance patch, Rework.
A buff makes something stronger and a nerf makes it weaker. Nerf is named after the foam toys: a nerfed weapon is one that now hits like foam.
Balancing a live game is mostly making these changes in small steps, because small numbers move results a long way. A nerf also feels like having something taken away, so a buff to everything else is often the better-tasting fix.
Not rework: A buff or a nerf changes a number. A rework changes what the thing does, usually because no number could fix it.
Seen in: Ultima Online is where the word reached online games. Its designer Raph Koster has confirmed it, after players complained that a change had left their swords useless, though Merriam-Webster says the origin isn't certain.
Say it: "Nerf the shotgun's range by 10%, and leave its damage alone. Change one number per patch, so we can read what it did from the win rates."
https://gamecraft.dqnotes.com/systems/buff-and-nerf

### The meta
Also: Metagame, Tier list, Flavour of the month.
What most people are playing at the moment, and the choices that makes best in reply. It is the game above the game: you don't only pick what's strong, you pick what beats what everyone else is picking.
It changes with no patch at all, as players copy what wins and then start beating it. A healthy one keeps moving. A stale one means everyone has stopped thinking and started copying a list.
Not dominant strategy: A dominant strategy is best no matter what. The meta only has to be best right now, against what's common, and it can be overturned by something that was bad last week.
Seen in: Hearthstone players follow deck tier lists built from ladder data, such as the Vicious Syndicate reports, and choose their decks by what they expect to meet.
Say it: "Don't patch yet. The top deck has been on top for only a week, and its counter is already rising. Check the shares again in ten days."
https://gamecraft.dqnotes.com/systems/the-meta

### Tech tree
Also: Skill tree, Research tree, Talent tree.
A map of upgrades where each one requires those before it, so that buying one opens the next. You can't afford all of it, so your path through it is your strategy, and it is visible to you and often to your rivals.
It paces new abilities so that nobody gets the strongest thing first, and gives every player a long-term plan. Choices made early shut out others, which is what makes the plan yours.
Not unlock list: A list opens things in a fixed order, and everyone gets them all. A tree branches, and what you take decides what you can't have yet.
Seen in: Sid Meier's Civilization made the idea famous: you can't research Gunpowder without what comes before it, and choosing what to study next is one of the main decisions of a game.
Say it: "The tree is too wide to matter. Cut it to three branches, four nodes deep, with 6 points to spend by the end of the match, so nobody can finish two."
https://gamecraft.dqnotes.com/systems/tech-tree

### Build diversity
Also: Build variety, Viable builds, Option space.
How many different ways of setting up your character or army are actually good enough to use. A game may offer thousands, and only the ones close to the best count.
It is the difference between having choices and appearing to have them. A single overpowered item can quietly wipe out most builds, since everything without it falls below the line, and nobody will mention that in the patch notes.
Not number of items: Adding items adds possible builds. It does not add viable builds, and it can reduce them if a new item is too strong to leave out.
Seen in: Path of Exile's huge passive skill tree is built for this: the same tree has to support very different characters, and its players publish and compare build guides for each new league.
Say it: "Check build diversity after every patch: count builds within 5% of the top DPS. If that drops below ten, something has become mandatory."
https://gamecraft.dqnotes.com/systems/build-diversity

### Min-maxing
Also: Dump stat, Optimising, Theorycrafting.
Making one thing as high as you can by making another as low as you dare. Which build is best follows from the formula, so changing the formula changes how people play.
Players will do this whatever you intended, so what matters is what your formulas reward. If stats add together, all your points go in one basket. If they multiply, a balance between them wins.
Not build diversity: Min-maxing is what a player does to a single build. Build diversity is how many different builds survive once everyone is doing it.
Seen in: Dungeons & Dragons players call the stat they neglect a dump stat, and for a fighter who plans to do no talking, Charisma is the classic one to sacrifice for more Strength.
Say it: "Everyone is dumping defence, because it adds so little. Make damage taken divide by defence rather than subtract it, so that the first points are always worth having."
https://gamecraft.dqnotes.com/systems/min-maxing

### Action economy
Also: Actions per turn, Tempo.
In a turn-based fight, how many things each side gets to do. A party of four with four actions usually beats one monster with four times the health and damage, because that single monster wastes damage and can be stunned out of its whole turn.
It is why boss fights that look fair on paper end in two rounds. It is also why a stun is so strong, and why anything giving an extra action tends to be the best thing in the game.
Not dps: DPS adds up total damage. Action economy is about how that damage is divided into turns, which decides how much of it is wasted or lost to one bad moment.
Seen in: Dungeons & Dragons 5th edition gives powerful solo monsters legendary actions, taken at the end of other creatures' turns, so that one creature isn't simply out-acted by a party.
Say it: "The solo boss dies in two rounds, because it acts once while the party acts four times. Give it three small actions a round, in place of one big one, and keep the total damage the same."
https://gamecraft.dqnotes.com/systems/action-economy

### DPS and time to kill
Also: TTK, Damage per second, Breakpoint.
DPS is average damage over time. Time to kill is how long it really takes to bring something down. They differ because damage comes in lumps: it takes a whole extra shot to deal the last point of damage.
These two numbers set the pace of a shooter or an RPG. Time to kill decides how much a mistake costs and whether you can react. And players will find every breakpoint where a tiny change means one fewer shot.
Not damage per hit: Damage per hit is one number on a weapon. It tells you little alone: a gun that hits for less but fires twice as fast has more DPS. Two guns can have the same DPS and different time to kill.
Seen in: World of Warcraft raiders run damage meters such as Details!, showing DPS for everyone in the group. Call of Duty players compare time to kill for every gun, and it is what they mean when they say a game feels fast.
Say it: "Time to kill is 0.3 s, and nobody has a moment to respond. Raise player health so the best gun takes about 0.6 s, and keep the shots-to-kill at four or five."
https://gamecraft.dqnotes.com/systems/dps-and-time-to-kill

### Perfect imbalance
Also: Rotating meta, Controlled imbalance.
Keeping a competitive game a little off balance on purpose, and changing which way it leans every so often. Players never settle on one solved answer, so there is always something new to work out. Extra Credits popularised the phrase in 2012.
A game in perfect balance can still go stale, once everyone knows the answer. Moving things regularly keeps old strategies from solidifying and gives people a reason to come back, as long as each change is small and nothing is left unplayable.
Not unbalanced: An unbalanced game has one answer and stays that way until someone fixes it. Perfect imbalance leaves many strategies close together, and moves the top spot on purpose.
Seen in: League of Legends ships balance patches about every two weeks, so the strongest champions keep changing and players keep having to solve the game again.
Say it: "Win rates are within 2% of each other, and nobody talks about the game any more. Pick a neglected champion every patch and nudge it up, so that the top three keep changing."
https://gamecraft.dqnotes.com/systems/perfect-imbalance

### Nash equilibrium
Also: Game theory optimal, GTO, Mixed strategy.
A set of strategies where nobody can do better by changing theirs alone. Often it means mixing your choices in exact proportions, so that the other side can't exploit you whatever they do.
It shows you where a game settles once players have worked it out. If the settling point is dull, or one option is never used in it, that option is broken. Being at it means nobody can punish you. It does not mean you are punishing them.
Not dominant strategy: A dominant strategy is best no matter what the other side does. An equilibrium is only stable: each side is doing the best it can, given what the other is doing.
Seen in: In 2015, researchers at the University of Alberta announced that their program Cepheus had essentially solved heads-up limit Texas hold'em, working out a strategy so near to unbeatable that a lifetime of play could not reliably show the difference.
Say it: "Run the match-ups through a solver. If the equilibrium uses the dagger under 2% of the time, the dagger is too weak. I don't care that some players enjoy it."
https://gamecraft.dqnotes.com/systems/nash-equilibrium

### Prisoner's dilemma
Also: Iterated prisoner's dilemma, Tit for tat, Cooperate or defect.
Two players each pick cooperate or betray. Betraying is always better for you whatever the other does, yet both end up worse off than if both had cooperated. Play it many times and things change, because betrayal can then be punished.
It explains why players grief, hoard and betray when the rules reward it, however nice they are. If you want co-operation in a game, the rules have to make the relationship last: reputation, repeated meetings, a future.
Not zero-sum game: In a zero-sum game, my gain is your loss, and co-operation is meaningless. In a prisoner's dilemma both can win or both can lose, which is why betrayal is such a temptation.
Seen in: The British game show Golden Balls ended with the actual dilemma, Split or Steal. In the 1980 tournaments run by Robert Axelrod, the winner for repeated play was Anatol Rapoport's Tit for Tat: start nice, then copy whatever the other did last.
Say it: "Players keep killing each other at the shared stash. Make it repeat: let squads stay together for several raids and show past betrayals, so that stealing has a price next time."
https://gamecraft.dqnotes.com/systems/prisoners-dilemma

## 6. The player's mind

Motivation, reward, memory and bias: what is going on in the head of the person holding the controller, and how sure anyone really is.

### Intrinsic motivation
Also: Playing for its own sake, Internal motivation.
Doing something because the doing is the reward. Nobody plays tag for the points.
It is what is left when the rewards run out. A game with none of it loses players the day the unlocks stop, however generous they were.
Not extrinsic motivation: Intrinsic is wanting to do the thing. Extrinsic is wanting what you get for doing it. Most play mixes the two, and the same activity can move from one to the other.
Seen in: People spent hours in Minecraft's creative mode, which has no goals, no score and nothing to unlock.
Say it: "Before we add any rewards, test the core with the score hidden. If nobody wants a second go at the jump with no points, a points system will only hide that."
https://gamecraft.dqnotes.com/mind/intrinsic-motivation

### Extrinsic motivation
Also: External rewards, Carrot.
Doing something for what it gets you: points, loot, money, a badge, or avoiding a penalty.
It gets people started on things they would not try otherwise, and that has real value. It also stops working when the reward does, so it can only carry a game for as long as you can keep paying.
Not intrinsic motivation: Extrinsic rewards come from outside the activity. They can sit on top of intrinsic enjoyment without harming it, particularly when they tell you that you did well.
Seen in: Xbox achievements, added with the Xbox 360 in 2005, gave people a score across every game for doing things they might never have tried, such as finishing on the hardest difficulty.
Say it: "Use the daily quest to introduce the new crafting system, then drop it after a week. If people stop crafting when the reward stops, the reward was the only reason they did it."
https://gamecraft.dqnotes.com/mind/extrinsic-motivation

### Self-determination theory
Also: SDT, Autonomy, competence, relatedness.
Edward Deci and Richard Ryan's theory that people thrive when three needs are met: autonomy (it was my choice), competence (I am getting good at this) and relatedness (other people matter here).
It gives you three concrete questions to ask of a dull section, and it has far more research behind it than most game-design frameworks. In 2006 Ryan, Rigby and Przybylski surveyed players and found that felt autonomy and competence predicted enjoyment and wanting to keep playing. That is correlation from surveys, so treat it as a guide.
Seen in: Breath of the Wild is strong on two of the three: you can go anywhere from the start (autonomy), and its physics rewards your own ideas (competence). It is a single-player game and does little for the third, which shows you need not hit all of them.
Say it: "This tutorial fails on autonomy: it is a corridor with one solution. Let players choose between two routes, and show their best time at the end so they can see themselves improve."
https://gamecraft.dqnotes.com/mind/self-determination-theory

### Overjustification effect
Also: Undermining effect, Crowding out.
Rewarding people for something they already enjoy can make them do less of it once the reward stops. They come to see the reward as the reason they were doing it.
It warns you that bolting rewards onto something already fun can cost you. Its scope is argued over. Lepper, Greene and Nisbett showed it in 1973 in preschoolers expecting an award for drawing. Deci, Koestner and Ryan's 1999 review of 128 experiments found tangible, expected rewards do undermine interest, while Cameron and Pierce argued the harm is limited and avoidable. Surprise rewards and praise did not show the same harm in those reviews.
Not hedonic adaptation: Adaptation means a reward stops feeling good because you are used to it. Overjustification means the reward changes why you think you are doing the activity.
Say it: "Do not pay out coins for finishing the story chapters. People are already reading them because they want to know what happens. Give a surprise gift at the end instead of a promised reward up front."
https://gamecraft.dqnotes.com/mind/overjustification

### Flow
Also: In the zone, Flow channel, Flow state.
The state of being fully absorbed in a task that stretches your ability without overwhelming it, so that time seems to pass differently. Mihaly Csikszentmihalyi described it from interviews and experience sampling, from 1975 onward.
It explains two different ways to lose a player: too hard, and they leave angry; too easy, and they leave bored. Skill grows, so difficulty has to grow too. The channel is a model and its edges have never been measured for any given game. Reviews find that balance between challenge and skill relates to flow only moderately.
Not fun: Flow is one kind of engagement, and some people enjoy games without ever being in it. Calm games and social games can be a joy with hardly any challenge at all.
Seen in: Tetris gets faster as you clear lines, which keeps the challenge rising with your skill. Jenova Chen built his 2006 thesis game flOw around the idea, letting players choose their own difficulty as they go.
Say it: "Players are bored by world two and overwhelmed by world four. Smooth the ramp: add one new enemy type per level, and raise speed only after the player clears a level without being hit."
https://gamecraft.dqnotes.com/mind/flow

### Fiero
Also: Triumph, Hard fun.
The burst of triumph after beating something hard, often with arms in the air. Nicole Lazzaro used the Italian word for pride in her 2004 "Four Keys to Fun" research, where it is the reward of what she called hard fun.
You cannot hand it out. It needs a real chance of failure first, so if you remove all the frustration you remove this too. Lazzaro's framework came from observing players and is widely used, but it is a descriptive model, not an experimental finding.
Not power fantasy: Fiero is earned against resistance and needs you to have struggled. A power fantasy gives you the strength from the start and lets you enjoy it.
Seen in: Dark Souls is built around it. The words VICTORY ACHIEVED barely matter on their own. They matter because you have been dying to that boss for an hour.
Say it: "The boss goes down too easily for the music and banner we gave it. Make the player lose to it at least once on average, then the same fanfare will actually land."
https://gamecraft.dqnotes.com/mind/fiero

### Power fantasy
Also: Empowerment, Horde mode feeling.
Designing the game so that the player gets to feel overwhelmingly strong. The enemies exist to be swept aside.
It is a legitimate pleasure and a tired player may want nothing else. Done without any contrast it goes flat, which is why many games make you weak first, so you have something to compare with.
Not fiero: A power fantasy gives you the feeling for free. Fiero makes you earn it. Plenty of games move from the second to the first as you level up.
Seen in: Vampire Survivors starts you weak and ends with the screen full of your attacks and enemies evaporating in crowds. The Dynasty Warriors series has built itself on one fighter against hundreds.
Say it: "For the last hour, stop scaling enemies. Let the player one-shot the things that killed them in hour one, so they can see how far they have come."
https://gamecraft.dqnotes.com/mind/power-fantasy

### Bartle's player types
Also: Achievers, explorers, socialisers, killers, Bartle taxonomy.
Richard Bartle's 1996 paper sorted players of MUDs, which are text-based online worlds, into four kinds: achievers, explorers, socialisers and killers.
It gives a team useful words for "not everyone wants what you want". Use it for that only. It came from long discussions among the administrators of one kind of game, and it was not built from data. Nick Yee's surveys later found the parts of each type do not reliably go together, and the online Bartle test forces people to pick. Bartle himself never meant it to cover all games.
Not audience segment: A segment comes from data about your own players. Bartle's types are a thirty-year-old theory about MUD players, so do not plan a marketing budget around them.
Seen in: Bartle's own case was MUD1, the text world Roy Trubshaw began in 1978 and Bartle went on to finish, which is a long way from the games the types are now applied to.
Say it: "Everything we have planned rewards completion. Add one thing for people who like poking at how the world works: a hidden area with no marker and no reward. Do not call them explorers on the slide as if that were science."
https://gamecraft.dqnotes.com/mind/bartle-types

### Operant conditioning
Also: Reinforcement, Reward learning, Shaping.
Behaviour followed by a reward becomes more likely, and behaviour followed by nothing or by punishment becomes less likely. B. F. Skinner studied it from the 1930s onward.
It is one of the best-established findings in psychology, and every tutorial uses it. It also explains how players learn the wrong lesson: whatever they happened to be doing as the reward landed gets the credit.
Not classical conditioning: Classical conditioning pairs two things that happen to you, such as a bell and food. Operant conditioning pairs something you do with what follows it.
Seen in: Super Mario Bros. teaches that coins are good and that jumping at blocks can pay, all within its first minute and with no text, by rewarding you for trying.
Say it: "Play the reward sound on the same frame as the hit, not when the score updates at the end of the wave. Otherwise players credit the wrong action."
https://gamecraft.dqnotes.com/mind/operant-conditioning

### Skinner box
Also: Operant chamber, Reinforcement schedule.
A sealed cage with a lever and a food dispenser that Skinner used to study how the timing of rewards changes behaviour. He preferred the name "operant chamber". In games, "Skinner box" is an insult for a game that is mostly a schedule.
The chamber's lesson is the schedule: when rewards come shapes behaviour as much as how big they are. Ferster and Skinner recorded these patterns in 1957, mainly with pigeons, so apply them to people with care. John Hopson brought the ideas to game makers in a 2001 article.
Not compulsion loop: A Skinner box is about the schedule of rewards. A compulsion loop is about the cycle of activities around them. Calling a game a Skinner box is a judgment; calling something a loop need not be.
Seen in: Cow Clicker, which Ian Bogost made in 2010, was a joke at this design's expense: you click a cow, then you wait six hours to click it again. People played it in earnest anyway.
Say it: "Pull the daily reward chest out of the schedule. Right now people log in, press a button and leave. Tie the reward to a match played, so it rewards doing something."
https://gamecraft.dqnotes.com/mind/skinner-box

### Variable reward
Also: Variable ratio schedule, Random reward, Intermittent reinforcement.
Rewards that come after an unpredictable number of actions. It is the same pay on average as a fixed schedule, but you can never tell when the next one lands.
In Ferster and Skinner's work this schedule produced fast, steady responding that kept going long after rewards stopped, because a long dry run looks like bad luck. It is a real source of excitement in loot, and it is also what makes gambling hard to leave, so decide which you are building.
Not random reward with a floor: A pure variable schedule has no limit on how long you can go dry. A pity timer caps it, which keeps the surprise while removing the worst runs.
Seen in: Diablo drops are the classic game example: the next monster could be the one with the good axe. It costs you nothing but time there, whereas a paid slot charges for every pull.
Say it: "Use a variable drop for bonus loot, but keep the main progress bar fixed, so nobody ever feels stuck. Never sell a chance at the reward for real money."
https://gamecraft.dqnotes.com/mind/variable-reward

### Compulsion loop
Also: Core reward loop, Engagement loop.
A repeating cycle in which an action earns a reward, and the reward makes the next action possible or more appealing: fight, loot, upgrade, fight something harder.
It is how many games stay interesting between big moments, and nearly every action RPG has one. The word is industry slang and the psychology behind "compulsion" is loose. Use it to ask a real question: when does this loop let the player stop?
Not core loop: A core loop is what you repeat in a game, and it can be purely enjoyable. People say compulsion loop when the pull is about the reward more than the activity.
Seen in: Destiny is a well-known example, with missions dropping gear that makes you able to take on harder missions that drop better gear.
Say it: "Cap the loop with a natural end: after three missions, the camp scene plays and saving is suggested. People should be able to finish a session feeling done."
https://gamecraft.dqnotes.com/mind/compulsion-loop

### Near-miss effect
Also: Near miss, So close.
A loss that looks almost like a win, such as two jackpot symbols and a third one row off. The claim is that this feels closer to winning than other losses and makes people play on.
The evidence is split, so be careful. Reid wrote about it in 1986, Kassinove and Schare reported that near misses raised persistence in 2001, and Clark and colleagues in 2009 found near misses recruited reward-related brain areas. But Pisklak, Yong and Spetch in 2019 reviewed half a century of this work and found no conclusive, replicable evidence of more persistence. The feeling is real; its effect on behaviour is unsettled.
Not close game: Losing a race by a metre tells you something about your skill. A slot machine's near miss tells you nothing, since the reels were never close to anything.
Seen in: When you run out of moves in Candy Crush Saga with a level nearly done, it offers to sell you more moves. That is a different thing from a slot machine, since you really did nearly do it, but the sales pitch arrives at the same moment.
Say it: "When a player loses a skill-based match, show the true margin. Do not ever fake a close result, and do not put a purchase on the losing screen."
https://gamecraft.dqnotes.com/mind/near-miss-effect

### Hedonic adaptation
Also: Hedonic treadmill, Reward fatigue, Getting used to it.
Good things stop feeling as good as you get used to them. A reward is felt against what you have come to expect, not on its own.
It is why a game's numbers have to keep climbing, and why the loot that thrilled you in hour one is vendor trash by hour twenty. Brickman and Campbell proposed the treadmill in 1971. Later long-term studies, such as work by Lucas and by Diener, show that adaptation is often incomplete and differs from person to person.
Not overjustification effect: Adaptation is about how strong a reward feels over time. Overjustification is about what happens to your interest in the activity.
Seen in: Diablo games escalate damage numbers from tens into the billions across their lifespans, largely to keep the same feeling alive.
Say it: "Do not raise the reward each time. Change its kind: after five gold drops, the next prize should be a new ability. Variety wears out more slowly than size."
https://gamecraft.dqnotes.com/mind/hedonic-adaptation

### Endowed progress effect
Also: Head start effect, Artificial advancement.
People work harder towards a goal, and finish it more often, if they are given a head start, even when the amount left to do is the same. Joseph Nunes and Xavier Drèze named it in 2006.
It costs nothing and it works on anything with a bar. In their car wash study, 34% of ten-stamp cards with two free stamps were redeemed against 19% of plain eight-stamp cards, over nine months, from 300 cards handed out. Both needed eight paid washes. The result is cited a lot and a reason for the head start seems to help. The headline figure is a single field study, so measure it in your own game.
Not goal gradient: Endowed progress is about where you start. Goal gradient is about speeding up as you near the end. A head start works partly by putting you further along that gradient.
Seen in: Profile-completion bars on sites such as LinkedIn start partly full, counting things you have already done as progress.
Say it: "Start the first quest at 2 of 10 with a line saying "You already found two in the tutorial". It should be true, and the player still has eight left to find."
https://gamecraft.dqnotes.com/mind/endowed-progress

### Goal gradient
Also: Goal-gradient hypothesis, Speeding up near the end.
Effort rises as you get closer to a goal. Clark Hull proposed it in 1932 and tested it with rats in a runway, which ran faster the nearer they were to food.
It tells you that the middle of a long bar is where people drop off, so break big goals into smaller ones. Ran Kivetz, Oleg Urminsky and Yuhuang Zheng revived it in 2006: café customers shortened the time between coffees by about 20% between the first stamp and the last, and customers on a 12-stamp card with 2 free finished ten purchases in 12.7 days against 15.6.
Not endowed progress effect: A goal gradient is the speeding up on the way. The endowed progress effect is the head start at the beginning.
Seen in: World of Warcraft's experience bar is split into 20 visible bubbles, so there is always a small goal nearby, and people play on to reach the next one.
Say it: "The 100-level pass loses people around level 40. Split it into ten chapters of ten levels with a reward at the end of each, so nobody is ever more than ten levels from a finish."
https://gamecraft.dqnotes.com/mind/goal-gradient

### Zeigarnik effect
Also: Unfinished business, Ovsiankina effect (the resuming half).
The claim, from Bluma Zeigarnik's experiments published in 1927, that people remember unfinished tasks better than finished ones. Her colleague Maria Ovsiankina reported the related urge to go back and finish.
The popular version overclaims. A 2025 meta-analysis by Ghibellini and Meier found no general memory advantage for unfinished tasks, but it did find a general tendency to resume them. So do not say players "remember" an open quest better. Say that an interrupted task tends to pull you back, which is the part that seems to hold.
Not cliffhanger: A cliffhanger makes you want to know what happens next. The urge to resume is about a task you started and did not finish, even a dull one.
Seen in: Civilization's "one more turn" works this way: at the end of any turn something is nearly finished, a building or a research, so there is no clean moment to leave.
Say it: "When the player quits, save mid-quest and show that quest first when they come back. Do not nag with a notification: being asked is not the same as the pull of an unfinished thing."
https://gamecraft.dqnotes.com/mind/zeigarnik-effect

### Completionism
Also: Completionist, 100%, Collect-a-thon, Coupon collector.
The urge to finish everything a game offers: every item on the checklist, every marker on the map, the number to 100%.
A visible checklist turns optional content into unfinished business, and some people love that. Others feel it as a chore they cannot stop doing, so keep it opt-in. If items are random, the last few cost far more than the first, which is where collection turns into a money trap.
Not mastery: A completionist wants nothing left unticked. A master wants to do one thing perfectly. Someone can have every achievement and still be bad at the game.
Seen in: Hollow Knight lets you push completion to 112%, which is a joke and a dare. Xbox achievements gave completion a public score starting in 2005.
Say it: "Show the collection percentage only after the credits. During the story it pushes people off the path. And never give random drops to a complete-the-set reward: use a shop for the last few."
https://gamecraft.dqnotes.com/mind/completionism

### Curiosity gap
Also: Information gap, Mystery box.
The itch of knowing that there is something you do not know. George Loewenstein described curiosity in 1994 as a gap between what you know and what you want to know.
It only works if the player already knows a bit: with nothing to go on there is no question, and with everything there is no reason. It is cheap to build and it does not wear out as rewards do. The trap is a promise you cannot keep, since players remember a mystery that led nowhere.
Not clickbait: Both open a gap. A curiosity gap pays it off with something worth knowing. Clickbait opens it just to collect the click.
Seen in: Outer Wilds runs almost entirely on this. Nothing you gain makes your ship stronger, only what you know changes, and each answer raises a new question.
Say it: "Put a locked door with a light under it in the first room, in the player's line of sight. Do not mark it on the map. Let them find the key an hour later."
https://gamecraft.dqnotes.com/mind/curiosity-gap

### Loss aversion
Also: Loss framing, Gain framing, Framing effect, Prospect theory.
Losing something hurts more than gaining the same thing feels good. Daniel Kahneman and Amos Tversky built this into prospect theory in 1979, and in 1981 showed that wording alone can flip a choice.
It explains why players hate penalties that are mathematically the same as bonuses. Tversky and Kahneman estimated in 1992 that losses weigh roughly twice as much as gains. Gal and Rucker argued in 2018 that this is overgeneralised and depends on context. Yet a 2020 replication across 19 countries by Ruggeri and colleagues supported prospect theory's main predictions, so it stands as strong but not universal.
Not risk aversion: Risk aversion is preferring a sure thing. Loss aversion is about gains and losses feeling different, and it can make people take big risks to avoid a loss.
Seen in: World of Warcraft's rest system is the industry's favourite story here. Developers have described how in beta a tired character earned 50% and a rested one 100%, which players hated. It was relabelled as 100% and a 200% bonus with the same result, and players liked it. It is told as an anecdote, and the exact numbers vary between accounts.
Say it: "Do not say "lose 20% XP if you leave the match". Say "finish the match for a 25% bonus". It's the same maths and nobody will write an angry post about it."
https://gamecraft.dqnotes.com/mind/loss-aversion

### Sunk cost fallacy
Also: Sunk cost, Already paid, In too deep.
Carrying on with something because of what you have already put in, even though that is gone whatever you do next. Only what comes from here on should count.
Arkes and Blumer showed it in 1985 with theatre season tickets: people who paid full price went to more shows than those given a discount. Many people have played a game they stopped enjoying because of the hours already logged, and games can lean on this, which is why it is worth knowing.
Not loss aversion: Loss aversion is about feeling a future loss too strongly. Sunk cost is about counting a past cost that cannot be recovered.
Seen in: Genshin Impact's pity counter carries over between banners of the same type. A player 70 pulls in has a reason to stay that a new player does not, and part of it is real because the odds really will improve, which makes it a hard case to reason through.
Say it: "Show players what they would still need to do, not what they have put in. "12 hours left" helps someone decide. "You have invested 200 hours" is only there to stop them."
https://gamecraft.dqnotes.com/mind/sunk-cost

### Endowment effect
Also: Ownership bias, Mine is better.
People value a thing more once they own it. Kahneman, Knetsch and Thaler's 1990 study gave students mugs, and owners asked about twice what others would pay for the same mug.
It tells you that a player will care more about a thing they feel is theirs. It also has limits: Plott and Zeiler showed in 2005 that with careful procedures the gap can shrink or disappear, so how general it is, and why it happens, stays debated.
Not sunk cost fallacy: Sunk cost is about what you have paid. The endowment effect can happen with something you were given a minute ago for free.
Seen in: XCOM lets you name your soldiers and change how they look. Losing Sergeant Dave, whom you named after your brother, hurts far more than losing a unit.
Say it: "Let players name their first ship and paint it before the first mission, with no stats changing. Then when it gets damaged, they will care about repairing it."
https://gamecraft.dqnotes.com/mind/endowment-effect

### Gambler's fallacy
Also: Due for a win, Law of averages.
Believing that a run of one outcome makes the other more likely next, when each roll is independent. After five misses, a true 50% shot is still 50%.
Your players will believe this however good your random numbers are. Tversky and Kahneman described the root in 1971 as a belief that small samples should look like the whole. That is why fair randomness feels broken, and why many games quietly stop being fair.
Not pity timer: The fallacy is wrongly believing a game has memory. A pity timer gives it one, so believing a win is due is then correct.
Seen in: XCOM's players routinely complain about missing several 80% shots in a row, even though with enough shots that is bound to happen to someone.
Say it: "Players are reporting the drops as bugged. Keep the 10% average, but add a counter that raises the odds after each miss and guarantees a drop by 25 tries."
https://gamecraft.dqnotes.com/mind/gamblers-fallacy

### Fudged odds
Also: True hit, Hidden assist, Lying about probability.
Players read probability badly, so games often show one number and use another. A shown 75% may really be 88%.
Sid Meier told GDC in 2010 about testing combat for Civilization Revolution. Players were delighted to win at long odds, accepted a loss at 2-to-1, and got angry at a loss at 20-to-10, which is the same ratio. His conclusion was that a purely mathematical system would not do, and player psychology has to count. Fudging is a kindness in a game with nothing at stake. It stops being one when players pay on the basis of the number.
Not rubber banding: Rubber banding changes the game based on how you are doing. True hit is a fixed lie about the numbers, and it applies whoever you are.
Seen in: Most Fire Emblem games from The Binding Blade on average two random rolls for hits, so a displayed 75 really hits about 88% of the time. Some later ones changed the method: Fates uses a hybrid, Engage a modified single roll.
Say it: "Average two rolls for all hit chances shown to the player, so 90% rarely misses and 10% rarely lands. Never apply this to anything bought with real money: paid odds must be exactly what we state."
https://gamecraft.dqnotes.com/mind/fudged-odds

### Peak-end rule
Also: Duration neglect, End on a high.
People judge a past experience largely by its most intense moment and how it ended, and they pay surprisingly little attention to how long it lasted. Kahneman and colleagues showed it with cold water in 1993, and with colonoscopy patients in 1996.
A player reviewing your game is reviewing the memory of it. It is a strong rule of thumb, not a law: a 2022 meta-analysis by Alaybek and colleagues found a large peak-end effect and almost no effect of duration, but also found that a plain average of the whole experience predicted memory about as well.
Not pacing: Pacing shapes how the experience feels while it is happening. The peak-end rule is about what is left when it is over.
Seen in: Peggle plays Ode to Joy in slow motion as your last peg goes. After a so-so level, that ending is what you remember.
Say it: "The session currently ends on a defeat screen and a loading bar. Move the XP tally and level-up after the match and add the best moment replay, so the last thing they see is good."
https://gamecraft.dqnotes.com/mind/peak-end-rule

### Attribution of failure
Also: Attribution theory, Fair death, That was my fault.
What people decide caused their failure: themselves or the game, something they can change or something they cannot. Bernard Weiner built this into a theory of motivation in the 1980s.
Players who think "I jumped too late" try again. Players who think "that was cheap" quit, and the same death can land either way. Weiner's framework comes from classroom research, so applying it to games is reasoning by analogy, though designers have found it useful for a long time.
Not difficulty: Difficulty is how often you fail. Attribution is who you blame. A very hard game can feel fair, and an easy one can feel like it cheated.
Seen in: Super Meat Boy kills you constantly, restarts you at once, and at the end replays every one of your failed attempts at the same time, so you can see that the mistakes were yours and that you got better.
Say it: "When an unseen enemy kills the player, they blame the game. Add a half-second warning sound, and show the killer on the death screen with what hit them and for how much."
https://gamecraft.dqnotes.com/mind/attribution-of-failure

### Cognitive load
Also: Mental load, Working memory limit, Too much at once.
How much a person has to hold in mind at once. John Sweller developed cognitive load theory in 1988, from research on how people learn.
Attention is limited, and the limit is smaller than designers think. Cowan argued in 2001 that people hold about four chunks at a time. Everything on your screen and in your tutorial uses some of that. Load itself is hard to measure, and researchers argue about how to do it, but the limit is easy to feel.
Not difficulty: Difficulty is how hard the challenge is. Load is how much you must keep track of to engage with it. A chess puzzle is hard with little to track; a cluttered menu is easy but heavy.
Seen in: Portal teaches one new idea per room, each tested alone before it is combined with others. By the end you are managing several at once without ever having been overloaded.
Say it: "The tutorial asks them to learn movement, crafting and the map in one screen. Split it: walking only in room one, crafting only after the first enemy, map when they first get lost."
https://gamecraft.dqnotes.com/mind/cognitive-load

### Mental model
Also: What the player thinks is going on, Player's theory of the game.
The picture a player builds of how the game works. It is only ever a guess, and it steers everything they try next.
Players act on their model, not your code. If they believe that hiding in grass makes them invisible, your stealth system is being judged against a rule you never wrote. Don Norman's The Design of Everyday Things is the usual reference: the design has to communicate the model, because you won't be there to explain.
Not tutorial: A tutorial is one way of teaching the model. The model is what the player actually ended up believing, which you only discover by watching them play.
Seen in: In Super Mario Bros. the first Goomba walks straight at you, so you find out what enemies do, and the first mushroom slides off a pipe towards you, so you find out mushrooms help.
Say it: "Testers think the shield blocks everything, but it only blocks the front. Show a visible arc around the player when blocking and a spark on any blocked hit, so the model comes from the screen."
https://gamecraft.dqnotes.com/mind/mental-model

### FOMO
Also: Fear of missing out, Limited-time pressure, Daily login pressure.
The worry that you are missing something other people are getting. Przybylski and colleagues gave it a formal scale in 2013. In games it is deliberately created by things that will not come back.
It brings people back, and it can also make the game feel like a job. The scale measures a general anxiety about social life, not a response to game design, so treat claims about what game FOMO does to players with care. What you can say for certain is what the design asks of them: look at the calendar below.
Not live event: A one-off in-game concert is a shared moment that you can miss with no lasting penalty. FOMO design makes the missing cost you something permanent, like an exclusive reward.
Seen in: Destiny 2's seasonal content used to be removed when seasons ended, and many players wrote about the pressure to finish it in time. Fortnite's battle pass rewards run on the same kind of deadline.
Say it: "Do not make the limited skin vanish forever. Put it in the shop a year later for everyone. And let streaks survive one missed day in seven, so holidays do not break them."
https://gamecraft.dqnotes.com/mind/fomo

### Dark pattern
Also: Deceptive design, Manipulative design, Roach motel, Confirmshaming.
A design that gets people to do something they would not choose if they saw clearly. Harry Brignull coined the term in 2010. Zagal, Björk and Lewis applied it to games in 2013.
It makes this quarter's numbers go up by spending your players' trust, and regulators now act on it. The FTC's 2022 case against Epic over Fortnite's purchase design ended in an order, finalised in March 2023, to pay $245 million in refunds. The test: if you would be embarrassed to explain how the button works, it's a dark pattern.
Not persuasive design: Persuasive design makes the thing you already want easier. A dark pattern depends on you not noticing, or on making your own choice hard to act on.
Seen in: The FTC said Fortnite's button layout led players to buy things by accident with a single press, and that accounts were locked when people disputed charges.
Say it: "Cancel must take the same number of taps as subscribing. Same size button, same contrast, no guilt copy. Any screen in the purchase flow that we would not show a journalist comes out."
https://gamecraft.dqnotes.com/mind/dark-pattern

## 7. Story

How games tell stories that films and books cannot: agency, structure and the things only play can say.

### Agency
Also: Player agency, Meaningful choice.
The player's power to change what happens in a way that matters to them. It is measured by what is different afterwards, not by how many buttons were offered.
Agency is the one thing a game has that a film does not. Take it away at the moment the player cares most and they stop feeling like the cause of anything.
Not freedom: Freedom is how much you can do. Agency is how much of it changes anything. A huge open world where nothing reacts has plenty of the first and little of the second.
Seen in: Undertale lets you spare every monster you meet. The ending, the way characters treat you and even the shape of some boss fights follow from what you did.
Say it: "Give the player real agency over the fire: which houses burn depends on where they send the water, and the closing scene names each house they saved."
https://gamecraft.dqnotes.com/story/agency

### Ludonarrative dissonance
Also: Story and gameplay clash.
What the rules make you do contradicts what the story says is going on. Clint Hocking named it in 2007 in a blog post about BioShock.
Players believe what they do over what they are told. If the script calls your hero gentle and the game pays per kill, the script loses, and every later cutscene sounds slightly false.
Not plot hole: A plot hole is the story contradicting itself. Dissonance is the rules contradicting the story, even when the story is airtight.
Seen in: Hocking's point was that BioShock's rules reward looking after yourself while its story asks you to help Atlas with no choice in the matter. Harvesting a Little Sister pays 160 ADAM against 80 for rescuing her.
Say it: "The hero is written as someone who hates killing, but kills are our only source of XP. That is ludonarrative dissonance. Pay 10 XP for sparing and 2 for killing, or rewrite the hero."
https://gamecraft.dqnotes.com/story/ludonarrative-dissonance

### Ludonarrative harmony
Also: Ludonarrative resonance.
The rules and the story say the same thing, so playing the game is itself the telling of the story.
When the rules carry the theme, the writing has less to do. Players feel the point in their hands and nobody has to explain it.
Not tutorial in the fiction: Dressing a button prompt up as a character speaking is only a coat of paint. Harmony means the rule itself means what the story means.
Seen in: In Hades, Zagreus keeps trying to escape the underworld, and dying sends him back through the blood pool to the House of Hades, where characters talk about how the last attempt went. Failure is the plot.
Say it: "Make death part of the fiction: the player wakes at the inn, the innkeeper mentions how they died, and each death moves one story beat forward instead of reloading."
https://gamecraft.dqnotes.com/story/ludonarrative-harmony

### Emergent narrative
Also: Embedded narrative (its opposite), Player stories, Systemic storytelling.
Embedded narrative is written in advance and waits to be found, as a cutscene or an audio log. Emergent narrative comes out of the rules running, so players end up telling stories nobody wrote. Henry Jenkins set out both in a 2004 essay.
Embedded story gives you control and costs words. Emergent story costs systems instead, and the player owns it, but you cannot promise it will land or be any good.
Not branching narrative: A branch is still written by someone, in advance. An emergent story was never written at all: only the rules that produced it were.
Seen in: Middle-earth: Shadow of Mordor's Nemesis system lets an orc who kills you get promoted, remember the fight and carry the scars when you meet again. BioShock's audio diaries are the embedded kind.
Say it: "I want emergent stories from the village: give each person a grudge list and a hunger value, let theft raise the grudge, and print what happens as a plain log. Don't script any of it."
https://gamecraft.dqnotes.com/story/emergent-narrative

### Show, don't tell
Also: Show don't tell.
Let the audience see a thing happen instead of being informed of it. Games have a stronger form: let the player do it.
A caption asks to be believed. A town that clears the street when you walk in does not need to ask.
Not environmental storytelling: Show, don't tell is the general rule, for any scene in any medium. Environmental storytelling is one way of following it, using the place.
Seen in: Inside tells its whole story without one word of dialogue or text. You work out what is being done to those people from what you see them made to do.
Say it: "Delete the line saying the guards fear her. When she walks into the barracks, have them stop talking, stand up and leave the room."
https://gamecraft.dqnotes.com/story/show-dont-tell

### Environmental storytelling
Also: Staged spaces, What happened here.
Arranging a place so that the player works out what happened there. Don Carson brought the idea to games in a 2000 article, borrowing it from theme park design.
Players trust what they deduce more than what they are told, and the player keeps moving while it happens. A few moved props can cost less than one line of voice acting.
Not set dressing: Set dressing makes a place look nice. Environmental storytelling makes it say something: take one prop away and a fact about the past goes with it.
Seen in: Portal's test chambers look spotless, until a loose panel shows the scrawled den of someone who got out, with "the cake is a lie" on the wall. Nobody says a word, and you stop trusting the voice guiding you.
Say it: "Stage this cabin so players can tell someone left in a hurry: two bowls on the table, one chair jammed under the door handle, boots still by the bed. No notes, no voice-over."
https://gamecraft.dqnotes.com/story/environmental-storytelling

### Branching narrative
Also: Branching story, Choose your own adventure, Time cave.
The story splits at each choice into separately written paths. Pure branching never rejoins, so every decision really does send you somewhere else.
Nothing else makes a choice feel so large, or makes so much of your writing invisible. Most of what you write, most players never read.
Not foldback: A branch stays split. A foldback rejoins. Real games mix them, because pure branching gets too expensive within a few choices.
Seen in: The Witcher 2 splits on one decision: side with Roche or with Iorveth, and the entire second act plays out somewhere else, with different people and quests. You have to replay the game to see the other half.
Say it: "Use real branching only for the Act 1 choice. It leads to two separate second acts, about 40 scenes each, and both rejoin before the finale."
https://gamecraft.dqnotes.com/story/branching-narrative

### Branching explosion
Also: Combinatorial explosion, Geometric explosion.
Each extra layer of choices multiplies the writing. Two options per scene, ten choices deep, is 2,047 scenes, and one playthrough reads eleven of them.
This is the sum that decides most story structures. It explains why games fold their branches back together and why a promised "every choice matters" so rarely survives production.
Not content bloat: Bloat means there is too much for one player to get through. The explosion means there is too much to write, however little of it each player sees.
Seen in: Disco Elysium's script runs to over a million words, and that took holding the branches together with checks and flags. One playthrough still reads only part of it.
Say it: "Three options at each of six choices is 1,093 scenes, which we cannot afford. Fold back after every second choice and keep the difference in flags."
https://gamecraft.dqnotes.com/story/branching-explosion

### Foldback
Also: Branch and bottleneck, Bottleneck, Gated branching.
Branches rejoin at a shared scene, so the writing stays affordable. Flags carry what you did through the join. Chris Crawford named it, and he thought it cheated.
It is how nearly every story-heavy game pays its bills. Without flags, though, a foldback quietly erases the choice before it, which was Crawford's complaint.
Not illusion of choice: With a foldback, the paths really do differ before they rejoin, and flags can carry that forward. An illusion of choice never differed at all.
Seen in: In The Walking Dead by Telltale, you save either Carley or Doug in the first episode. Whichever you saved is shot in the third, and the story carries on along one line.
Say it: "Fold both paths back at the gate scene. Keep a "paid" flag and a "showed letter" flag, and write one extra line in the gate scene for each, so it still feels like your journey."
https://gamecraft.dqnotes.com/story/foldback

### Illusion of choice
Also: Fake choice, Choice without consequence.
The player picks between options that all lead to the same thing, with the difference limited to a line or two. It feels like a decision until someone replays it.
Used well it is cheap and gives a scene some personality, since a player choosing a tone is doing something real. Discovered, it can sour every other choice in the game.
Not agency: Agency changes what happens next. An illusion of choice changes how the same thing felt while it happened.
Seen in: The Walking Dead by Telltale puts "will remember that" on screen after your choices. It makes each one feel recorded, while most of the plot goes on the same way.
Say it: "These three replies are an illusion of choice, and that is fine: let each one set the tone of the next line, but do not promise on screen that anyone will remember it."
https://gamecraft.dqnotes.com/story/illusion-of-choice

### String of pearls
Also: Rivers and lakes, Beads on a string.
A fixed sequence of open areas. Inside each pearl you can roam and do things in any order. The string between them is the plot, always in the same order.
It gives the player freedom where it is cheap and keeps control where the story needs it. The writer knows what has happened at each beat of the plot, every time.
Not linear: A corridor gives you no room to wander at all. A string of pearls is just as fixed in its order, but with room inside each stop.
Seen in: BioShock moves you through Rapture one district at a time in a fixed order, and within each district you can go back and forth to search and scavenge before moving on.
Say it: "Build the story as a string of pearls: five explorable zones, in a fixed order, with the plot only advancing at the door between zones. Side content stays inside its zone."
https://gamecraft.dqnotes.com/story/string-of-pearls

### Hub and spoke narrative
Also: Hub structure, Spoke and hub.
A central place the player keeps returning to, with separate stories leading off it that can be taken in any order.
Choosing your own order is real freedom at low cost. But if the spokes refer to one another, the writing has to allow for every combination, so spokes tend to stay sealed off from each other.
Not open world: An open world has no required centre. A hub and spoke sends you back to one place between stories, which is also where the writer can catch up with you.
Seen in: Mass Effect lets you take Therum, Feros and Noveria in any order, returning to your ship between each. That ship is where your crew can react to what you have just done.
Say it: "Make the ship a hub with four spoke missions. Each spoke sets one flag, and the crew get one line per flag, so we write 4 lines and not one for each of the 16 combinations."
https://gamecraft.dqnotes.com/story/hub-and-spoke

### State tracking
Also: Flags, Story variables, Quality-based narrative.
The game keeps a list of what has happened: did you pay, who did you lie to, how many times have you died. Scenes then read that list to decide what to say.
One flag can buy a whole scene's worth of memory for one extra line. Forget a flag and a character will thank you for something you never did, which players do notice.
Not save file: A save file stores the whole game so that you can come back to it. Flags are the story's own memory, and a save just happens to contain them.
Seen in: Undertale keeps some memories even if you reset your save: Flowey can tell you what you did on a timeline you just erased. Mass Effect 2 and 3 read your earlier save and bring your decisions forward.
Say it: "Set a flag "lied_to_warden" when they lie at the gate. In the prison scene, if that flag is set, the warden skips the greeting and says "I know what your word is worth.""
https://gamecraft.dqnotes.com/story/state-tracking

### Dialogue tree
Also: Conversation tree, Dialogue wheel, Branching dialogue.
A conversation laid out as lines and replies: you pick what to say, the character answers, and new options open or close depending on what has been said.
It gives a conversation shape and a way to hide information behind questions. Done badly it becomes a menu you click through until the options run out.
Not barks: In a dialogue tree you stop and choose what is said. A bark fires by itself while you carry on playing.
Seen in: Firewatch runs on one: you answer Delilah over a walkie-talkie while you walk, against a timer, and you can let it run out and say nothing. Disco Elysium's white checks can be retried later, and its red checks cannot.
Say it: "Keep this dialogue tree to three questions. Asking about the river opens a follow-up, and "Who drowned?" only appears after that. Leave one option open that can end the talk at any time."
https://gamecraft.dqnotes.com/story/dialogue-tree

### Barks
Also: Ambient dialogue, Reactive dialogue, Dynamic dialogue.
Short lines characters say by themselves in response to what is happening: "Reloading!", "That was close." The trigger comes from game state, not from a script.
Barks make the world feel as if it is paying attention, for very little writing per line. If they repeat, they become the most annoying thing in the game, so they need cooldowns and memory.
Not cutscene: A cutscene plays once at a planned time. A bark can fire at any moment, again and again, so it must make sense wherever the player happens to be.
Seen in: Left 4 Dead matches hundreds of world facts against thousands of lines and picks the most specific rule that fits, as Valve's Elan Ruskin explained at GDC in 2012. Hades has about 21,000 voice lines, so characters rarely repeat themselves.
Say it: "Add companion barks fired from state: low health, a second hit within 5 seconds, a coin found. Choose the most specific rule, wait at least 8 seconds between barks, and never play a line twice in a row."
https://gamecraft.dqnotes.com/story/barks

### Cutscene
Also: Cinematic, In-engine scene.
A stretch of the game that takes control away and plays like a film. It gives you exact framing and timing, at the cost of the player doing nothing.
It is the most reliable way to land a beat, since nothing can go wrong. It is also a point where the game stops being a game, so players skip them if there are too many.
Not scripted event: A scripted event plays out while you keep control and can look away. A cutscene takes the controller from you, so you cannot miss it, and you cannot do anything either.
Seen in: Metal Gear Solid 4 holds the Guinness record for the longest cutscene in a game, at 27 minutes. Half-Life went the other way and tells its story with the camera staying in your own eyes.
Say it: "Do not cut away for the bridge collapse. Let the player keep control, drop the planks behind them on a timer, and save the one cutscene for the moment nothing can be done."
https://gamecraft.dqnotes.com/story/cutscene

### Quick time event
Also: QTE, Press X to not die.
A button prompt pops up during a scene, and you have a moment to press it. The name comes from the "quick timer events" in Shenmue's manual, though Dragon's Lair played this way back in 1983.
It keeps hands on the controller through a scene that would otherwise be a cutscene. The catch is that the player is watching for the prompt and not at the scene, so they miss what you staged.
Not rhythm game: A rhythm game teaches a pattern you can learn and get better at. A QTE usually asks you to react to something you could not have predicted.
Seen in: Resident Evil 4 builds a whole knife fight with Krauser out of prompts in a cutscene, and failing one kills you. It also put them into scenes you thought you could put the controller down for.
Say it: "No QTE for the boulder. If it stays, show the button a full second ahead, allow 900 ms to press, and let a miss cost health, not a reload."
https://gamecraft.dqnotes.com/story/quick-time-event

### Diegetic
Also: In-world, Inside the fiction, Diegesis.
Something that exists inside the story's world, so the characters could see or hear it. A radio playing in the room is diegetic. The film score is not.
Moving information into the world keeps players looking at the game and not at a corner of the screen. It is also harder to read at a glance, so it can cost you clarity.
Not non-diegetic: A health bar on the screen is non-diegetic: the hero cannot see it. Non-diegetic is not worse, and it is often clearer.
Seen in: Dead Space puts Isaac's health in a strip of lights down the spine of his suit, and his ammo count is shown on the gun. There is no health bar on the screen.
Say it: "Make the health readout diegetic: a strip of ten lights down the character's back that goes out from the top, and no bar on the screen. Keep the camera behind them so it stays visible."
https://gamecraft.dqnotes.com/story/diegetic

### Avatar versus character
Also: Silent protagonist, Player character, Blank slate.
An avatar is a blank the player fills: they are the hero. A character is someone else with their own past and voice, whom the player steers. A silent protagonist is the usual way to keep the blank blank.
The choice decides what your writer is allowed to do. An avatar can never surprise the player with a feeling, and a character can say something the player would never have chosen.
Not customisation: Choosing a face does not make an avatar. Commander Shepard can look like anyone and still has a voice and a will of her or his own.
Seen in: Gordon Freeman in Half-Life and Chell in Portal never speak, so the other characters do the talking at you. Geralt in The Witcher 3 is fully a character: he has a voice and a history, and you decide what he does about it.
Say it: "She is a character, not an avatar. Give her a voice and a sister she misses, and only offer the player choices she might really make."
https://gamecraft.dqnotes.com/story/avatar-vs-character

### Narrative design
Also: Narrative designer, Game writing (the neighbouring job).
Deciding how the story works as a game: what the player learns when, what a choice can change, which systems carry which beats. Game writing is producing the words themselves.
Studios that treat these as one job tend to hire a writer too late and hand them a finished game with gaps in it. The structure is what decides whether any of the good lines can land.
Not game writing: The writer decides what the character says. The narrative designer decides that a character should speak here, what triggers it, and what happens if the player has already walked off. Small teams have one person do both.
Seen in: Inkle's 80 Days was written by Meg Jayanth, on top of the studio's own scripting language. The words and the machine deciding which words you see are two separate pieces of work.
Say it: "We need narrative design before we hire a writer: a list of every flag, what each quest can change, and where the player can walk off in the middle of a line."
https://gamecraft.dqnotes.com/story/narrative-design

### Quest
Also: Mission, Side quest, Task.
A unit of story with a goal, some obstacle and an outcome, tracked by the game. A side quest is one the main plot does not require.
Quests are how a large game hands out story in pieces the player can pick up and put down. A good side quest has its own small arc and leaves a mark on the world, and a poor one is a shopping list.
Not objective: An objective is one step: go here, kill that. A quest is the whole story those steps belong to.
Seen in: Mass Effect 2's loyalty missions are optional side quests, one for each crew member. Whether you did them helps decide who survives the final mission, so skipping the side stories has a real cost.
Say it: "Every side quest needs to change one thing in the world: a shop opens, a road is safer, or someone is gone. If it changes nothing, cut it."
https://gamecraft.dqnotes.com/story/quest

### Fetch quest
Also: Go-and-get quest, Courier quest.
A quest whose whole content is going somewhere, getting a thing and bringing it back.
They are cheap and they get the player out into the map, which is why every game has them. But walking is not a story, and with no turn in the middle the player has nothing to decide and nothing to remember.
Not trading sequence: A trading sequence chains fetches into one long puzzle, where working out the next swap is the point. A plain fetch quest gives you nothing to work out.
Seen in: Late in The Wind Waker, the Triforce hunt makes you find charts, pay Tingle to decode them and then go looking for each piece, and it is often cited as the game's low point.
Say it: "This fetch quest needs a turn. When the player reaches the well, the bucket is in use by someone who needs it more, and the player has to decide what to do."
https://gamecraft.dqnotes.com/story/fetch-quest

### Three-act structure
Also: Inciting incident, Setup, confrontation, resolution.
A story in three parts: set up a normal life until an inciting incident breaks it, struggle with the consequences, then resolve them. Syd Field made the shape standard in screenwriting.
It gives you a schedule for what the player should feel, and a way to spot a game with a good first ten hours and nothing to resolve. The inciting incident is the most useful piece: if nothing has broken yet, the player has no reason to care.
Not hero's journey: Three acts is a schedule for any story. The hero's journey is one particular plot, with particular events in it, that can fit on that schedule.
Seen in: Portal is close to a clean version: test chambers as the setup, your escape from the fire at chamber 19 as the turn, and the fight with GLaDOS as the end. In Half-Life, the resonance cascade is the inciting incident.
Say it: "The inciting incident arrives too late. Keep the first act to 20 minutes: the village is attacked at the end of the first quest, and the player cannot go home after that."
https://gamecraft.dqnotes.com/story/three-act-structure

### Hero's journey
Also: Monomyth.
A plot pattern set out by Joseph Campbell in The Hero with a Thousand Faces (1949): a call to adventure, a crossing into the unknown, an ordeal, and a return home changed.
It is a dependable skeleton for a game about getting stronger, since levelling up and growing up line up well. It is also so familiar that following it step by step makes a story feel generic.
Not three-act structure: Three acts tells you when things should happen. The hero's journey also tells you what should happen.
Seen in: Jenova Chen has said the idea behind Journey was to put two players through the hero's journey together. The mountain gives you the call, the descent and the climb without one word of dialogue.
Say it: "Use the hero's journey as a skeleton: the call comes in the first ten minutes, the ordeal at about 70% of the way, and the return shows the starting village again with everything that has changed."
https://gamecraft.dqnotes.com/story/heros-journey

### Kishotenketsu as a story form
Also: Four-act structure, Ki-sho-ten-ketsu.
A four-part form from Chinese and Japanese writing: introduce, develop, twist, reconcile. The drama comes from a turn that changes how you see the earlier parts, so it needs no conflict at all.
It is a good answer to the idea that every story needs a villain. It suits small stories and puzzles, and short games can use it to end on understanding, not on a fight.
Not three-act structure: Three acts run on conflict and its resolution. Kishotenketsu runs on recontextualising: nothing has to be in conflict for the third part to change everything.
Seen in: Koichi Hayashida, director of Super Mario 3D Land, has cited it as a model for level design, using the form of four-panel comics: a new idea, more of it, a twist on it, and a last flourish.
Say it: "Build this chapter as kishotenketsu: four short scenes of her daily routine, the third reveals who the meals were for, and the fourth repeats the first scene word for word."
https://gamecraft.dqnotes.com/story/kishotenketsu

### In media res
Also: Into the middle of things, Cold open.
Starting the story partway through the action and explaining how you got there later. The Latin means "into the middle of things" and comes from Horace.
Players will forgive any amount of confusion if something is happening, and they will not forgive a slow start. It also buys you a question, how did this happen, that carries the next few hours.
Not flashback: A flashback is a scene set earlier. In media res is a choice of where to start. It usually needs a flashback afterwards to fill things in.
Seen in: Uncharted 2 opens with a wounded Drake waking up in a train car hanging off a snowy cliff, and you climb out of it before the game goes back to explain how he got there.
Say it: "Start in media res: the player is already hanging from the wrecked train with the controls live. Go back to the museum job afterwards, and cut the first 15 minutes of walking."
https://gamecraft.dqnotes.com/story/in-media-res

### Stakes
Also: What is at risk, What you stand to lose.
What will be lost if the hero fails, and what makes that matter to this particular player. Small and personal usually beats vast and abstract.
A world in danger has been used so often that it means nothing. A dog the player has been feeding for ten hours means a great deal, and in a game it has to be something the player knows they can actually lose.
Not macguffin: The MacGuffin is the thing everyone is chasing. The stakes are what happens if they fail to get it.
Seen in: Mass Effect 2's final mission can kill any of your crew for good, including Shepard, and the dead stay dead when you bring your save into the next game. The whole game has been getting you ready for that risk.
Say it: "Raise the stakes: the companion the player has spent five hours with can die in this fight, permanently, and it depends on whether they finished the side quest."
https://gamecraft.dqnotes.com/story/stakes

### Chekhov's gun
Also: Plant and payoff, Setup and payoff, Foreshadowing.
If something is made prominent early, it should matter later. It comes from Anton Chekhov's advice that a loaded rifle on stage has to go off.
A setup is what makes a payoff feel earned, and a payoff with no setup feels like cheating. In a game it is also a promise: players do go back to try the thing you showed them.
Not red herring: A Chekhov's gun pays off. A red herring is planted on purpose to mislead you. A gun you forgot about is neither, and is just a loose end.
Seen in: BioShock has Atlas say "would you kindly" over and over from the first minutes, as ordinary politeness, and it means something else entirely when the game finally explains it.
Say it: "If the rope in the finale saves them, show it in the first room and let the player try it there. Don't let it appear for the first time when they need it."
https://gamecraft.dqnotes.com/story/chekhovs-gun

### MacGuffin
Also: Plot coupon, The thing everyone wants.
An object that everyone in the story is chasing, whose actual nature hardly matters. Alfred Hitchcock made the word famous.
It gets characters moving and in each other's way without any explanation. If players start asking what it actually does, you have either a plot problem or a better story than you thought.
Not chekhov's gun: Chekhov's gun has to do something when the time comes. A MacGuffin only has to be wanted.
Seen in: Uncharted 2 sends everyone racing after the Cintamani Stone, and it matters mostly because the people with guns want it before Drake gets to it.
Say it: "The stone is a MacGuffin, so do not write a lore page for it. Spend those words on why these three people are willing to kill each other over it."
https://gamecraft.dqnotes.com/story/macguffin

### Dramatic irony
Also: Audience knows more, The bomb under the table.
The audience knows something the characters do not. Alfred Hitchcock's example: a bomb under a table is fifteen seconds of shock if it is a surprise, and fifteen minutes of suspense if you have seen it.
It turns boring scenes into tense ones at no cost. Telling the player early is often worth more than saving a twist, because a twist only works once.
Not twist: A twist hides something from you until the moment it pays off. Dramatic irony shows you first, then makes you watch people who do not know.
Seen in: What Remains of Edith Finch tells you at the start that this family dies young. Each playable story is then spent inside someone happily unaware of the end you are expecting.
Say it: "Show the player the bomb in the first shot. Then let the two characters have their small talk for a full 15 seconds with the timer on screen."
https://gamecraft.dqnotes.com/story/dramatic-irony

### Unreliable narrator
Also: Lying narrator.
The one telling you the story is wrong, biased or lying, and you are meant to catch them at it.
In a game, the player has a second witness: the game they are playing. When the voice and the play disagree, the gap can be the most interesting thing in the story.
Not ludonarrative dissonance: Dissonance is an accidental disagreement between rules and story. An unreliable narrator is a disagreement you planned, and the player is meant to notice it.
Seen in: In Call of Juarez: Gunslinger, old Silas Greaves tells his exploits in a saloon as you play them. When he is challenged or changes his mind, the level changes around you to match the new version.
Say it: "Have the narrator say the leap was perfect while the player is climbing out of the ditch. Never correct him on screen, and let the player catch him."
https://gamecraft.dqnotes.com/story/unreliable-narrator

### Morality system
Also: Karma, Good and evil meter, Paragon and Renegade.
The game scores your actions on a good-and-evil scale, then changes how the world treats you. It turns ethics into a number, and players treat numbers as things to win.
It is cheap to build and gives clear feedback, but once a reward sits at the end of the bar, players stop asking what is right and start asking what the bar wants.
Not reputation: Morality scores you against the game's idea of right and wrong. Reputation is what particular people think of you, and two groups can disagree about the same act.
Seen in: In Mass Effect, high Paragon or Renegade scores open conversation options, which encourages you to pick one colour and stay with it. BioShock's Little Sisters are a simpler moral test whose result decides the ending.
Say it: "Drop the karma bar. Track what they did as specific flags, so helping a beggar does not undo a murder, and let the people involved react to the thing they saw."
https://gamecraft.dqnotes.com/story/morality-system

### Multiple endings
Also: Branching endings, True ending.
The story can finish in more than one way, depending on what the player did. The choice may be made in the final minute or built up over the whole game.
It rewards replaying and gives choices a place to land. The cost: most players see one ending, so the rest of the work is hidden, and a last-minute choice lets someone reload and see them all in a minute.
Not new game plus: Multiple endings are different ways for the story to finish. New Game Plus is a way to start over and keep things. Some games use the second to reach the first.
Seen in: Nier: Automata has 26 endings named A to Z: five that carry the story, and the rest mostly jokes and sudden deaths. Undertale's ending depends on how many monsters you killed over the whole game.
Say it: "Do not make the ending a final menu. Add up three flags from across the game and pick one of four endings from that, so nobody can reach another ending by reloading in the last room."
https://gamecraft.dqnotes.com/story/multiple-endings

### New Game Plus
Also: NG+, Second playthrough.
After finishing the game, you start over and keep something from the first run, such as your levels, items or knowledge.
It asks very little of you and doubles the life of everything you have built. It works best when the second run reveals new things, and doesn't just inflate the enemies' numbers.
Not multiple endings: New Game Plus changes how you begin. Multiple endings change how you finish.
Seen in: Chrono Trigger gave the option its name in 1995, and you need it to see most of its endings. Nier: Automata goes further: its later playthroughs show you the same events from another character's side.
Say it: "Add New Game Plus: keep levels and the cellar key. The key opens a door that sits locked through the first run, and characters who remember you get a new line."
https://gamecraft.dqnotes.com/story/new-game-plus

## 8. Worldbuilding

How to make a place feel older and larger than what the player will ever see.

### Setting
Also: Backdrop, Time and place.
Where and when the game takes place, and the conditions of life there. It decides what can happen, who is around and what everyone takes for granted.
A good setting generates plot and rules for free. A city under the sea explains its leaks, its isolation and its madness without a line of dialogue.
Not lore: The setting is what you can see and stand in. Lore is what is known or written about it.
Seen in: BioShock is set in Rapture, an underwater city built for people who wanted no government above them, and by 1960 it is falling apart. The setting is the argument the whole game is having.
Say it: "Change the setting from a generic castle to a salt mine that is still working. Now every room has a reason to be underground, and every guard has a reason to be tired."
https://gamecraft.dqnotes.com/worldbuilding/setting

### Lore
Also: Backstory, Mythos.
Everything known about the world beyond the immediate plot: its history, gods, wars and customs. Players usually meet it in pieces and mostly optionally.
Lore rewards the players who want more without holding up those who do not. It becomes a problem when you require it, or when you write it instead of making the game.
Not story: The story is what happens during the game. Lore is everything that had already happened, and the player can miss it and still finish.
Seen in: Dark Souls tells almost none of its history directly. Fans pieced it together from item text and dialogue, and arguing over it became a large part of how the game is enjoyed.
Say it: "Keep lore optional: nothing on the main path needs it, and every entry should be short enough to read in the time it takes to pick up the object."
https://gamecraft.dqnotes.com/worldbuilding/lore

### Canon and retcon
Also: Canon, Retcon, Retroactive continuity.
Canon is what officially counts as having happened in the world. A retcon changes that after the fact, so something established is now said to have been different.
Every fact you publish is a thing later work must agree with. A retcon can save a series, but its cost is everything that was built on the old fact, and fans keep count.
Not plot hole: A plot hole is an accident that nobody has addressed. A retcon is a deliberate change that says: this is how it was all along.
Seen in: The Elder Scrolls II: Daggerfall has several endings, so the series later decided they all happened: a "Dragon Break" called the Warp in the West let time itself split, turning a problem of canon into part of the lore.
Say it: "Check what depends on this before we retcon it. If the king was never poisoned, three quests and the whole of Act 2 need new reasons, so I would rather add a second poisoner."
https://gamecraft.dqnotes.com/worldbuilding/canon-and-retcon

### World bible
Also: Story bible, Series bible, Lore document.
The reference document that holds the true answers about your world: the timeline, the names, the rules, how things are spelled. It is written for the team, not for players.
Two writers will each answer an unasked question differently, and the bigger the team, the sooner it happens. A bible settles it once, as long as someone keeps it up to date.
Not game design document: A design document says how the game works and what you will build. A world bible says what is true in the fiction, whether or not any of it is ever shown.
Seen in: Halo has a story bible, started at Bungie and carried on at 343 Industries, kept so that the games and books agree. Outside authors are sent the parts they need for their own work.
Say it: "Before the two writers start, put a one-page world bible together: the calendar, how magic is paid for, who rules what, and how every place name is spelled. Whoever finds a gap fills it in the bible first."
https://gamecraft.dqnotes.com/worldbuilding/world-bible

### Iceberg principle
Also: Theory of omission, Iceberg theory.
Know far more about your world than you show. Hemingway's version: an iceberg moves with dignity because only one-eighth of it is above water, and the rest can be felt.
What is left out does work, as long as someone knows what it is. Hemingway's own warning was that leaving out what you never knew only makes hollow places.
Not vagueness: Vagueness means there was nothing behind the line. The iceberg means a specific thing is behind it, and that is why the line can be so short and so sure of itself.
Seen in: A Dark Souls item description might give a few sentences about a knight, naming a lord and a betrayal it never explains. Enough other items agree with it that you can tell the rest of it exists.
Say it: "Apply the iceberg principle here: write all twelve facts about the siege in the bible, and let the player see only the one on the sword: "Its owner never came back to collect it.""
https://gamecraft.dqnotes.com/worldbuilding/iceberg-principle

### Internal consistency
Also: World logic, Consistent rules.
The world obeys its own rules every time, whatever those rules are. A world can be as strange as you like, as long as it stays strange in the same way.
Players rely on the rules to plan. Break one for the sake of a scene and they lose more than the scene, since they stop trusting everything else you have told them.
Not realism: Realism is matching our world. Consistency is matching your own. Gravity can reverse every Tuesday, provided it really does so on every Tuesday.
Seen in: The solar system in Outer Wilds runs the same way on every 22-minute loop: the sand moves between the twin planets on the same schedule, and Brittle Hollow breaks up the same way. Because it is so consistent, you can plan around it.
Say it: "The bell always rings two seconds before the flood. No exceptions, not even for the boss. If we want a surprise, we build it out of that rule, by making the player unable to hear the bell."
https://gamecraft.dqnotes.com/worldbuilding/internal-consistency

### Suspension of disbelief
Also: Willing suspension of disbelief.
The player agrees to accept things that are not true, in exchange for a story. Samuel Taylor Coleridge named it in 1817.
It is a loan, not a gift. Players will accept one huge impossible thing with no fuss, and then walk out over a small careless one, because the small one shows nobody was paying attention.
Not verisimilitude: Suspension of disbelief is what the player does for you. Verisimilitude is what you do to make that easy.
Seen in: Portal asks you to accept a gun that makes holes in space, then takes care over a smaller question: Chell wears springs on her heels, the reason a fall does not hurt her. Portal 2 calls them Long Fall Boots.
Say it: "Nobody minds the dragons. They mind that the guards ignore you carrying a ladder through the throne room. Fix the guards and the dragons stay fine."
https://gamecraft.dqnotes.com/worldbuilding/suspension-of-disbelief

### Verisimilitude
Also: The appearance of truth, Believability.
The quality of seeming true, which is not the same as being true. Small, specific details make a made-up thing feel as if it has been observed.
Players cannot check your world against reality, so they judge it by how sure of itself it sounds. One odd specific detail convinces more than a page of accurate summary.
Not realism: Realism is about matching the facts. Verisimilitude is about feeling right, and it sometimes means getting the facts wrong: a film gunshot sounds nothing like a real one, and people believe the film.
Seen in: In Red Dead Redemption 2, Arthur's hair and beard keep growing, and camp life goes on around you, chores and all. None of it is needed for the plot, and it is why people believe in the place.
Say it: "Do not research the real price of grain. Just give each shopkeeper one specific complaint, like the mule that bit her or the tax she refuses to pay, and they will read as real."
https://gamecraft.dqnotes.com/worldbuilding/verisimilitude

### Tone
Also: Mood, Register.
The attitude a work takes to its own events: grim, silly, cosy, cruel. The same facts can be told in any tone and produce a different game.
Everyone on the team makes tone decisions every day, in every line and texture, whether they know it or not. If nobody names it, the game wanders from one to the next and players feel it as cheapness.
Not genre: Genre is what kind of thing it is. Tone is how it feels about being that thing. Horror can be deadly serious or a comedy.
Seen in: Fallout tells a story about nuclear ruin in the voice of a cheerful 1950s advert, with Vault Boy smiling and giving a thumbs up. The joke and the horror depend on each other, and the tone is the series' identity.
Say it: "Fix the tone: dry and quietly funny, like someone who has seen worse. No puns during the funeral. Jokes come from people being stubborn, not from winking at the player."
https://gamecraft.dqnotes.com/worldbuilding/tone

### High concept
Also: Hook, Elevator pitch, One-line premise.
A premise so clear it can be understood and wanted from one sentence. It gives everyone on the team the same picture of the game.
If you cannot say it in a sentence, you will struggle to sell it, and your team will each build a different game. It also works as a filter, since any feature that does not serve the line is suspect.
Not tagline: A tagline is a piece of marketing that can be written after the game is finished. A high concept is the idea the game is made from.
Seen in: Superhot: time moves only when you move. Papers, Please: you are a border inspector in a failing state, checking passports. In each case the one line tells you the verbs, the tension and why you would want to play.
Say it: "Pin the high concept to one sentence: "You run a ferry across the river of the dead, and you cannot refuse anyone." If a feature does not serve that line, it goes."
https://gamecraft.dqnotes.com/worldbuilding/high-concept

### Factions
Also: Guilds, Reputation groups, Sides.
Groups with their own goals, beliefs and grudges, often against each other. Helping one usually costs you with another.
Factions turn a backdrop into politics, and they make your choices cost something. They are also an efficient way to write a world: one good disagreement can drive dozens of quests.
Not morality system: A morality meter has one judge. Factions have many, and an act that one group loves is the reason another group shoots you.
Seen in: Fallout: New Vegas sets the NCR, Caesar's Legion and Mr. House against one another over one city, and each faction keeps its own opinion of you. Yes Man offers a fourth way, which is to push all of them out.
Say it: "Give the river towns three factions: the ferrymen, the bridge builders and the toll office. Each favour moves your standing with all three, and no one can be friends with all of them by the end."
https://gamecraft.dqnotes.com/worldbuilding/factions

### Hard and soft magic
Also: Sanderson's First Law, Magic systems.
Hard magic has clear rules the audience knows. Soft magic stays mysterious. Brandon Sanderson's First Law: an author's ability to solve conflict with magic is directly proportional to how well the reader understands it.
If players know the rules, a clever solution is theirs to anticipate and it feels earned. If they do not, the same solution feels like the author cheating. Soft magic is fine, as long as it creates problems instead of solving them.
Not science fiction versus fantasy: This has nothing to do with genre or how serious the magic looks. It is only about how much the audience can predict.
Seen in: Divinity: Original Sin 2 has very hard magic: fire ignites oil, rain makes water, and lightning shocks anyone standing in that water. Once you have seen those rules, you can plan combos the designers never specifically wrote.
Say it: "The player must be able to solve the final puzzle with the rules we have taught. State the spell's three limits in the first hour, then build the finale so the third limit is the answer."
https://gamecraft.dqnotes.com/worldbuilding/hard-and-soft-magic

### History layers
Also: Strata, Layered ruins, Deep time.
A place that shows several past eras at once: a modern town on an older town on something older still, each leaving marks the next one built around.
Real places make no sense until you know what used to be there, and that is why they feel real. An oddity with a cause, like a street that bends for no reason, suggests a history you do not need to narrate.
Not timeline: A timeline lists what happened. Layers are what is left of it, sitting where the player can touch it.
Seen in: Breath of the Wild's Hyrule shows at least three ages at once: the abandoned towns from the Calamity 100 years ago, the Sheikah shrines from 10,000 years ago, and the villages people have rebuilt since.
Say it: "Give this town three layers. The market street bends because the old river did, the church is built on the old fort's foundation, and the well is older than both. Nothing needs a label."
https://gamecraft.dqnotes.com/worldbuilding/history-layers

### Ecology
Also: Ecosystem, Food chain.
What everything in the world eats, and what eats it. A place with real ecology has creatures with reasons to be where they are, doing things unrelated to you.
A monster that stands in a room waiting for the player is a target. A monster that is hunting something else is a part of a world, and it gives the player a whole lot to observe and use.
Not enemy placement: Enemy placement decides where to put a threat for the player. Ecology decides where a creature would be if the player did not exist.
Seen in: Rain World's creatures search for food and shelter on their own, with their AI running even when they are off screen. In Monster Hunter: World, large monsters fight each other over territory in the "turf wars", and you can use it.
Say it: "Give the swamp a food chain: frogs eat flies, herons eat frogs, and the big predator hunts herons at dawn. Let the player find that out by watching, and use it as bait."
https://gamecraft.dqnotes.com/worldbuilding/ecology

### Used future
Also: Lived-in world, Dirty sci-fi, Wear and tear.
A world whose things are old, dented and repaired, so they seem to have been in use for years. George Lucas used the phrase about Star Wars in 1977, and it applies to any setting, not only the future.
New things look like props. Wear is evidence of people: where they walk, what breaks often, what no one bothered fixing. It suggests a past without saying a word.
Not post-apocalypse: A post-apocalypse is a world that has ended. A used future is a world still at work, with things simply getting old in use.
Seen in: Alien: Isolation set itself the rule of building nothing that could not have been built on the set of the 1979 film. Its chunky CRT monitors and worn corridors feel like a real ship owned by a mean company.
Say it: "Make this ship feel lived in. Wear the floor where the crew walk from bunks to galley, tape the kettle's handle, and leave the panel that always breaks with its screws missing."
https://gamecraft.dqnotes.com/worldbuilding/used-future

### Naming
Also: Naming language, Phonotactics, Name generator.
Naming people and places using a consistent set of sounds, so that they sound as if they come from one culture. The rules for which sounds can sit together are called phonotactics.
Players read a culture from its names before anything else. Random fantasy names sound like random fantasy names, and one small rule set makes two peoples sound different from one another.
Not conlang: A naming language only needs sounds and a few word pieces, and it can be built in an afternoon. A conlang also has grammar and vocabulary, enough to write and speak in.
Seen in: The Elder Scrolls: Balmora, Sadrith Mora and Ald'ruhn in Morrowind sound like one people, and Whiterun, Windhelm and Riften in Skyrim like another. You can tell which province you are in by the sign.
Say it: "Build a naming language for the coast people: only soft consonants, syllables that end in a vowel or n, and 2 to 3 syllables. Every place and person on the coast gets named from that."
https://gamecraft.dqnotes.com/worldbuilding/naming

### Conlang
Also: Constructed language, Fictional language.
A language invented for a work, with its own sounds, grammar and words. Real ones are also given a history, with sound changes that split one old tongue into several new ones.
Most games do not need a full one, but a little goes a long way: shared roots across place names, or two dialects that are clearly related, can suggest a thousand years of history.
Not cipher: A cipher swaps the letters of a real language, so a player can decode it back to English. A conlang is its own language, with its own words and rules.
Seen in: Far Cry Primal's Wenja was built for the game by the linguists Andrew and Brenna Byrd, from reconstructed Proto-Indo-European, and the actors learned to perform in it. In Skyrim, the dragon shouts are words in the dragons' own language.
Say it: "We do not need a full conlang. Make 40 root words and two sound-change rules, so the northern and southern place names look related but clearly split a long time ago."
https://gamecraft.dqnotes.com/worldbuilding/conlang

### In-world documents
Also: Audio logs, Item descriptions, Found text, Collectible lore.
Letters, audio logs, signs and item descriptions that exist inside the fiction. Each is written by someone, for someone, for a reason, and that is what makes them feel found and not delivered.
The player can take them or leave them, so they do not hold up anyone in a hurry. Better still, several documents that do not quite agree give players the pleasure of working out what actually happened.
Not codex: A codex entry is written for the player by the game, like an encyclopedia. An in-world document was written by someone in the story, who may be lying.
Seen in: Dark Souls hides its history in item descriptions. Its director Hidetaka Miyazaki has said that as a child he read books too hard for him, and that his imagination filled the gaps. BioShock tells Rapture's fall in audio diaries you find on the ground.
Say it: "Put the backstory in six item descriptions. No item tells the whole story, two of them contradict each other about the fire, and each stays under 40 words."
https://gamecraft.dqnotes.com/worldbuilding/in-world-documents

### Cultural texture
Also: Customs, Everyday culture, Anthropology.
The ordinary habits that make a people particular: what they eat, how they greet each other, what is rude, what they swear by. It is mostly small and rarely explained.
Without it, every town is the same town with different signs. Details that nobody explains make a culture feel as if it existed before you came in, and does not need you.
Not faction: A faction is a group that wants something. Culture is how people live whatever they want, and one culture can hold several rival factions.
Seen in: Morrowind's Dunmer live in hollowed-out giant crab shells and grown mushroom towers, travel on enormous insects, and speak of living gods, and the game does not stop to explain it as strange.
Say it: "Give the river people three customs players will see without any explanation: they never say goodbye, they pay in salt, and they leave one seat empty at every table."
https://gamecraft.dqnotes.com/worldbuilding/cultural-texture

### The map as a story
Also: Map logic, Geography as history, Fantasy map.
Geography explains history: towns grow at river crossings, borders follow mountains, roads go around the marsh. A map that follows those reasons tells a story before anyone speaks.
Players read maps as they read faces. A town placed for no reason feels like a level marker, and one sitting on the only ford for 50 miles comes with its own wars, tolls and rich families.
Not level layout: Level layout is about how a space plays. A map as a story is about why anyone would have settled here at all.
Seen in: Hollow Knight makes its maps part of the place: you buy them from Cornifer, a cartographer you track down in each area, by following the trail of paper he leaves behind.
Say it: "Put the towns where people would have settled: at the river mouth, at the one ford and at the mountain pass. Then the road between them explains itself, and so does why they hate each other."
https://gamecraft.dqnotes.com/worldbuilding/map-as-story

### Fridge logic
Also: Icebox scene, Plot hole you notice later.
Something that seemed fine at the time, and only falls apart later, when you are at the fridge and have stopped to think. Alfred Hitchcock is credited with calling these icebox scenes.
Pace can hide a hole, which is useful, since some holes are too costly to fill. It only works in the moment, though, and players now take their theories online.
Not plot hole: A plot hole can be spotted while it is happening. Fridge logic is specifically the kind you get away with until afterwards.
Say it: "This is fridge logic: nobody asks why the guards did not just lock the door while the building is on fire. Keep the scene moving at this pace, and don't stop to linger on the door."
https://gamecraft.dqnotes.com/worldbuilding/fridge-logic

### Scale
Also: Compressed space, Game scale.
How big the world claims to be against how big it actually is. Game worlds are heavily compressed: a capital of thousands is a few streets, and a kingdom takes an afternoon to walk.
Compression is fine, and it is needed, because real distances are boring. Trouble comes when the words and the space disagree: a "great city" with twelve houses reads as a joke unless you handle it.
Not map size: Map size is a number on a design document. Scale is how big the place feels, and that depends on what fills it and how long it takes to cross.
Seen in: Whiterun in Skyrim is treated as one of the great cities of its province and is only a handful of streets. Players accept it, because the game's mountains, music and sparse wilds all make the walk there feel long.
Say it: "Never give a number for the city's population. Say "the city" and show forty houses. If a book must mention it, write "a few hundred souls"."
https://gamecraft.dqnotes.com/worldbuilding/scale

## 9. Looks

Why one screenshot reads in a blink and another reads as mud: the words for art direction, light, colour, pixels and animation.

### Art direction
Also: Visual direction, Art style, Visual identity.
The set of decisions that makes every picture in a game look as if it belongs to the same game: palette, shapes, lighting, level of detail, how things move. An art director writes those rules and holds everyone to them.
Without it, ten good artists make ten good pieces that clash. A clear direction also costs less, because artists stop having to guess and a small team can look deliberate.
Not graphics: Graphics are what the hardware can draw. Art direction is what you choose to draw with it, and a strong direction often looks better on weaker hardware.
Seen in: Mirror's Edge is almost entirely white, with red marking the routes you can run. The art director's brief was reportedly to make the game recognisable at a glance, and one developer account says the clean city also answered motion sickness in the grittier early prototype.
Say it: "Art direction for this game: three main colours and one accent, rounded shapes only, no outlines, light always from the top left. Anything that breaks those four rules goes back."
https://gamecraft.dqnotes.com/looks/art-direction

### Stylisation
Also: Stylised art, Abstraction.
Deliberately leaving out and exaggerating parts of reality so that what remains says more. A stylised tree is not a worse tree; it is a choice about which few things make a tree.
Realism needs enormous budgets and dates quickly, because the bar keeps rising. A strong style has its own rules, so it is cheaper to make and tends to age well.
Not low quality: Stylised art is simple on purpose, and every remaining shape is doing a job. Low-quality art is simple because time ran out, and it usually still copies reality, just badly.
Seen in: After a realistic-looking Zelda tech demo at Space World 2000, Nintendo gave The Wind Waker flat colours and huge cartoon eyes. Fans were divided at the time, and it has since been remastered with that look intact.
Say it: "Stylise the trees: one trunk shape and two or three clumps of leaves in two flat values. No individual leaves. A player should recognise it at 64 pixels tall."
https://gamecraft.dqnotes.com/looks/stylisation

### Readability
Also: Visual clarity, Legibility, Read.
How quickly a player can tell what they are looking at and what it means for them: what can hurt me, what can I touch, where can I stand. Artists say a thing "reads" when it is understood in a glance.
When a game is hard to read, players get hit by things they never saw, and they blame the game, correctly. Beautiful detail that hides the enemy is a bug.
Not detail: Detail adds information everywhere. Readability ranks information, so the things that matter win. Adding detail usually makes it worse.
Seen in: Cuphead colours the things you can parry pink, so in a screen full of bullets the ones you can slap are picked out at once.
Say it: "Enemies and hazards have to read first. Drop the background to 40% contrast and desaturate it, keep full saturation for anything that can hurt the player, and give pickups their own colour."
https://gamecraft.dqnotes.com/looks/readability

### Silhouette
Also: Silhouette test, Outline read.
The outline of a character or object, filled in solid. The silhouette test asks whether you can still tell who it is with all colour and interior detail gone.
In a real fight, characters are small, far away and moving, and the outline is all you get. If two characters share a shape, players will mistake one for the other.
Not rim light: A silhouette is the shape itself. Rim light is one way of separating that shape from a dark background.
Seen in: Valve's 2007 paper Illustrative Rendering in Team Fortress 2 says the nine class silhouettes were carefully designed to be very distinct, because telling classes apart is the core of the game.
Say it: "Fill all four heroes in black and put them side by side. If you cannot name each one in under a second, change the outline: bigger shoulders, a hat, a long weapon. Do not fix it with colour."
https://gamecraft.dqnotes.com/looks/silhouette

### Value structure
Also: Squint test, Value, Light and dark pattern, Notan.
The pattern of light and dark in a picture, ignoring colour. Value means how bright a colour is. The squint test is the old painter's check: narrow your eyes until detail goes, and see whether the picture still holds together.
The eye reads value before hue, so a picture with weak values turns to mud however lovely the colours are. It is also what survives for players with colour blindness.
Not saturation: Saturation is how strong a colour is. Value is how light or dark it is. Two very different hues can have the same value and melt together in grey.
Seen in: Limbo has no colour at all, so everything it tells you, from depth to danger, has to be carried by value.
Say it: "Do a squint test on this level. Keep three value groups: a dark foreground for the play space, a mid-grey background, and bright only where the player should look. Check it in greyscale before adding colour."
https://gamecraft.dqnotes.com/looks/value-structure

### Focal point
Also: Point of interest, Visual hierarchy, Leading the eye.
The one place the eye lands first. You make it with the strongest contrast, the only saturated colour, an isolated shape, or lines that point at it.
A picture with no focal point makes players search, and they stop looking. In a level, a focal point is a way to give directions without an arrow.
Not waypoint marker: A marker is a sign painted on top of the world. A focal point is the world itself arranged so you look there.
Seen in: Mirror's Edge turns the next useful pipe or door red against its white city, a system it calls Runner Vision, so your eye finds the route while you are running.
Say it: "Make the tower the focal point: it is the only warm thing on screen, it sits against the brightest sky, and the road bends towards it. Everything else stays at half contrast."
https://gamecraft.dqnotes.com/looks/focal-point

### Shape language
Also: Shape design, Circles, squares, triangles.
Building a design out of one family of simple shapes so that it carries a meaning before any detail. Circles tend to read as friendly, squares as sturdy, and triangles as dangerous or quick.
It tells the player what something will do before it does it, and it gives a whole cast or world a family resemblance.
Not silhouette: A silhouette is one thing's outline and asks "can I tell who this is?" Shape language is the vocabulary behind many outlines and asks "what kind of thing is this?"
Seen in: Kirby is a circle and Bowser is all spikes and horns. You know who to trust before either of them moves.
Say it: "Use shape language: anything safe is built from circles, anything solid from squares, anything that can hurt you has a point. No round enemies, no spiky platforms that are safe to stand on."
https://gamecraft.dqnotes.com/looks/shape-language

### Palette
Also: Colour palette, Limited palette, Swatches.
The fixed set of colours a game, a level or a character is allowed to use. Old hardware forced a small one; artists now often choose one anyway.
Fewer colours make everything agree with everything else. A limited palette is one of the cheapest ways to make a small team's art look intended.
Not colour grading: A palette is chosen before you draw and limits what you can draw with. Grading is applied over the finished picture and shifts what is already there.
Seen in: A Super Mario Bros. sprite gets three colours plus transparency, the NES limit per sprite palette. Downwell chooses a similar limit by default: black, white and red.
Say it: "Limit each level to 16 colours, with the same four shadow colours everywhere. Only the player and pickups may use the accent red."
https://gamecraft.dqnotes.com/looks/palette

### Colour harmony
Also: Colour scheme, Colour wheel, Complementary colours.
Choosing colours by where they sit on the colour wheel. Neighbours (analogous) feel calm, opposites (complementary) feel tense and bright, and three spaced evenly (triadic) feel playful.
It gives you a reason for a colour, not only a taste. It also lets you plan a mood: a level can start analogous and turn complementary when things go wrong.
Not palette: A palette is the list of colours you are allowed. Harmony is the relationship that makes them work, and a palette can have good or bad harmony.
Seen in: Team Fortress 2 keys its two teams warm and cool: RED's bases are warm brown and rust, BLU's are cool steel and slate, so a glance at the walls tells you whose base you are in.
Say it: "Use a split between warm and cool: orange for the player and everything friendly, teal for the world, and nothing else above 30% saturation."
https://gamecraft.dqnotes.com/looks/colour-harmony

### Colour script
Also: Colour keys, Mood strip, Color script.
The whole story laid out as a strip of small paintings, one per beat, so you can read its emotional arc as colour before anything is built. It answers: what does each part of this feel like, and how does it change?
Each level looks fine alone, but played end to end they all feel the same. A colour script gives the game a rhythm, so the bleak part earns the sunrise.
Not colour grading: A colour script is the plan for the whole game, made first. Grading is one of the tools that carries it out, applied at the end.
Seen in: Ralph Eggleston painted Pixar's first colour script for Toy Story, in pastel on long strips of black paper. In games, Journey's palette shifts along a curve its artists set to follow the story; art director Matt Nava says they wanted colour to guide players' emotions.
Say it: "Make a colour script before we build levels: twelve thumbnails, warm and gold for the opening, grey-blue through the middle, one red peak at the boss, pale dawn at the end."
https://gamecraft.dqnotes.com/looks/colour-script

### Foreground framing
Also: Framing, Foreground elements, Natural frame.
Placing dark shapes close to the camera, such as branches, rocks or a doorway, around the edge of the view so they frame what matters.
It gives a flat picture a sense of depth and of being somewhere, and it points the eye inward. Used carelessly, it covers the action, so keep it out of the play space.
Not vignette: A vignette is an even darkening of the screen corners added afterwards. Foreground framing uses real objects in the world, which move as the camera moves.
Seen in: Ori and the Blind Forest places dark out-of-focus shapes close to the camera, and creatures even walk along the foreground before stepping onto Ori's plane.
Say it: "Add foreground framing to the vista: near-black branches in the top left and rocks along the bottom, moving at 1.5 times camera speed, blurred, and never over the player's path."
https://gamecraft.dqnotes.com/looks/foreground-framing

### Atmospheric perspective
Also: Aerial perspective, Depth haze.
Far things look paler, bluer and lower in contrast, because there is more air between you and them. Painters have used this for centuries to make a flat surface look deep.
It separates layers without outlines, and the play layer gets the contrast by default. Without it, a mountain and a nearby rock argue for the same attention.
Not distance fog: Distance fog is often there to hide what the machine cannot draw, as in the fog of the original Silent Hill. Atmospheric perspective is the natural effect that fog imitates, and it is about depth, not hiding.
Seen in: Limbo layers its grey world so each step back is paler than the last, with the near plane almost black. It has no colour to separate its layers, so this does all the work.
Say it: "Add atmospheric perspective: each background layer shifts 25% towards the sky colour and loses contrast, so the farthest ridge is barely darker than the sky."
https://gamecraft.dqnotes.com/looks/atmospheric-perspective

### Parallax
Also: Parallax scrolling, Layer scrolling.
Background layers move slower than the foreground as the camera moves, the same way that fence posts rush past a train window while hills barely shift. It fakes depth on a flat screen.
It is a cheap, strong depth cue that tells you which things are close. Too much, or layers moving at the wrong speed, makes the world feel like sliding paper or even makes people feel sick.
Not layers: Layers are the separate pictures. Parallax is only their different speeds. Layers all moving together are just one flat picture.
Seen in: Moon Patrol (1982) is often credited with introducing full parallax scrolling, with three background layers moving at different speeds. Jump Bug (1981) had used a limited form. Ori and the Blind Forest pushes the idea to hundreds of layers, by one reviewer's count.
Say it: "Use four parallax layers: sky fixed, mountains at 10% of camera speed, trees at 40%, foreground bushes at 130%. The play layer is always 100%."
https://gamecraft.dqnotes.com/looks/parallax

### Pixel art
Also: Sprite art, 8-bit style, 16-bit style.
Art made at a low resolution with every pixel placed on purpose, then shown enlarged with hard edges. Each pixel is a decision, which is what separates it from a shrunken image.
It can be quick to make per asset and hard to make well. It only looks right if every asset shares one pixel size, so mixed pixel sizes are the classic mistake.
Not low resolution: Shrink a big drawing and you get low resolution: blurry edges and hundreds of accidental colours. Pixel art is built at that size, with a chosen colour in every square.
Seen in: Small Mario in Super Mario Bros. is 16 by 16 pixels, and Shigeru Miyamoto has said he got his cap and moustache because hair and a mouth were too hard to show at that size.
Say it: "Pixel art at 320 by 180, scaled up by whole numbers only. One pixel size for everything, with no rotated or smoothly scaled sprites, and characters 24 pixels tall."
https://gamecraft.dqnotes.com/looks/pixel-art

### Palette swap
Also: Recolour, Colour swap, Palette shift.
The same image data shown with a different set of colours. The picture stores numbers that point into a palette, so changing the palette recolours everything at once and costs no extra art.
It turns one sprite into a family: a stronger enemy, a second player, a frozen version. On hardware with little memory it was a free character.
Not colour grading: Grading shifts the whole finished picture. A palette swap changes one thing's colours exactly, and can change any index into any colour at all.
Seen in: Luigi is Mario with a different palette in Super Mario Bros., and that game's clouds and bushes are the same shape in different colours. Mortal Kombat built Scorpion and Sub-Zero from one recoloured ninja.
Say it: "Do the elite enemy as a palette swap of the grunt: swap the body red to purple and the eyes to white. No new frames, and the player learns it is the same enemy but worse."
https://gamecraft.dqnotes.com/looks/palette-swap

### Dithering
Also: Ordered dither, Bayer dither, Error diffusion, Floyd-Steinberg.
Faking colours you do not have by mixing dots of the ones you do. From a distance, a checkerboard of black and white looks grey, so a smooth gradient survives a tiny palette.
Without it, a gradient breaks into hard stripes called banding. It also suits a style, though it can flicker badly once things move.
Not banding: Banding is the problem: visible steps where a gradient ran out of colours. Dithering is the fix, trading those steps for fine grain.
Seen in: Return of the Obra Dinn renders in pure black and white. Lucas Pope began with a Bayer pattern, moved to his own blue-noise variant, and worked to stop the pattern flickering as the camera moves.
Say it: "Quantise the fog to 8 levels and use a 4 by 4 Bayer dither, fixed to the screen. If it crawls during camera moves, try blue noise."
https://gamecraft.dqnotes.com/looks/dithering

### Sprite
Also: Sprite sheet, Sprite strip, Flipbook.
A 2D image drawn into the game at some position, usually a character or object. A sprite sheet is one image holding all its frames; playing the animation means showing a different rectangle of it each moment.
A single sheet loads as one file, stays together, and is fast to draw. Keep each frame the same size and anchored to the same spot, or the character jitters as it plays.
Not texture atlas: A sprite sheet usually holds one thing's animation, in a tidy grid. An atlas packs many unrelated images together, wherever they fit, to save draw calls.
Seen in: Metal Slug is famous for the sheer number of hand-drawn sprite frames behind each soldier and tank, down to the way they idle.
Say it: "Give me a sprite sheet of 8 run frames at 32 by 32, in one row, with feet on the same baseline in every frame, played at 12 frames a second."
https://gamecraft.dqnotes.com/looks/sprite

### Tileset
Also: Tilemap, Autotile, Tile bitmask.
A set of small square images that fit together edge to edge, used to build big levels from a small amount of art. Autotiling picks the right edge or corner piece for you by checking each tile's neighbours.
One tileset can make hundreds of levels, in very little memory. Without autotiling, a level designer places every corner by hand, and every map edit means fixing the edges around it.
Not sprite: Tiles stay on a grid and make up the level. Sprites are free to move anywhere on top of it.
Seen in: RPG Maker's editor calls them autotiles: you paint with one grass or water tile, and it picks the shores and corners for you.
Say it: "Use 16 pixel tiles with a four-neighbour autotile. That is 16 pieces of ground, chosen by checking up, right, down and left, and three variants of the plain middle so large areas do not look stamped."
https://gamecraft.dqnotes.com/looks/tileset

### Texture atlas
Also: Atlas, Sprite atlas, Texture packing.
One big image with many smaller images packed into it. Switching images is slow for a graphics card, so if everything lives in a single image the game can draw it all in one go.
Every switch can force a new draw call, which is a separate instruction to the graphics card. Atlases turn hundreds of draw calls into a few, which is often the difference between a smooth game and a stuttering one.
Not sprite sheet: A sprite sheet keeps one thing's frames in a neat grid for animation. An atlas packs anything together, in any arrangement, for speed.
Seen in: Stardew Valley keeps its items together on one sheet, springobjects.png, and item mods have long patched that single file.
Say it: "Pack all the UI icons into one 2048 pixel atlas with 2 pixels of padding, so the inventory screen is a single draw call and not 60."
https://gamecraft.dqnotes.com/looks/texture-atlas

### Low poly
Also: Polygon count, Flat shaded, Poly count.
Models built from few triangles, with the facets left showing. Once a hardware limit, it is now often a chosen style, because each flat face catches light as a single colour.
It is fast to make and to draw, and it still looks intentional, which suits small teams. The catch is that every triangle shows, so poor shapes cannot hide under textures.
Not low resolution: Polygon count is how many faces a model has. Resolution is how many pixels you draw it with. A game can be low poly and razor sharp.
Seen in: Star Fox (1993) drew flat-shaded polygons on a Super Nintendo with the help of a Super FX chip inside the cartridge, and its faceted ships still set the look many low-poly games now choose on purpose.
Say it: "Low poly, flat shaded, no textures: each rock is under 40 triangles and each face a single colour. Keep the budget under 500 triangles per character."
https://gamecraft.dqnotes.com/looks/low-poly

### PBR
Also: Physically based rendering, Metalness and roughness, Physically based shading.
Shading materials by rules of real light, usually described with two main numbers: how rough a surface is and whether it is metal. Because the material is described and not painted, it looks right under any light.
Before it, artists faked highlights to suit one lighting setup, and the fake fell apart in the next room. With PBR, one material works everywhere, and artists across a team get consistent results.
Not realism: PBR describes how light behaves, not what the art looks like. Cartoon worlds use it too. It is a rule set, not a style.
Seen in: Brent Burley's Disney paper of 2012 set out the artist-friendly model for Wreck-It Ralph, and games followed: Ready at Dawn built The Order: 1886 on a physically based pipeline, presented at SIGGRAPH in 2013.
Say it: "Use PBR with metalness and roughness. Rusty iron is metal at 0.9 and roughness at 0.7, polished stone is metal at 0 and roughness at 0.2, and no one paints highlights into textures."
https://gamecraft.dqnotes.com/looks/pbr

### Normal map
Also: Bump map, Surface detail map.
A texture that stores which way the surface faces at every pixel, as a colour. Lighting reads it, so a flat wall gets the shading of bricks and cracks while the geometry stays flat.
It gives you the look of millions of triangles for the price of a handful. It gives itself away only at the edges, where the outline is still straight.
Not height map: A height map stores how high a surface is and can actually move geometry. A normal map stores only the direction it faces, so it changes the lighting and nothing else.
Seen in: Doom 3 (2004) built its look on normal maps baked from much denser models, so low-polygon characters and walls catch light as though they are full of detail.
Say it: "Bake a normal map from the 2 million triangle sculpt onto the 3,000 triangle model. Keep the silhouette correct in the real mesh, and let the normal map do the wrinkles."
https://gamecraft.dqnotes.com/looks/normal-map

### Cel shading
Also: Toon shading, Toon shader.
Shading in a few flat steps of colour instead of a smooth gradient, often with an ink outline, so a 3D model looks like a hand-painted animation cel.
It gives a game a strong, lasting look. It is also unforgiving: every small lump in a surface becomes a hard blotch of shadow, so the surface directions have to be tended by hand.
Not outline: Outlines are a separate pass around a shape. Cel shading is the stepped lighting inside it. Each can appear without the other.
Seen in: Jet Set Radio (2000) made the look famous. Guilty Gear Xrd took it to a point where its 3D fighters pass for 2D; Junya Motomura's GDC 2015 talk explains that the slightest difference in a surface normal can end up as a huge blotch.
Say it: "Cel shade the characters with three light steps and a hard edge between them, plus a dark outline two pixels thick. Hand-edit the face normals so the nose shadow stays clean."
https://gamecraft.dqnotes.com/looks/cel-shading

### Rim light
Also: Rim lighting, Back light, Edge light, Fresnel highlight.
A thin line of light along the edge of a shape, as if something bright stands behind it. In games it is often calculated: the more a surface turns away from you, the brighter it gets.
It pulls a dark character away from a dark background without changing the mood of the scene. It is one of the cheapest ways to protect a silhouette.
Not outline: An outline is a drawn line of constant colour. A rim light behaves like light: it reacts to the shape and the angle.
Seen in: Team Fortress 2 adds rim highlights to its characters. Valve's own commentary says they emphasise the outline, and that without them a character blends into the scenery.
Say it: "Add a cool rim light to the player, fresnel power 3 at 60% strength. Keep it on in dark rooms, so the silhouette never melts into the background."
https://gamecraft.dqnotes.com/looks/rim-light

### Ambient occlusion
Also: AO, SSAO, Contact shadow.
Soft darkening where surfaces meet: in creases, corners and under objects, where less of the surrounding light can reach. It is a shadow with no particular light causing it.
It stops objects looking as though they float above the floor, and it makes shapes solid even in flat lighting. Overdone, it looks like dirty halos around everything.
Not shadow: A shadow comes from one light and has a direction. Ambient occlusion comes from all around, and shows how exposed a spot is.
Seen in: Crysis (2007) is credited as the first game to use screen-space ambient occlusion, a technique developed by Vladimir Kajalin at Crytek. It estimates the effect each frame from the depth of nearby pixels.
Say it: "Add ambient occlusion, baked into the level for corners and SSAO at half resolution for moving characters. Keep the radius to 30 cm so there are no halos."
https://gamecraft.dqnotes.com/looks/ambient-occlusion

### Bloom
Also: Glow, Light bloom, Real-time glow.
Bright parts of the picture bleed soft light into the pixels around them, the way light scatters in your eye or in a camera lens. It works by copying the brightest pixels, blurring them, and adding them back.
Screens cannot show how bright the sun is, and bloom makes a thing feel brighter than the pixel really is. Too much and everything looks as if the lens has been smeared with butter.
Not lens flare: Bloom is a soft glow around a bright thing. A lens flare is streaks and circles caused by light bouncing inside a camera, and it can sit far from the light.
Seen in: The developers of Tron 2.0 published their Real-Time Glow method in 2004, and that article spread the technique across the industry. Heavy bloom soon became a mark of its era.
Say it: "Add bloom only above a brightness of 1.0, so only lights and magic glow, not white walls. Spread it over five blur levels at low intensity, and keep it under 15% on the HUD."
https://gamecraft.dqnotes.com/looks/bloom

### Colour grading
Also: LUT, Lookup table, Post-processing colour.
Shifting all the colours of the finished picture to set a mood, as a last step. A LUT, or lookup table, stores the shift as a small cube of colours: each input colour simply looks up its output.
One small image can change the whole mood of a level, cheaply, at any time, and can be crossfaded with another. It can also flatten your careful value work if pushed too far.
Not tone mapping: Tone mapping squeezes bright light into the range a screen can show. Grading then chooses the look. Tone mapping is usually a technical step and grading a creative one.
Seen in: Fallout 3's wasteland has a green cast that players still argue about, and mods such as Fellout exist to remove it. Fallout: New Vegas went warm, with browns and oranges, so the two games feel apart at a glance.
Say it: "Grade the flashback with a LUT: lift the blacks, pull saturation down 30%, push shadows towards teal. Crossfade it in over two seconds rather than cutting."
https://gamecraft.dqnotes.com/looks/colour-grading

### Mipmap
Also: Mip levels, Mips, Trilinear filtering.
A stack of smaller, already blurred copies of each texture, and the graphics card picks the size that fits how far away the surface is. "Mip" comes from the Latin multum in parvo, much in a small space. Lance Williams described it in 1983.
Without them, a distant surface picks a few random pixels from a huge image, and the result shimmers as the camera moves. They also make rendering faster. They cost about a third more memory.
Not level of detail: Mipmaps swap a texture for a smaller version. Level of detail, or LOD, swaps a whole model for a simpler one. The idea is the same, but the things being swapped are different.
Seen in: The original PlayStation had no mipmapping in hardware, part of why its distant textures crawl as you move. The Nintendo 64 did have it.
Say it: "Generate mipmaps for all world textures and use trilinear filtering, so the distant floor stops shimmering. Leave them off for pixel-perfect UI."
https://gamecraft.dqnotes.com/looks/mipmap

### Skybox
Also: Sky dome, Cube map, 3D skybox.
A giant picture of the distant world wrapped around the camera, usually on the inside of a cube, and drawn behind everything else. It moves with you, so you can never get closer to it, which is how it seems infinitely far away.
It gives you mountains, a city or a whole galaxy for the cost of six images. If the corners or the horizon show, or the ground does not meet it, the illusion breaks.
Not parallax: Parallax layers move at different speeds to suggest depth. A skybox does not shift at all as you walk, only when you turn, which is exactly what very distant things do.
Seen in: Half-Life 2's Source engine adds a 3D skybox: a miniature model of the distant city, drawn around the real level so its skyline shifts slightly as you move.
Say it: "Use a skybox for the mountains and sky, a 2048 pixel cube map, fixed to the camera position. Match its horizon colour to the fog so the ground has no visible seam."
https://gamecraft.dqnotes.com/looks/skybox

### Billboarding
Also: Billboard, Impostor, Camera-facing sprite.
A flat picture in a 3D world that keeps turning to face the camera, so you never catch it edge-on. Trees, grass, particles and old-style enemies are often done this way.
One flat card costs two triangles, against thousands for a modelled tree, so a forest becomes affordable. The illusion breaks if you look straight down, or if you can see it swivel.
Not sprite: A sprite is the flat image. Billboarding is the rule that makes a flat image turn towards the camera in a 3D world.
Seen in: Doom's monsters are flat pictures that always face you, drawn from eight angles so they appear to turn. Super Mario 64's trees are flat cards that turn to follow the camera.
Say it: "Billboard the distant trees past 50 metres: one camera-facing card with two triangles, turning only around the vertical axis so they do not tip when you look up."
https://gamecraft.dqnotes.com/looks/billboarding

### The twelve principles of animation
Also: Twelve basic principles, Disney's principles.
Twelve rules for making drawings seem alive, worked out by Disney's animators and set down by Frank Thomas and Ollie Johnston in their 1981 book The Illusion of Life. They are still how animators talk about motion.
They name what is missing when a thing moves correctly and still looks dead. Nearly all "it feels stiff" notes are one of these twelve.
Not game feel: The principles describe how motion looks. Game feel also covers how it answers your hands. A long anticipation can be a great animation and a terrible control.
Seen in: Cuphead, drawn by hand to match 1930s cartoons, shows them openly: nearly everything on screen squashes, stretches, anticipates and overshoots.
Say it: "The boss looks stiff. Apply the principles: half a second of anticipation before the slam, slow in and slow out on the arm, an arc on the swing, and follow-through on the cape."
https://gamecraft.dqnotes.com/looks/twelve-principles

### Timing and spacing
Also: Easing, Slow in, slow out, Ease in and out.
Timing is how many frames a move takes. Spacing is how far it travels between each of those frames. Same timing with different spacing gives very different motion.
Weight and mood live here. Even spacing looks mechanical, and clustered frames at each end look natural. Most "it looks robotic" notes are a spacing problem.
Not frame rate: Frame rate is how many frames the whole game shows each second. Timing is how many of them this one movement takes.
Seen in: Richard Williams's The Animator's Survival Kit (2001) teaches this with charts of ball positions on a line, and animators still sketch those ticks when they plan a move.
Say it: "Keep the door at 12 frames, but change the spacing: ease out hard, so 70% of the move happens in the first 3 frames, then let it settle into the frame."
https://gamecraft.dqnotes.com/looks/timing-and-spacing

### Pose to pose
Also: Keyframing, Key poses, Inbetweening, Tweening.
Drawing the most important poses first, then filling in the frames between them. Its opposite is straight ahead, drawing each frame after the last from start to finish.
You can check that the move reads and lands on time before doing the expensive work, and game timing depends on landing on the right frame. Straight ahead is livelier and harder to control.
Not straight ahead: Straight ahead starts at frame one and finds the move as it goes, which suits fire and water. Pose to pose plans the destination first, which suits anything that must hit a mark.
Say it: "Block the attack pose to pose first: wind-up, contact, recovery. Send me just those three held poses at the right timing before anyone adds inbetweens."
https://gamecraft.dqnotes.com/looks/pose-to-pose

### Animating on twos
Also: On ones, On twos, Limited animation, Held frames.
Showing each drawing for two frames of a 24-frame second, so there are 12 new drawings a second. "On ones" means a new drawing every frame, and "on threes" every third. The holds are a choice, not a slowdown.
It halves the drawing work, and it can look better: snappier and more graphic, where smooth inbetweens might turn to mush. It also separates the character's rhythm from the game's frame rate.
Not low frame rate: A low frame rate makes the whole game choppy and slower to respond. Animating on twos leaves the game running at full speed and holds only the drawings.
Seen in: Studio MDHR has said Cuphead was drawn on ones, 24 drawings a second, for extra fluidity. Guilty Gear Xrd goes the other way on purpose, using limited animation on 3D models so that they feel hand-drawn.
Say it: "Animate the character on twos at 24, so 12 new poses a second, with attacks on ones for the three frames around contact. The game still runs at 60."
https://gamecraft.dqnotes.com/looks/animating-on-twos

### Smear frames
Also: Smear, Motion smear, Multiples.
One or two distorted drawings, shown so briefly you never quite see them, that stretch an object along its path during a very fast move. It is the hand-drawn version of motion blur.
A very fast move in crisp frames looks like a teleport, because the eye sees no link between the poses. A smear bridges them in a single frame, and you feel the speed.
Not motion blur: Motion blur is calculated by the renderer and softens everything. A smear is drawn, on purpose, and can bend, stretch and be as stylish as any other pose.
Seen in: Chuck Jones's The Dover Boys at Pimento University (1942) is the classic early example. In games, Hollow Knight's nail swings use them, as sword attacks often do.
Say it: "The sword swing looks like it teleports. Add one smear frame between wind-up and contact: a crescent along the blade's path, held for a single frame at 24."
https://gamecraft.dqnotes.com/looks/smear-frames

### Skeletal animation
Also: Frame-by-frame animation, Bone animation, Cutout animation, Traditional animation.
Two ways to make a character move. Frame by frame means a new drawing for each pose. Skeletal means one drawing (or model) attached to bones, with the computer moving the bones.
Frame by frame gives you total freedom, at the cost of redrawing everything for each new outfit or move. Skeletal gives you smooth motion at any speed, blending and reuse, but it can look like a paper puppet.
Not rigging: Skeletal animation is the way of moving a character. Rigging is the job of building the bones and controls it needs.
Seen in: Every frame of Cuphead was drawn by hand and inked on paper. Rayman Origins took the other route, with Ubisoft's UbiArt Framework animating the hand-drawn art by puppeting it.
Say it: "Use skeletal animation for all enemies, because we need 40 of them and blending between moves. Do the player frame by frame, at 12 drawings a second, since that is what people will look at most."
https://gamecraft.dqnotes.com/looks/skeletal-animation

### Rigging
Also: Rig, Inverse kinematics, IK, Forward kinematics, FK.
Building the skeleton and the controls that move a character. With forward kinematics you rotate each joint in turn, from the shoulder down. With inverse kinematics you place the hand, and the solver works out the joints.
A good rig is what makes animating quick. IK also works at runtime, so feet land on real slopes and hands reach real door handles, things no pre-made animation can know about.
Not skinning: Rigging builds the bones and controls. Skinning sets how the surface bends with those bones. The word rigging often covers both.
Seen in: Spore had to animate creatures that players had built themselves, so a 2008 SIGGRAPH paper describes retargeting its motions to any body at runtime, solving limbs by inverse kinematics. Rain World animates its creatures in code too, rather than from drawn frames.
Say it: "Rig the legs with IK at runtime. Cast down from each hip, place the foot on what you hit, and lower the pelvis so the longer leg can still reach on stairs."
https://gamecraft.dqnotes.com/looks/rigging

### Blend tree
Also: Blend space, Animation blending, Locomotion blending.
A small graph that mixes animations by a number from the game. Feed in speed and it returns, say, 30% walk and 70% run, averaged joint by joint and kept in step, so there is no separate clip for every speed.
Without blending, a character snaps between clips and pops at every change of speed. It also means a few clips can cover thousands of in-between moves.
Not state machine: A state machine picks which thing the character is doing and handles the switch. A blend tree mixes clips continuously within one of those states.
Seen in: Unity's Animator has a component literally called a Blend Tree, and Unreal Engine's equivalent is called a Blend Space.
Say it: "Use a 1D blend tree on speed: idle at 0, walk at 1.5 m/s, run at 5 m/s, with foot phases synced. Then add a lean as a second axis, driven by turn rate."
https://gamecraft.dqnotes.com/looks/blend-tree

### Root motion
Also: In-place animation, Animation-driven movement, Foot sliding.
Letting the animation decide how far the character moves. The animator moves the whole body in the clip, and the game follows it. Its opposite is in-place animation, where code moves the character and the clip marches on the spot.
Root motion plants the feet and gives moves real weight, but the controls then have to wait for the animation. In-place responds at once, but the feet slide if code and clip disagree on speed.
Not foot sliding: Foot sliding is the symptom: a planted foot skating across the floor. It appears whenever the speed in the code does not match the speed in the clip.
Seen in: Both big engines treat it as a per-animation choice: Unreal's documentation says root bone movement does not affect the character until you enable Root Motion, and Unity has an Apply Root Motion setting.
Say it: "Use root motion for heavy attacks and climbing, so the feet stay planted. Use in-place for the run, driven by code, and match the clip's stride to 5 m/s so that the feet do not slide."
https://gamecraft.dqnotes.com/looks/root-motion

## 10. Sound

Music, effects and silence: the half of the game people feel without noticing, and the words for asking for it.

### Waveforms
Also: Sine, square, triangle, saw, Oscillator shapes, Timbre.
The basic repeating shapes a synthesiser starts from. Each has its own stack of harmonics, the quieter pitches above the main one, and that stack is what makes it sound the way it does.
It gives you words for tone. "Thin and pure" is a sine, "hollow and woody" is a square or a triangle, "buzzy and bright" is a saw, and a sound designer can act on any of those.
Not pitch: Pitch is how high a note is. The waveform is what kind of sound plays at that pitch: a flute and a violin can play the same note.
Seen in: The NES plays music from pulse waves, a triangle and noise, which is much of why its games sound the way they do. Its bass is usually on the triangle, the softest of its voices.
Say it: "Use a sine for the coin so it sounds clean, and a saw through a filter for the laser. Keep square waves for the retro menu sounds only."
https://gamecraft.dqnotes.com/sound/waveforms

### ADSR envelope
Also: Envelope, Attack, decay, sustain, release, Volume envelope.
Four numbers that shape a note's loudness over time. Attack is how fast it rises, decay how fast it falls to the sustain level, and release how long it takes to fade once the note ends.
The same oscillator can be a pluck, an organ or a slow pad, depending only on these numbers. A short attack also feels instant, while a slow one on a button sound feels like lag.
Not reverb: An envelope shapes the sound itself. Reverb adds the room's echoes after it, so a short note in a large hall still rings on.
Seen in: The Commodore 64's SID chip gives each of its three voices its own attack, decay, sustain and release, in hardware.
Say it: "Give the UI click a 2 ms attack and an 80 ms decay with no sustain, so it feels instant. The menu pad can take 700 ms to fade in."
https://gamecraft.dqnotes.com/sound/adsr-envelope

### Low-pass filter
Also: LPF, High-cut, Muffling.
A filter that lets low frequencies through and cuts high ones above a chosen point, the cutoff. Close it and a sound goes dull, as if heard through a wall.
It is the cheapest way to say "this is far away, behind something, or underwater." Moving it over time also changes a sound's energy without touching the volume.
Not volume: Turning down makes everything quieter equally. A low-pass removes the brightness and leaves the weight, which is why a muffled explosion still shakes.
Seen in: Hitman's sound propagation, described at GDC Europe in 2015, treats rooms and portals as a graph and applies a low-pass filter to sounds heard through them.
Say it: "When the player pauses, slide a low-pass down to 800 Hz over 200 ms, so the world goes muffled and the menu sounds sit on top of it."
https://gamecraft.dqnotes.com/sound/low-pass-filter

### Chiptune
Also: 8-bit music, Chip music.
Music written for, or in the style of, the sound chips of old consoles and computers. Those chips have only a few voices, each playing one note at a time, so every limit is heard.
It is an instant signal of an era, and the limits make it very cheap to run. Fast arpeggios that fake chords are its signature, born because no channel could hold a chord.
Not retro-sounding music: True chiptune stays within a chip's real channels. Plenty of retro-styled scores use modern synths and reverb, which the old hardware could not have played.
Seen in: Shovel Knight's score, by Jake Kaufman with Manami Matsumae, was written for the NES chip plus the VRC6 expansion, eight channels in all. The VRC6 only ever shipped on the Japanese Famicom, so the sound is authentic to hardware most players never owned.
Say it: "Write the score for NES limits: two pulses, a triangle and noise, one note per channel. Fake the chords with arpeggios and put the bass on the triangle."
https://gamecraft.dqnotes.com/sound/chiptune

### Ambience
Also: Ambient sound, Atmos, Room tone, Soundscape.
The constant quiet sound of a place: a low bed that never stops, plus small sounds scattered at random around you. Nobody notices it until it stops.
Without it, a world feels like an empty set, however good it looks. Ambience sets mood and place before anything happens, and it costs far less than music.
Not music: Ambience belongs to the place and has no tune or beat. Some of the best game scores blur the line on purpose.
Seen in: Limbo's sound, by Martin Stig Andersen, works as ambient, processed sound of the world more than as tunes. For Inside he went further, playing synth sounds through a human skull with transducers and contact microphones and recording what came out.
Say it: "Give the cave a bed of 170 Hz rumble that never loops audibly, and a random drip every one to three seconds, panned somewhere different each time."
https://gamecraft.dqnotes.com/sound/ambience

### Foley
Also: Footsteps and cloth, Movement sounds.
Sounds of bodies and the things they touch, usually recorded by a person acting them out: footsteps, clothes, a hand on a door. The name comes from Jack Foley, who did this for early film.
Foley makes characters feel present and physical. Footsteps also carry information: who is near, how fast they move, what they are walking on.
Not sound effect: Foley is the small human movement, recorded to match it. Effects are the bigger designed events, such as explosions and magic.
Seen in: For NieR:Automata, lead sound designer Misaki Shindo's team recorded the footsteps of 2B and 9S with real shoes, first for cutscenes and then in the game as well.
Say it: "Record six footsteps per surface for gravel, wood, metal and snow, and add a quiet cloth layer on each step so the character has a body."
https://gamecraft.dqnotes.com/sound/foley

### Diegetic and non-diegetic sound
Also: In-world sound, Source music, Underscore.
Diegetic sound exists inside the game's world, so its characters could hear it: a radio, a footstep. Non-diegetic sound lays over the world for you alone: the score, a menu click.
Moving a sound between the two is a powerful trick. Music that turns out to come from a real radio makes a moment feel suddenly real, and the player can walk away from it.
Not ambience: Ambience is one kind of diegetic sound. Diegetic covers anything in the world, including a song playing on a jukebox.
Seen in: Silent Hill's pocket radio plays static when monsters are near, without saying where: a warning that exists in the character's hands. Grand Theft Auto's music mostly comes from the car radio, so you can change the station or get out and leave it.
Say it: "Play this song from the radio on the desk, not over the scene: band-pass it to 350 Hz to 2.8 kHz, attenuate it with distance, and let the player switch it off."
https://gamecraft.dqnotes.com/sound/diegetic-sound

### Attenuation
Also: Falloff, Rolloff, Distance curve.
How a sound gets quieter as you move away from it. A curve maps distance to volume, usually with a range inside which it is at full volume and a range beyond which it stops.
Attenuation is how players locate things by ear. The curve is a design choice, not physics: an important enemy can be audible from further than reality would allow.
Not occlusion: Attenuation depends only on distance. Occlusion is what happens when something solid stands in between.
Seen in: Overwatch departs from realism on purpose: enemy footsteps are mixed louder than allies' footsteps, because they matter more. Its designers described this at GDC in 2016.
Say it: "Enemy footsteps stay at full volume to 3 m, fall off on an inverse curve, and stop at 25 m. Distant footsteps should still be locatable."
https://gamecraft.dqnotes.com/sound/attenuation

### Occlusion
Also: Muffling through walls, Obstruction, Sound propagation.
Muffling a sound because something solid stands between it and the listener, usually with a low-pass filter and a drop in volume. Games test this with lines or room graphs.
In any game with hearing, it tells players where something is relative to the walls. Without it, enemies in the next room sound as if they are beside you.
Not obstruction: Trade usage varies, but occlusion usually means the sound and the listener are in separate spaces, with the sound passing through the wall. Obstruction means something blocks the direct path while the sound still reaches you around it.
Seen in: Rainbow Six Siege passes sound through propagation nodes and destructible connection points, so breaking a wall changes what you hear. Its audio director Louis Philippe Dion separated occlusion from obstruction, because occlusion alone would have let players hear through walls too well.
Say it: "If there is no clear line from the sound to the listener, close a low-pass to 500 Hz over 100 ms and drop it 9 dB. Check each door when it opens."
https://gamecraft.dqnotes.com/sound/occlusion

### Reverb
Also: Reverberation, Room sound, Convolution reverb.
The thousands of echoes that follow a sound as it bounces around a space, blurred into a tail. Its length and colour tell your ears the size and material of the room.
It places a sound in a space, even with your eyes shut. Dry sound in a cathedral looks wrong without anyone being able to say why. Too much turns everything to soup.
Not echo: An echo is a distinct repeat you can pick out. Reverb is so many reflections so close together that you hear one smooth fade.
Seen in: Forza Horizon 5 was reported to use ray tracing for audio, which Playground Games compared to echolocation. Microsoft's Project Acoustics takes another route, simulating sound waves through a level ahead of time and handing designers the results as reverb settings.
Say it: "The hall needs about 1.8 s of reverb, the closet about 0.3 s. Send footsteps and voices to it at 25%, and keep gunfire drier so it stays punchy."
https://gamecraft.dqnotes.com/sound/reverb

### Pitch randomisation
Also: Round robin, Variation, Random container.
Playing a repeated sound slightly higher or lower each time, or cycling through several recordings of it, which is called round robin. The two are usually used together.
The ear spots an identical sound repeated at once, and a rifle turns into a stutter: the machine-gun effect. Small variations hide it, and cost almost nothing.
Not musical feedback: Randomising hides the pattern in a repeated sound. Musical feedback makes a pattern on purpose, so that the repeats climb and mean something.
Say it: "Every footstep picks one of six samples, never the same one twice running, and shifts pitch by up to 1.5 semitones either way. Do the same for gunfire."
https://gamecraft.dqnotes.com/sound/pitch-randomisation

### Voice limiting
Also: Polyphony limit, Voice stealing, Virtual voices, Max instances.
A cap on how many sounds can play at once. When a new sound needs a voice past the limit, the least important one is stopped: it is stolen.
Without a cap, a hundred coins at once sum into distortion and waste processing on sounds nobody can pick out. With priorities, the important ones always get through.
Not mix: A mix sets how loud each sound is. Voice limiting decides which sounds get to exist at all.
Seen in: On the NES, a sound effect takes over one of the music's channels while it plays, so the tune loses a part for a moment. Battlefield: Bad Company culls sounds by loudness, so anything drowned out by a louder one is not processed at all.
Say it: "Limit coin pickups to 8 voices, and steal the oldest first. Dialogue and the player's own shots have top priority and are never stolen."
https://gamecraft.dqnotes.com/sound/voice-limiting

### Mix
Also: Mixing, Balance, Snapshot, Bus.
The balance of every sound against the others, so that what matters can be heard. In a game it must change as play changes, so groups of sounds get saved sets of levels called snapshots.
Every sound can be great and the game can still be a wall of noise. A mix is a series of decisions about what the player needs to hear now, and what has to give way.
Not mastering: A mix sets each sound against the others. Mastering is the final pass on the whole result, for overall level and tone.
Seen in: Overwatch's "importance system" changes the mix by threat, using whether an enemy is looking at you, shooting at you and how far away it is. A distant enemy aiming at you sits above a nearby one who is not. Scott Lawlor and Tomas Neumann explained it at GDC in 2016.
Say it: "Add three mix snapshots. Exploring keeps music up. Combat drops pads by 16 dB and pushes drums and impacts. Dialogue pulls everything down at least 12 dB."
https://gamecraft.dqnotes.com/sound/mix

### Dynamic range and loudness
Also: Loudness, LUFS, Compression, Night mode.
Dynamic range is the distance from the quietest sound to the loudest, in decibels. Loudness is how loud the whole game seems over time, measured in a unit called LUFS.
Wide range feels cinematic in a quiet room, and on a phone it means whispers you cannot hear and explosions that hurt. Games offer settings for this because the room is unknown.
Not volume: Volume is one number at one moment. Dynamic range is the spread between the extremes, and it stays the same when you turn your speakers up.
Seen in: PlayStation's target, per Wwise's documentation, is -24 LUFS on average, give or take 2. An earlier Sony working group set -23 for PS3 and -18 for Vita. Battlefield: Bad Company took another path with "HDR audio", which DICE described as measuring loudness at the listener and scaling every sound, not compression.
Say it: "Master the game to -24 LUFS on console. Add a night mode that squeezes the range from 39 dB to about 12 dB, so whispers carry and explosions do not wake anyone."
https://gamecraft.dqnotes.com/sound/dynamic-range

### Ducking
Also: Sidechain, Sidechain compression, Auto-ducking.
Turning one sound down automatically while another plays, most often music under speech. A sidechain is the wire that does it: one sound's level controls another's volume.
It keeps dialogue and big moments clear without anyone hand-mixing each one. If it comes back too quickly the music pumps, which is why the release is slower than the attack.
Not mix snapshot: A snapshot is a set of levels you choose for a situation. Ducking follows a live signal and reacts to whatever is actually playing.
Seen in: Wwise has both auto-ducking and sidechaining, the latter built with a meter effect on one bus driving a game parameter on another. DICE contrasted its HDR audio with traditional ducking, which shows how standard ducking had become.
Say it: "Sidechain the music to the dialogue bus: 9 dB down with a 50 ms attack and a 400 ms release, so it never pumps between words."
https://gamecraft.dqnotes.com/sound/ducking

### Silence as a tool
Also: Audio rest, Negative space, Dropping out.
Removing sound on purpose: cutting the music, or even the ambience, before or during a moment. Silence is something you can play, as long as you have made noise beforehand.
Ears adapt to loudness, so after a quiet moment the same sound lands harder. It also frightens: players know a score going quiet means something is about to change.
Not missing audio: A deliberate rest has a shape: it comes straight after sound, for a reason. An empty scene with nothing at all just feels unfinished.
Say it: "Cut all the music 1.8 s before the boss's first slam, and keep only a low wind. Then let the slam hit at the same level as always."
https://gamecraft.dqnotes.com/sound/silence

### Earcon
Also: Audio icon, UI sound, Notification sound.
A short abstract sound with a learned meaning, like a menu blip that means yes. It is a word in the game's language, and nothing in the real world sounds like it.
It tells you something happened while you are looking elsewhere. A good one takes one or two hearings to learn, and then you recognise it for the rest of the game.
Not auditory icon: An auditory icon sounds like its meaning: a paper crumple for the bin. An earcon has no natural link, so its meaning is taught through use.
Seen in: The Legend of Zelda plays a short rising jingle when you find a secret, and it means the same thing across decades of the series.
Say it: "Confirm is two notes rising a fifth, back is the same two falling, and error is one low note repeated. All under 300 ms, all from the same synth."
https://gamecraft.dqnotes.com/sound/earcon

### Stinger
Also: Musical sting, Hit point, Musical punctuation.
A short burst of music played over the score to mark a moment: a discovery, an alarm, a victory. It is written to fit the score, and can wait for the next beat.
It lets the score react to an event instantly without changing what it is playing. On the beat it sounds composed. Off the beat it responds faster, but it can clash.
Not earcon: An earcon is an interface signal and sounds the same anywhere. A stinger is a piece of the music, and matches the score's key and tempo.
Seen in: Metal Gear Solid, in 1998, pairs a "!" over a guard's head with a short, sharp alert sound when you are spotted, and the sound has barely changed through the series since.
Say it: "When the player is spotted, play a brass stinger in A minor on the next beat, at most 300 ms late. The loop carries on underneath it."
https://gamecraft.dqnotes.com/sound/stinger

### Leitmotif
Also: Theme, Motif, Character theme.
A short tune tied to one character, place or idea, and brought back changed as the story goes. The same notes can be heroic in one key and broken in another.
It tells a story without words. A tune you learned in a happy scene, played slowly in a minor key, hurts more than any line of dialogue could.
Not main theme: A main theme is one piece for the whole game. A leitmotif is a building block that turns up inside many different pieces.
Seen in: Austin Wintory built Journey's score around a single theme, with a solo cello representing the player. It became the first game score nominated for a Grammy, in December 2012.
Say it: "Write a seven-note motif for the sister. Play it on a music box when we meet her, on solo cello when she is lost, and on full brass in the last fight."
https://gamecraft.dqnotes.com/sound/leitmotif

### Vertical layering
Also: Vertical remixing, Stems, Layered music.
Splitting one piece of music into layers that all play at once in sync, then fading them in and out with the game. The music can change at any moment without a seam.
It responds instantly and cannot go out of time. The catch is that all the layers must work together, so the melody and harmony cannot really change: only the intensity can.
Not horizontal resequencing: Vertical changes which layers play at the same moment. Horizontal changes which section comes next.
Seen in: Red Dead Redemption's score, by Bill Elm and Woody Jackson, was recorded as stems all in A minor at a fixed tempo so they could be overlaid. Rockstar says it was limited to five stem tracks at a time.
Say it: "Write the exploration music as four stems in the same key and tempo. Keep the pad always on, bring in the bass near danger, and add drums when a fight starts."
https://gamecraft.dqnotes.com/sound/vertical-layering

### Horizontal resequencing
Also: Branching music, Music states, Transition on the bar.
Cutting music into sections and letting the game choose which plays next. Requests wait for a musically sensible point, usually the next bar line, so changes sound written.
It allows real changes of melody and key, which layers cannot do. The cost is delay: the music may need a second or two to answer, so it cannot mark a split-second event.
Not crossfade: A crossfade blends two tracks whenever it is told to, whether or not they fit. Resequencing waits for a moment when they do.
Seen in: Monkey Island 2, the first game to use LucasArts' iMUSE system by Michael Land and Peter McConnell. In the town of Woodtick each building plays a variation of the theme on a different instrument, and leaving ends on a closing flourish back into the main theme. Earlier games could only fade or cut.
Say it: "When combat starts, change to the fight section on the next bar line, at most 2.4 s later. Play a stinger on the next beat to cover the wait."
https://gamecraft.dqnotes.com/sound/horizontal-resequencing

### Adaptive music
Also: Interactive music, Dynamic music, Reactive score.
A score that responds to what the player is doing, by bringing layers in and out, moving between sections, or both. The composer writes rules as well as notes.
A film composer knows what happens at each second, and a game composer does not. Adaptive music lets the score stay matched to a scene that plays differently every time.
Not dynamic mix: A mix changes how loud sounds are. Adaptive music changes what the music itself plays, and when.
Seen in: LucasArts' iMUSE, first used in Monkey Island 2, has been likened to a pit orchestra: it plays shorter or longer sections while it waits for key events. Red Dead Redemption instead layers stems in one key and tempo, so they can be combined during play.
Say it: "Drive the music from one danger value. Above 20%, fade in bass, above 45% drums, above 65% lead. Above 75%, move to the alert section at the next bar line."
https://gamecraft.dqnotes.com/sound/adaptive-music

### Musical feedback
Also: Rising pitch, Combo pitch, Scale on hits.
Playing the sounds of repeated actions as notes, often climbing a scale as a combo grows. The action becomes a tune, and the player can hear how well it is going.
It carries a count without a number on screen, and each rising note makes you want the next one. Using a real scale keeps it pleasant however long the chain runs.
Not pitch randomisation: Random pitch hides repetition, and means nothing. Rising pitch is a signal: the higher it is, the better you are doing.
Seen in: Peggle raises the pitch of its peg sound with each peg a shot hits, so you can hear a good shot building. In Peggle 2 and Peggle Blast, Guy Whitmore tied those notes to the harmony of the music playing underneath.
Say it: "Each pickup in a chain plays the next note of A minor pentatonic, starting from A4. Reset after one second without a pickup, and top out after 12 notes."
https://gamecraft.dqnotes.com/sound/musical-feedback

## 11. Interface

How a game speaks to the player and how the player answers: HUDs, menus, saves, teaching and accessibility.

### HUD
Also: Heads-up display, Non-diegetic UI, Overlay.
The information laid over the picture of the game: health, ammo, score, map. The name comes from fighter planes, whose pilots read figures on the glass without looking down. Elements that exist only for the player, outside the world, are also called non-diegetic.
A HUD is the clearest way to tell a player anything, and each piece of it takes eyes off the world. Too little and players feel blind; too much and they play the corners of the screen.
Not ui: UI is the whole interface, menus and all. The HUD is the part on screen while you play.
Seen in: Super Mario Bros. keeps score, coins, world and time in a single row across the top. Skyrim hides its health, magicka and stamina bars when they are full and brings them back when they are needed, while its compass stays. Journey has no HUD at all.
Say it: "Cut the HUD to health and ammo. Fade both out after three seconds at full, and bring them back the moment either changes."
https://gamecraft.dqnotes.com/interface/hud

### Diegetic UI
Also: In-world UI, Diegetic interface.
Interface that exists inside the game's world, so the character could see it too: a meter on a suit, a map held in the hand. Fagerholt and Lorentzon's 2009 thesis Beyond the HUD names four kinds of game UI, sorted by two questions: is it in the story, and is it in the world's space?
It keeps your eyes on the world and makes the interface part of the fiction. The cost is clarity: when it is dark, or the action is fast, the information is harder to read and easier to miss.
Not spatial ui: Both sit in the game's 3D space. Diegetic UI is something the character knows about. Spatial UI is there only for you.
Seen in: Dead Space puts Isaac's health in a strip down the spine of his suit and his stasis energy beside it, and ammo counts on the weapon. In Metro 2033 the watch on your wrist shows how long your gas mask filter has left. Far Cry 2 has you look down at a map held in your hand.
Say it: "Move the health bar onto the character's back as a strip of lights, and add a limp under 30%. Keep a plain bar as an option for people who can't read that fast."
https://gamecraft.dqnotes.com/interface/diegetic-ui

### Spatial UI
Also: World-space UI, Spatial interface.
Interface placed in the game's 3D space that the character can't see: an outline round an enemy, a glowing trail on the ground. It exists in the space but not in the story.
It puts information where you're already looking, so you don't have to glance at a corner. Used everywhere, it covers the world you built with scribbles.
Not diegetic ui: A trail on the ground that nobody in the story can see is spatial. If the characters could point to it, it would be diegetic.
Seen in: Splinter Cell: Conviction projects its objectives as words onto the walls of the level. Fable III lays a glowing trail on the ground to your objective.
Say it: "Remove the objective text from the corner. Project it on the wall of the room it refers to, in the same font, and let the camera find it."
https://gamecraft.dqnotes.com/interface/spatial-ui

### Meta UI
Also: Meta interface, Screen-space effects.
Interface that stands for something in the story but is drawn on the screen rather than in the world: blood on the camera when you're hurt, a phone that fills the screen. It is in the story but not in the space.
It tells you something urgent without a single number, and it is hard to miss. It says little, though, and it covers the picture just as you need to see, so it needs a cap and an off switch.
Not hud: A health bar means nothing in the story, so it is non-diegetic. Blood on the screen stands for a wound, so it is meta.
Seen in: Call of Duty: Modern Warfare 2 shows damage as blood spreading over the screen, with no health bar. Grand Theft Auto IV has you take out a phone in the world, and then uses it through a flat screen overlay.
Say it: "Show low health with a red vignette that starts at 40% and grows to the corners by 10%, and never let it cover more than a third of the picture."
https://gamecraft.dqnotes.com/interface/meta-ui

### Information hiding
Also: Hidden information, Withheld stats.
Deliberately leaving out a number the game knows, so the player has to read it from the world or work it out. In other words, deciding what the HUD should keep in the fog.
Everything you show becomes what people look at. Show the boss's health and they watch a bar; hide it and they watch the boss. Hide something the player can't work out any other way, though, and it just feels unfair.
Not fog of war: Fog of war hides parts of the map until you scout them. Information hiding is about stats and states, like the health of the thing right in front of you.
Seen in: Monster Hunter shows no health bar for its monsters. You learn to read them: they limp, they tire and drool, their parts break. The trap in this toy works the same way as the game's, which only captures a monster that is badly hurt.
Say it: "Don't show the boss's health. Give it three visible stages instead: cracks at 60%, a slower walk at 40%, a limp under 25%."
https://gamecraft.dqnotes.com/interface/information-hiding

### Text readability
Also: Legibility, Contrast, HUD readability.
How quickly and reliably someone can read what the interface is telling them while busy doing something else. For text it comes down to size, weight and contrast against whatever happens to be behind it.
Your interface will be read over every frame of your game, not over the flat grey it was designed on. Text that looks fine in a mockup disappears on snow, and a player mid-fight will not squint.
Not ui scale: Text readability is the whole question of how easy the words are to read. Scale is one answer: how big they are.
Seen in: The Last of Us Part II has a High contrast display option, a separate way of drawing the game that mutes the scenery and gives allies, enemies and items strong, distinct colours.
Say it: "All HUD text gets a two pixel dark outline, and anything longer than three words sits on a plate at 75% black. Check it against the brightest level we have."
https://gamecraft.dqnotes.com/interface/readability

### UI scale
Also: HUD scale, Text size, Ten-foot UI.
How big the interface is drawn, and usually a setting that lets players change it. What matters is how large it looks from where they sit, which depends on the screen's size and their distance from it.
A game built on a monitor at arm's length will be played on a TV across a room, and on a handheld with a much smaller screen. Text that can't be read is a game that can't be played.
Not resolution: Resolution is how many pixels there are. UI scale is how many of them each letter uses. A sharper picture with the same small letters is still hard to read from a sofa.
Seen in: Dead Rising set its text so small that people on standard-definition TVs struggled to read its missions at all. The Last of Us Part II offers a HUD scale setting, Default or Large.
Say it: "Set body text to 32 px at 1080p and offer a 150% UI scale in settings. Test it on a 55 inch TV from three metres before we call it done."
https://gamecraft.dqnotes.com/interface/ui-scale

### Safe area
Also: Title safe, Action safe, Overscan.
The part of the picture you can trust a screen to show. Many TVs blow the picture up slightly and crop its edges, which is called overscan, so anything important stays inside a margin. The broadcast standard keeps titles within the middle 90%.
Without it, a HUD can be cut off on someone's TV and you will never see it on your own monitor. Phones have the same problem with notches and rounded corners.
Not ui scale: UI scale changes how big things are. The safe area changes where they are allowed to sit.
Seen in: PlayStation consoles have a system setting, Display Area, where you drag the picture's corners until they fit your TV, because the console can't tell by itself how much it crops.
Say it: "Keep every HUD element inside the middle 90% of the screen, and add a safe-area slider to the first-run settings so people can pull it in further."
https://gamecraft.dqnotes.com/interface/safe-area

### Health bar
Also: Life bar, Ghost bar, Delayed health bar, Damage trail.
A shape that shows how much damage you can still take. A common refinement is the ghost bar: when you are hit, the chunk you lost stays visible in a paler colour for a moment, then drains away.
A bar that simply jumps shorter tells you where you are, but not what just happened. The ghost shows the size of the hit, and it holds through a burst of small hits so you can see the whole of it.
Not recoverable health: A ghost bar is only a display. Tekken 8's grey section looks similar but is real: it is health you can win back by attacking.
Seen in: Elden Ring shows the chunk you just lost in a paler colour before it drains away. Hollow Knight skips the bar altogether and gives you masks, so you always know exactly how many more hits you can take.
Say it: "Add a ghost bar: after a hit, hold the lost chunk in white for half a second, restart that wait if another hit lands, then drain it over a quarter of a second."
https://gamecraft.dqnotes.com/interface/health-bar

### Damage numbers
Also: Floating combat text, Hit numbers, Pop-off numbers.
Numbers that jump off an enemy each time you hit it. They turn a hidden sum into feedback: how hard, and whether it was a critical hit.
In games about upgrades, they are the proof that the upgrade worked. In a fast game, though, a stream of equal-sized numbers is just noise, so size and colour need to do the reading for you, and small rapid hits should add into one.
Not hit stop: Hit stop makes a blow feel heavy. A damage number tells you how heavy it was. One is a sensation, the other a measurement.
Seen in: Borderlands fills the screen with floating numbers that burst from enemies as you shoot them. They are as much a part of its look as its art.
Say it: "Scale damage numbers with the size of the hit, between 18 and 60 px. Crits go red with a short pop. Hits within 300 ms of each other add into one number."
https://gamecraft.dqnotes.com/interface/damage-numbers

### Minimap
Also: Radar, Mini-map.
A small map in a corner of the screen, showing what's nearby and where you are. It answers "where am I and where do I go?" without opening anything.
It makes large spaces easy to cross, and it is expensive. People end up steering by it, and can drive through your beautiful city while looking at a small square in the corner.
Not waypoint: A minimap shows you the surroundings, and you do the working out. A waypoint gives you the answer.
Seen in: Namco's Rally-X (1980) put a radar beside the maze showing the flags and enemy cars, and it is often credited as an early minimap. The Grand Theft Auto series made a corner map standard for open worlds.
Say it: "Use a 200 px circular minimap, player up, with a 150 m radius. Show roads and objectives, not enemies, so people still look at the street."
https://gamecraft.dqnotes.com/interface/minimap

### Waypoint
Also: Objective marker, Quest marker, Map pin.
A marker that shows exactly where to go next, usually with a distance. It can sit on a compass, on the map or in the world.
It removes any chance of getting lost, and with it the need to look around, ask or remember. A strong waypoint turns exploring into following an arrow, so decide whether you want that.
Not signifier: A signifier hints that something can be done. A waypoint points to the place you've been told to go.
Seen in: Dead Space's locator draws a line of light along the floor to your objective at the press of a button, which keeps it inside the world. Skyrim marks your quest on its compass. Elden Ring only rarely gives you a marker, and players keep notes instead.
Say it: "Show a waypoint on the compass only, and put a distance on it within 50 m. The player should have to look up from the compass to find the actual door."
https://gamecraft.dqnotes.com/interface/waypoint

### Screen edge indicator
Also: Off-screen indicator, Damage direction indicator, Threat arrow.
A mark at the edge of the screen pointing to something important that is off it, such as who just shot you or where your teammate is.
The camera can only show so much, and players blame the game for being hit from somewhere they couldn't see. An arrow at the edge is a very cheap way to tell them.
Not waypoint: A waypoint is about where you're going. An edge indicator is about what you can't currently see, and it is often about what's coming for you.
Seen in: Call of Duty marks incoming fire with a red marker that shows the direction the shots came from, together with blood on the screen.
Say it: "When something off screen fires at the player, show a red arrow on that side for a second, and make it twice the size if the shot is about to land."
https://gamecraft.dqnotes.com/interface/screen-edge-indicator

### Quest log
Also: Journal, Mission log, Objectives screen.
A screen that keeps track of what you've been asked to do and what you already know. It is the game's memory, so players don't need one of their own.
People come back after two weeks with no idea what they were doing, and a good log lets them pick up again instead of quitting. Its danger is the opposite: a long list of chores that reads like a to-do list at work.
Not waypoint: The log says what you are doing and why. A waypoint says where to walk. Many games have both, and some have only the second.
Seen in: Elden Ring has no quest log at all, on purpose. Players keep paper journals or use companion apps instead, which is the clearest test there is of what a log is worth.
Say it: "Each entry in the quest log gets one sentence on what to do next, and one on who asked and why. Open on the most recent quest, and keep finished quests in their own tab."
https://gamecraft.dqnotes.com/interface/quest-log

### Inventory
Also: Backpack, Grid inventory, Carry weight, Item slots.
The screen and the rules for what you're carrying. The rule is usually one of three: a set number of slots, a weight limit, or a grid in which each item has a shape.
The rule decides what question the player asks each time they find something. Slots ask if it's worth a slot, weight asks if it's worth its weight, and a grid asks how it'll fit. Without a limit, there is nothing to decide.
Not equipment: Inventory is everything you carry. Equipment is what you are wearing or wielding right now.
Seen in: The first Resident Evil gives Jill eight slots and Chris six, and a key takes one just as a shotgun does. Resident Evil 4 turned this into a grid, its attache case, where you rotate and fit each item. Skyrim gives you a weight limit.
Say it: "Use a 6 by 4 grid. Weapons take long thin shapes and potions take one square, with a tidy button so that fitting things stays a pleasure, and never a chore."
https://gamecraft.dqnotes.com/interface/inventory

### Tooltip
Also: Hover text, Nested tooltip, Keyword tip.
A small box explaining the thing you are pointing at. A nested tooltip goes one further: words inside the box have tooltips of their own.
It lets a rich game explain itself at the moment of need, with no manual and no tutorial. It relies on hover, though, which doesn't exist on a touch screen or a pad, so those need another way in.
Not tutorial: A tutorial teaches you something when the game decides. A tooltip answers a question when you ask it.
Seen in: Crusader Kings III is built on nested tooltips: hold your pointer over a term in a tip and it opens another tip, as deep as you care to go, so a very complicated game can explain itself in place. Slay the Spire explains keywords like Vulnerable right beside the card.
Say it: "Any game term in a tooltip gets its own tooltip, and it stays open while you move into it. On pad, the same tips open with a held button."
https://gamecraft.dqnotes.com/interface/tooltip

### Signifier
Also: Visual language, Yellow paint, Environmental cue.
Any perceivable hint that an action is possible here: paint on a ledge, a glowing crack, a handle on a door. Don Norman brought the word into design to separate the thing you can do from the sign that tells you so.
Without them, players must try everything or give up. With them, you can make huge spaces easy to read without a word of text. Be careful: once players learn that only painted ledges work, an unpainted one is a wall.
Not affordance: An affordance is what you can do with something. A signifier is what tells you. A ledge you can climb has the affordance whether you notice it or not; the paint on it is the signifier.
Seen in: Mirror's Edge turns useful scenery red to show you what to run along and climb, which it calls Runner Vision. It can be turned down or off once you know the language.
Say it: "Every climbable ledge gets a strip of worn yellow paint, the same across the game. Nothing else in the level may be that yellow."
https://gamecraft.dqnotes.com/interface/signifier

### Fitts's law
Also: Fitts law, Target size rule.
The time it takes to reach a target grows with distance and shrinks with size, in a precise way. Paul Fitts measured it in 1954: time = a + b × log2(2D ÷ W), so doubling the distance or halving the width adds the same fixed amount.
It tells you where to spend screen space. Common actions should be big and close, and anything rare can be small and far away. Screen edges and corners are effectively huge targets for a mouse, as the pointer stops there.
Not friction: Fitts's law is about the time for one move to one target. Friction is the whole cost of a task, including how many of those moves it takes.
Seen in: Radial menus such as Mass Effect's power wheel are Fitts's law at work: all the choices sit next to the centre, equally close, and each one is a wide wedge that's hard to miss with a thumbstick.
Say it: "Make the buy button at least 80 px tall and put it where the thumb already rests. Leave anything rare in a small corner icon."
https://gamecraft.dqnotes.com/interface/fitts-law

### Menu flow
Also: Screen flow, UI flow, Wireframe flow.
The map of every screen in a game and how you get from one to the next: what Back does, where the cursor lands, what comes after a game over.
Players visit menus hundreds of times. A small flaw in the route, such as a cursor that forgets where you were, is repeated all those times, and flow is what makes a menu feel fast long before anything is animated.
Not friction: Menu flow is the route itself. Friction is what it costs to travel it, and you can also find friction outside menus, in a long animation or a slow door.
Say it: "After a game over, put the cursor on Retry with one press to restart. Back always returns to the row you came from, and no screen is more than three presses from play."
https://gamecraft.dqnotes.com/interface/menu-flow

### Friction
Also: UX friction, Interface cost, Clicks to do it.
Everything that stands between wanting to do something and having done it: presses, waits, confirmations, screens. Some is deliberate, and most of it is an accident nobody noticed.
People don't decide a game is annoying all at once; it builds a press at a time. Friction on something done once is fine, but friction on something done a hundred times is what they will remember the game for.
Not difficulty: Difficulty is the challenge you came for. Friction is what you pay to reach it. Making a boss hard is design. Making the walk back to it long is friction, unless you've chosen it on purpose.
Seen in: Ocarina of Time's Water Temple is remembered for its iron boots: on the N64, each change meant pausing, going to the equipment screen, choosing them and coming back, over and over. The 3DS remake made them an item you tap on and off, and its director, Eiji Aonuma, said the boots had made the dungeon seem harder than it was.
Say it: "Changing boots takes four presses and we do it thirty times in this dungeon. Put them on a quick slot so it takes one press and doesn't pause the game."
https://gamecraft.dqnotes.com/interface/friction

### Pause
Also: Pause menu, Real-time menu, Active pause.
Stopping the game. It seems too obvious to need a word, except that many choices are hiding here: whether menus stop the world, what happens when the phone rings, and whether the game pauses itself when you lose focus.
Real life interrupts people, and a game that can't be paused can't be played by someone with a child or a doorbell. Keeping menus running is a powerful way to add tension, but it should be a decision, not an oversight.
Not hit stop: Hit stop is a pause of a few hundredths of a second that the game makes for effect. Pause is you asking the game to wait.
Seen in: Dead Space's menus are projected from Isaac's suit and the game carries on behind them, so a monster can reach you as you look for a medkit. Dark Souls doesn't stop for its menus either, and players have complained about having no pause for as long as the series has existed.
Say it: "Pause when the window loses focus or a controller disconnects. Inventory runs in real time, on purpose, but the options menu always stops the game."
https://gamecraft.dqnotes.com/interface/pause

### Loading screen
Also: Load time, Loading tips, Playable loading screen.
What the player sees while the game reads what it needs from the disc or drive. It can be a blank screen, a progress bar, a tip or even a bit of play.
A wait with nothing happening seems longer than the same wait with something to look at, and a black screen can't be told apart from a crash. The time itself matters most, since people quit over it, but showing progress is cheap and it always helps.
Not friction: Loading is a wait the machine imposes. Friction is any cost, including ones you could remove with a redesign tomorrow.
Seen in: Ridge Racer on the first PlayStation let you play Galaxian while the main game loaded. Namco patented the idea of games during loading screens, and that patent only expired in 2015. Bayonetta keeps you playing with its own moves during loads.
Say it: "Show a real progress bar that never goes backwards, and let the player keep practising combos on a dummy while it loads. Keep tips to one line."
https://gamecraft.dqnotes.com/interface/loading-screen

### Onboarding
Also: First-time user experience, FTUE, New player experience, First session.
Everything that takes a stranger to being someone who knows how to play and wants to keep going. The start of that, the first few minutes, is often called the first-time user experience, or FTUE.
New players are at their least patient and know the least, and you lose more of them here than anywhere. Each screen between launch and play was put there by someone with a good reason, and together they drive people away.
Not tutorial: A tutorial is one way to teach. Onboarding is the whole road in, including the logos, the name entry, the first reward and the first time you want to come back.
Seen in: Hearthstone opens with scripted matches against characters who chat and react, with most choices already made for you. You are playing, and winning, inside the first minutes, and only then does it open up.
Say it: "Have the player jumping within ten seconds of launch. Move the account, the difficulty and the story behind the first level, and ask for each one only when it matters."
https://gamecraft.dqnotes.com/interface/onboarding

### Tutorial
Also: Training level, Prompted teaching.
A part of the game set aside to explain how to play, usually with instructions. It can be a dedicated level, or short prompts that appear as they're needed.
People don't remember rules they read before they had any use for them. A prompt at the moment of need teaches far more with far fewer words, and an unskippable one punishes the people who already know how to play.
Not teaching by doing: A tutorial tells you, then has you try. Teaching by doing just sets things up so that you find it out for yourself.
Seen in: Breath of the Wild opens on the Great Plateau, a region you can't leave without a paraglider, earned by finishing four shrines that each hand you a key ability. It is a tutorial that feels like the open world.
Say it: "Cut the opening text screen. Show one short prompt when the player reaches each obstacle, never more than five words, and don't show it again once they've done it."
https://gamecraft.dqnotes.com/interface/tutorial

### Teaching by doing
Also: Show, don't tell, Silent teaching, Level as tutorial.
Building the level so the player works out a rule for themselves: first a safe place to try it, then a small test, then a real one. No text is needed.
What you work out yourself sticks, and it feels like your own cleverness. The first use has to be safe, though. If the first thing a new ability is needed for can kill you, you are testing people before you've taught them.
Not tutorial: A tutorial explains the rule. Teaching by doing arranges the world so that the rule explains itself.
Seen in: World 1-1 of Super Mario Bros. teaches running, jumping, blocks, mushrooms and enemies without a line of text, and almost everything later in the game is a variation on it. Portal's first test chambers introduce the portal gun through a sequence of simple rooms, each building on the last.
Say it: "Introduce the double jump in a room with no danger and one high ledge. Next room, the same ledge over a shallow pit. Only the third has a gap that kills."
https://gamecraft.dqnotes.com/interface/teaching-by-doing

### Button prompt
Also: Controller glyphs, Input icons, Press A to.
The on-screen icon that tells you which button to press. Glyphs are the icons themselves, and they have to match the device in your hands, because pads label the same positions differently.
Xbox's A is at the bottom of the diamond, and Nintendo's A is on the right. A prompt drawn for the wrong pad is worse than no prompt at all, since it confidently sends you to the wrong button.
Not input remapping: Remapping changes which button does what. Prompts have to follow it: if someone has moved jump to Y, the prompt must say Y.
Seen in: Steam Input can hand a game the right icons for the pad in use, yet many PC games still ship only Xbox prompts, which leaves everyone else translating in their heads.
Say it: "Show glyphs for whichever device was used last, and switch the moment that changes, even mid-prompt. Read them from the player's current bindings, not from a fixed list."
https://gamecraft.dqnotes.com/interface/button-prompt

### Input remapping
Also: Rebinding, Custom controls, Key bindings.
Letting players choose which button or key does what. Full remapping covers every action, including the menus.
Hands differ, keyboards differ, and some people physically can't reach some buttons. For them this setting decides whether they can play at all. It is also far cheaper if you plan for it from the start.
Not control scheme: A scheme is a ready-made layout to pick from. Remapping lets you make your own. Offering three schemes isn't the same as letting people choose.
Seen in: The Last of Us Part II lets you remap every command to any button, save the result under your own name, and turn held buttons into toggles and rapid tapping into a single hold.
Say it: "Every action gets rebindable, including menu confirm. When a key is already taken, swap the two instead of leaving something unbound, and provide a reset."
https://gamecraft.dqnotes.com/interface/input-remapping

### Subtitles
Also: Captions, Closed captions, Speaker labels.
Dialogue as text on screen. Captions go further, with sounds and music as well. Done well, they come with a choice of size, a background, the name of whoever is speaking, and where that person is.
Many people turn these on, far more than those who can't hear, including anyone playing at night or in a second language. Tiny white text with no background fails all of them, and it's the default that decides what most players get.
Not captions: Subtitles assume you can hear and give the words. Captions assume you can't and also describe sounds, such as footsteps behind you or a door opening.
Seen in: The Last of Us Part II offers three subtitle sizes, three background settings, a choice of colour, speaker names with their own colours, and an arrow that points to a speaker you can't see.
Say it: "Turn subtitles on by default, at 40 px at 1080p, on a 70% black plate, with speaker names in their own colours and an arrow for anyone off screen."
https://gamecraft.dqnotes.com/interface/subtitles

### Colour-blind modes
Also: Colour vision settings, Never colour alone, Redundant coding, Daltonism options.
Settings that change a game's colours for people who see them differently, about 1 man in 12 and 1 woman in 200. The stronger rule goes with it: never let colour be the only thing that carries a meaning, and back it with shape, icon or position.
Red against green is the classic choice for enemy and friend, and it is the pair most likely to fail. A mode helps those who find it, whereas a second signal helps everybody from the first second.
Not high contrast mode: A colour-blind mode shifts colours so that they stay distinct. High contrast separates what matters from what doesn't, for people with low vision as well.
Seen in: Overwatch offers filters for protanopia, deuteranopia and tritanopia, with a strength slider. Hue, a puzzle game all about colour, has a mode that adds a symbol to each colour. Forza Motorsport's racing line is a warning: players report its red and green are hard to tell apart, the whole message of the line.
Say it: "Enemies get spiked outlines as well as red. Run every screen through a deuteranopia filter before review, and if two things can't be told apart in greyscale, give them different shapes."
https://gamecraft.dqnotes.com/interface/colour-blind-modes

### Accessibility options
Also: Assist mode, Assist options, Difficulty options.
Settings that let more people play: for vision, hearing, movement and how fast you can think or react. The good ones change one specific barrier and leave the rest of the game as it is.
One wall stops a person who could happily play everything else, and it is usually cheap to remove. The extra options cost you nothing with those who don't use them, and you gain every player who couldn't have played otherwise.
Not easy mode: Easy mode changes the game for you as a single bundle. Accessibility options let you take down the specific wall that's in your way, and keep the rest of the challenge.
Seen in: Celeste's Assist Mode lets you slow the game to as low as 50%, give yourself infinite stamina and extra dashes, or turn on invincibility, and the screen recommends playing without it first, then offers it anyway. The Last of Us Part II has more than 60 accessibility settings.
Say it: "Add a game speed setting from 50 to 100% in steps of 10, separate from difficulty, and don't lock any achievements behind turning it off."
https://gamecraft.dqnotes.com/interface/accessibility-options

### Motion settings
Also: Reduce motion, Motion sickness options, Camera shake setting, Photosensitivity options.
Controls over how much the picture moves by itself: camera shake, head bob, motion blur, flashes, field of view. They are for people who get ill or disoriented from on-screen movement, and they should default to the system's reduced-motion setting where one exists.
Motion sickness is physical, and no amount of willpower helps. A player who feels ill after ten minutes won't give you a second chance, and they often won't know which of your effects was to blame.
Not screen shake: Screen shake is one effect. Motion settings are the controls over all such effects, and every shake you add should have one.
Seen in: The Last of Us Part II has a Motion sickness section: camera shake and motion blur on sliders from 1 to 10, camera distance, and full-screen effects you can switch off.
Say it: "Put camera shake, head bob and motion blur on sliders in the first-run settings, reading from the system's reduced-motion preference. Keep all flashes under three a second."
https://gamecraft.dqnotes.com/interface/motion-settings

### One-handed play
Also: One-handed mode, Single-hand control scheme, Thumb-only.
A way of playing with one hand, whether for someone who only has the use of one, or for a person holding a baby or a train strap. It usually means merging actions onto fewer buttons and letting the game do some of them for you.
Asking for two things at the same time is the first thing that shuts someone out. A small rethink of what really needs holding opens the game to them, and on a phone it can turn the game into something people play on the way to work.
Not input remapping: Remapping moves buttons around, but it can't help if two must be held at once. A one-handed mode changes what the game asks for.
Seen in: Super Mario Run was built to be held upright in one hand: Mario runs by himself and you only tap to jump. The Last of Us Part II has control schemes for Right Hand Only and Left Hand Only.
Say it: "Add a one-handed mode: firing is automatic, movement takes one thumb, and no action needs two inputs at the same time."
https://gamecraft.dqnotes.com/interface/one-handed-play

### Save system
Also: Autosave, Save points, Saving, Save slots.
The rules for how your progress is kept: whether by save points, by saving anywhere, or by autosave, where the game writes your progress for you every so often. Each choice sets what a mistake or a crash can cost.
It decides how much a player has to lose, and so how brave they're allowed to be. It also has to work: a crash that costs an hour of play is what people remember, and a lost save can end a game for them.
Not checkpoint: A save keeps your progress between sessions. A checkpoint is only where you come back after dying, and it may never be written to disk.
Seen in: The first Resident Evil makes you spend an ink ribbon at a typewriter for each save, so saving is itself a decision. Dark Souls autosaves constantly in one slot, so you can't reload your way out of a mistake. Breath of the Wild keeps a number of autosaves alongside a manual slot.
Say it: "Autosave on every room change and every 60 seconds at most, keep the last three in rotation, and show a small icon so that people know not to switch off."
https://gamecraft.dqnotes.com/interface/save-system

### Checkpoint
Also: Respawn point, Bonfire, Bench, Mid-level restart.
A place in the level that you return to when you die, so you don't have to start over. Its spacing sets the price of every death.
Too far apart and players spend their time replaying what they have already beaten, which tires them without testing them. Close together, a game can be very hard indeed and still feel fair, because each try costs so little.
Not save system: A checkpoint sets where you come back after dying. The save system sets what survives when you turn the machine off.
Seen in: Celeste restarts you at the start of the current room almost instantly, so hundreds of deaths never feel like a punishment. Hollow Knight sends you back to the last bench you rested at, and your money stays where you died, guarded by your shade until you fetch it.
Say it: "Put a checkpoint before each hard jump, about 20 seconds of play apart, and make the respawn take under one second."
https://gamecraft.dqnotes.com/interface/checkpoint

### Quicksave
Also: Save scumming, Quickload, Save anywhere, Reload abuse.
One button that saves right now, and another that takes you back. Save scumming is what people do with it: reloading again and again until the dice fall their way, or until they've got through without a scratch.
It respects your time and it also dissolves risk. A 30% shot is really only a 30% shot if you have to live with the miss. Many people will optimise the fun out of a game if allowed, so choose knowingly.
Not permadeath: Quicksaves make every choice reversible. Permadeath makes none of them reversible. Most games lie somewhere between.
Seen in: XCOM: Enemy Unknown stores its random number seed in the save, so reloading and taking the same shot gets the same result, and its Ironman mode allows no going back at all. The Enemy Within expansion added an optional setting that re-rolls on load, and its name is Save Scum.
Say it: "Store the random seed in the save, so a reloaded shot lands the same way unless the player does something different. Add an Ironman mode with a single save."
https://gamecraft.dqnotes.com/interface/quicksave

### Attract mode
Also: Demo mode, Advertise mode, Self-playing demo.
What a game does when nobody is playing: shows itself. Arcade cabinets looped their title, scores and a demo that played itself, to pull people in from across the room. This site's home page is doing the same thing right now.
A silent machine in a loud room earns nothing. The demo shows what the game is in seconds, and it does so to people who weren't looking for it. Any screen that waits for a stranger needs the same idea.
Not trailer: A trailer is an edited film about the game. An attract mode is the game itself, running where you can step in and take over.
Seen in: Pac-Man's attract mode introduces its four ghosts on screen, by character and by nickname, then shows a demo chase.
Say it: "After 20 seconds on the title screen, play a recorded run with sound off and a blinking press start. Any input drops the player straight into a new game."
https://gamecraft.dqnotes.com/interface/attract-mode

## 12. Genres

The names for kinds of game, what each one promises the player, and a tiny playable version of every one.

### Genre
Also: Game type, Category.
A family of games that share verbs and a core loop. Film genres are named for mood or setting, such as horror or western. Game genres are usually named for what you do, or after the game that first did it.
The name sets what a player expects before they start. Knowing the promise lets you keep it, or break it on purpose, instead of by accident.
Not setting: A setting is where the game takes place, such as space or the Wild West. A genre is what you do there. A space game may be a shooter, a 4X or a farming game.
Seen in: Early first-person shooters were simply called Doom clones, after the 1993 game that set their pattern. Roguelike, metroidvania and soulslike are still named this way, after the games they copy.
Say it: "Our genre is deckbuilder, so players will expect a run, a growing deck and a choice of card after each fight. If we drop the run, we need to say so on the store page."
https://gamecraft.dqnotes.com/genres/genre

### Platformer
Also: Platform game, Jump and run.
You move a character through space by running and jumping across gaps and hazards. It promises mastery of your own body: every death is a jump you can see how to do better.
The loop is run, jump, land, fail, retry, so everything rests on how the jump feels. A platformer with a bad jump has no other game to fall back on.
Not metroidvania: A platformer usually moves you forward through separate levels. A metroidvania is one connected world that you go back through as you gain abilities, and the jumping is only part of it.
Seen in: Donkey Kong (1981) made the jump the main verb, and its hero was called Jumpman. Super Mario Bros. (1985) added a scrolling world and a run that carries into the jump, and Celeste (2018) is the modern study of how fair a hard jump can feel.
Say it: "This is a precision platformer, so tune the jump first. I want respawn under a second, 100 ms of coyote time, and every death visibly caused by something on screen."
https://gamecraft.dqnotes.com/genres/platformer

### Metroidvania
Also: Search action.
One large connected map, where new abilities open paths you already saw and could not take. It promises the pleasure of remembering a door. The name joins Metroid and Castlevania.
The world itself becomes the progress bar. The ledge you could not reach an hour ago is a promise, and keeping it feels better than any reward handed to you.
Not open world: An open world lets you go nearly anywhere from early on. A metroidvania is locked by what you can do, so its order is hidden in plain sight and opens up slowly.
Seen in: Metroid (1986) and Castlevania: Symphony of the Night (1997) gave the genre its name. In Hollow Knight (2017), the Mantis Claw lets you cling to walls, and suddenly shafts you have walked past for hours go somewhere.
Say it: "Make it a metroidvania: show the high ledge in the first room, give the double jump 20 minutes later, and draw the ledge on the map so the player remembers to come back."
https://gamecraft.dqnotes.com/genres/metroidvania

### Bullet hell
Also: Shmup, Shoot 'em up, Danmaku, Curtain fire.
A shooter whose screen fills with slow, patterned enemy fire that you thread through. It promises calm inside chaos. Your real hitbox is far smaller than your ship, which makes it possible.
The patterns look impossible and are not. The gap between how close you seem to come and how safe you are is what makes the player feel superhuman.
Not bullet heaven: In a bullet hell, the screen full of bullets is aimed at you. In a bullet heaven, the screen full of bullets is yours.
Seen in: Cave's DoDonPachi (1997) helped set the style. Touhou Project games show your tiny real hitbox as a dot when you slow down, and they count and score each graze, so a close pass earns something instead of only being survived.
Say it: "Make the hitbox a 4-pixel dot at the ship's centre, drawn only while focused, and score each bullet that passes within 24 pixels as a graze."
https://gamecraft.dqnotes.com/genres/bullet-hell

### Bullet heaven
Also: Survivors-like, Vampire Survivors-like, Reverse bullet hell, Horde survival.
You only move. Your weapons fire by themselves at an endless swarm, you collect the gems they drop, and each level offers a choice of upgrade. It promises to turn a weak character into a screen-filling storm in minutes.
One control means anybody can play, and all the decisions move to the level-up screen. The swing from desperate to all-powerful inside one run is the whole appeal.
Not twin-stick shooter: In a twin-stick shooter you aim, and your skill is in your hands. In a bullet heaven the aiming is done for you, and your skill is the build.
Seen in: Vampire Survivors, by Luca Galante under the name poncle, reached version 1.0 in 2022. A run lasts 30 minutes, at which point Death arrives to end it, and weapons combine into stronger forms if you picked the right pairs.
Say it: "Bullet heaven: movement only, weapons auto-fire, a level-up every 20 seconds at the start with a choice of three, and 500 enemies on screen by minute ten."
https://gamecraft.dqnotes.com/genres/bullet-heaven

### Fighting game
Also: Fighter, Versus fighter.
Two characters, one screen, health bars and rounds. It promises a conversation between two people, conducted in punches, where you learn to read your opponent as much as your own buttons.
Everything comes down to frames: how long a move takes to start, and how long you are open after it. That is how a fast reflex game can become a slow game of guessing and punishing.
Not beat 'em up: A beat 'em up has you walking through crowds of weak enemies. A fighting game is one even duel, usually against another person.
Seen in: Street Fighter II (1991) set the form with eight playable characters and a stable of special moves. Its combos began as a quirk of how attacks could be cancelled into others, and players kept them, so later games built them in.
Say it: "Light punch is 4 frames of startup and safe on block. Heavy is 12 frames and punishable by 8, so a heavy has to be a read, not a reflex."
https://gamecraft.dqnotes.com/genres/fighting-game

### Rhythm game
Also: Music game, Rhythm action.
Press in time with the music, and you are judged by how close you were to the beat. It promises the feeling of playing the music yourself, even if you cannot play anything.
A timing window of milliseconds is fair only if the game's clock really is the audio's clock. Tie the notes to frames rather than the sound, and players will fail on beats they hit.
Not musical feedback: Any game can make its sounds fit the music. A rhythm game is judged on your timing against the music, and the music is the level.
Seen in: PaRappa the Rapper (1996) and Beatmania (1997) set out the idea, and Guitar Hero (2005) made it a party with a plastic guitar. Guitar Hero and Rock Band both include a calibration screen, because televisions and speakers add delay that the game has to measure.
Say it: "Time everything from the audio clock, not the frame count. Perfect is within 45 ms, good within 90 ms, and add a calibration screen for players on slow televisions."
https://gamecraft.dqnotes.com/genres/rhythm-game

### Puzzle game
Also: Puzzler, Logic game.
A game whose challenge is working something out, not doing it quickly. It promises the moment when the answer arrives and you feel clever, and that feeling is yours, because you got there.
A puzzle can be solved once, so each must teach you something the last one did not. Good ones forgive your slips and let you try ideas freely, since the challenge is thinking.
Not match-3: A logic puzzle has an answer that you work out, and it stays solved. A match-3 is an endless reaction game that happens to be called a puzzle game by app stores.
Seen in: Sokoban (1982) has been pushing crates into dead ends for decades. Portal (2007) teaches one idea per room, and Baba Is You (2019), where the rules themselves are blocks you can push, gives you free and unlimited undo, because a misstep there is not what you are being tested on.
Say it: "Give each room one idea. Three rooms to teach it, then one that combines it with the last idea. Undo is free and unlimited, so nobody ever has to replay a solved step."
https://gamecraft.dqnotes.com/genres/puzzle-game

### Match-3
Also: Match three, Tile matching, Swap puzzle.
Swap two neighbouring pieces to line up three of a kind. They vanish, the pieces above fall, and new lines may form by themselves. It promises small and constant rewards, plus sudden lucky cascades.
Cascades are the secret: they are free rewards you did not plan for. The number of colours is a real design dial, since fewer colours mean more accidental matches.
Not tetris-like: In Tetris you place falling pieces against gravity and the clock. In a match-3 the pieces are already there, and you rearrange them.
Seen in: Eugene Alemzhin's Shariki (1994) came first, PopCap's Bejeweled (2001) made it huge, and King's Candy Crush Saga (2012) added a limit on lives and a map of hand-made levels, turning it into a free-to-play business.
Say it: "Use six colours on an 8 by 8 board. Matching four makes a line-clearing piece, matching five makes a colour bomb, and raise the pitch a step on every cascade."
https://gamecraft.dqnotes.com/genres/match-3

### Stealth game
Also: Stealth action.
You avoid being seen rather than fighting. Guards have vision and hearing that you learn and exploit. It promises the tension of being one mistake away, and the satisfaction of never being noticed at all.
Stealth only works when the player can read what guards know. If you cannot tell why you were spotted, it stops being a skill and becomes a coin toss.
Not hiding: Lots of games let you hide for a moment. A stealth game makes avoiding detection the main verb, with guards built around it.
Seen in: Metal Gear (1987) made avoiding guards the point, and Thief: The Dark Project (1998) added light and sound. Mark of the Ninja (2012), being in 2D, draws vision cones and noise rings right on the screen, so you always know what the guards know.
Say it: "Show each guard's vision cone, and fill a detection meter over one second rather than failing instantly. If you get spotted, the guard checks the last place they saw you, then goes back."
https://gamecraft.dqnotes.com/genres/stealth

### Roguelike
Also: Traditional roguelike.
A turn-based dungeon crawl on a grid, generated anew each run, where death is permanent. It promises that you can never just memorise your way through: only what you have learned comes with you.
Permadeath gives every decision weight, and random maps stop it becoming rote. Together they turn knowledge of the rules into the only thing you keep.
Not roguelite: A roguelike keeps nothing between runs but what is in your head. A roguelite lets you carry permanent upgrades, so even a bad run moves you forward.
Seen in: It is named after Rogue (1980). At the International Roguelike Development Conference in 2008, developers wrote the Berlin Interpretation, whose high-value factors are random environment generation, permadeath, turn-based play, grid-based movement, non-modal play, complexity, resource management, hack and slash, and exploration and discovery.
Say it: "Keep it a true roguelike: grid-based, enemies move only when you do, a new map every run, and one death wipes the save file. No unlocks between runs."
https://gamecraft.dqnotes.com/genres/roguelike

### Roguelite
Also: Rogue-lite.
Takes the runs and the permanent death of a roguelike, and adds something that stays: currency or unlocks that make the next run stronger. It promises that no attempt is wasted.
It removes the bitterness of losing an hour, so far more players stick around. The risk is that grinding starts to replace skill, and the player only wins by waiting.
Not roguelike: In a roguelike, you get better. In a roguelite, you and your character both get better, and it can be hard to tell which one beat the boss.
Seen in: In Rogue Legacy (2013), the gold you bank upgrades the family manor for your next heir. In Hades (2020), each death sends you back to the house, where Darkness buys permanent upgrades at the Mirror of Night and the story moves on, so even failing feels like progress.
Say it: "Make it a roguelite: runs reset, but bank 30% of the gold into a permanent shop. By run 10 a player should have about 50% more health than run 1."
https://gamecraft.dqnotes.com/genres/roguelite

### Soulslike
Also: Souls-like, Soulsborne.
Slow, heavy third-person combat, where stamina limits every swing and enemies have long tells you must learn. Death drops what you were carrying where you fell. It promises that hard things can be beaten by paying attention.
Difficulty comes from commitment, not from fast reactions: every swing locks you in. The dropped souls make walking back to the same boss feel tense and worth doing.
Not hard game: Plenty of hard games are not soulslikes. The genre is a specific set of rules: stamina, committed attacks, checkpoints that bring enemies back, and a run back to what you lost.
Seen in: FromSoftware's Demon's Souls (2009) and Dark Souls (2011) set the pattern. Resting at a bonfire in Dark Souls saves you and brings the enemies back, and if you die before reaching your dropped souls, they are gone for good.
Say it: "Soulslike rules: every attack and roll spends stamina, a boss wind-up lasts at least 600 ms, and dying leaves your currency at that spot. You get one chance to fetch it."
https://gamecraft.dqnotes.com/genres/soulslike

### Immersive sim
Also: Immersive simulation, Imsim.
A world built from systems that follow consistent rules, in which problems have many answers and none is written in by hand. It promises that if your idea should work, it will.
When a door can be picked, burned, blown or bypassed, players start to feel like authors. It is costly, because every new thing must work with all the others, and most players see only one path.
Not open world: An open world is about how big the space is. An immersive sim is often a few small dense rooms, and is about how many ways the rules let you cross them.
Seen in: Ultima Underworld (1992) and System Shock (1994) laid the ground. The first level of Deus Ex (2000), on Liberty Island, can be fought through or sneaked through by several routes. In Prey (2017), the foam of the GLOO Cannon hardens into stairs you can climb.
Say it: "Do not script the door. Give wood a burn value and locks a pick difficulty, and let the guard flee fire. If a tester gets through in a way we did not plan, that is a pass."
https://gamecraft.dqnotes.com/genres/immersive-sim

### JRPG
Also: Japanese role-playing game, Turn-based RPG, Console RPG.
A party-based role-playing game in the line of Dragon Quest: a set story and cast, turn-based menu battles, and levels that you raise by fighting. It promises a long journey with people you grow fond of.
Its loop is plain: fight, level up, reach the next town. The story pulls you along while the numbers promise that the boss who beat you today will not beat you tomorrow.
Not crpg: A JRPG usually tells you a set story about set characters. A CRPG usually lets you create your character, and then bends the story around your choices. The name describes a tradition, and some are not made in Japan.
Seen in: Dragon Quest (1986) and Final Fantasy (1987) set the form, and Final Fantasy VII (1997) took it worldwide. In Persona 5 (2016), hitting an enemy's weakness earns you another turn, so the menu becomes a hunt for weak spots.
Say it: "Standard JRPG battle: four commands, fire deals double to ice enemies, and hitting a weakness grants another turn. A random fight should take 30 seconds at most."
https://gamecraft.dqnotes.com/genres/jrpg

### CRPG
Also: Computer role-playing game, Western RPG, Isometric RPG.
A role-playing game in the tradition of pen and paper: you make a character from stats, and the rules, often with visible dice, decide how they do. It promises a story that answers who you chose to be.
Your stats change what you can say and try, not only how hard you hit. This makes a second playthrough a different game, and also means vast amounts of writing that most players never see.
Not jrpg: A CRPG is about the character you built and the choices you made. A JRPG is more about the set character you are travelling with.
Seen in: Baldur's Gate (1998) put Dungeons & Dragons rules on a computer. Baldur's Gate 3 (2023) rolls a large d20 on screen when you try something in conversation. Disco Elysium (2019) uses two six-sided dice, and lets a failed roll be as interesting as a success.
Say it: "Show the dice. A speech check is a d20 plus Charisma against a difficulty of 15, the player sees the percentage before choosing, and a failure opens a different scene, not a dead end."
https://gamecraft.dqnotes.com/genres/crpg

### MMO
Also: MMORPG, Massively multiplayer online game.
A persistent online world shared with thousands of players at once, which carries on when you log off. It promises a place, and a society, more than a story: other people are the content.
Other people give you the economy, the groups and the status that a solo game has to fake. The cost is servers and a live team for as long as it runs, and an empty MMO feels dead.
Not online game: A match of an online shooter ends and its world disappears. An MMO's world keeps going, and what you did in it remains.
Seen in: Ultima Online (1997) and EverQuest (1999) set up the form, and World of Warcraft (2004) brought it to millions. Its first big raid, Molten Core, took 40 players at once, which no one could do alone. EVE Online keeps all its players on a single shard.
Say it: "This MMO's first world boss should need about 20 players and spawn every two hours, so that word spreads, people gather and guilds have a reason to exist."
https://gamecraft.dqnotes.com/genres/mmo

### 4X
Also: Civ-like, Explore, expand, exploit, exterminate.
Empire-building strategy, usually turn-based, whose name means explore, expand, exploit and exterminate. It promises to take you from a single settler to a world power, and it keeps you up with "one more turn".
Growth compounds, and there is always something finishing next turn, so there is never a good place to stop. The catch is the late game: once you are clearly ahead, the rest can be a long chore.
Not rts: A 4X usually gives you all the time you want and a whole world to manage. An RTS is a short and urgent fight, where speed is part of the skill.
Seen in: Alan Emrich coined the term in 1993, previewing Master of Orion for Computer Gaming World. Sid Meier's Civilization (1991) is the best-known example: a settler, a fog-covered map and thousands of years.
Say it: "Make it a 4X: start with one city in fog of war, give each tile a yield so that expansion compounds, and always have something finish on the next turn."
https://gamecraft.dqnotes.com/genres/4x

### Real-time strategy
Also: RTS.
Gather resources, build a base, train an army and destroy the enemy, all at once in a continuous clock. It promises to make you a general, and tests how many things you can do at the same time.
Every worker you build delays a soldier, and every second you spend looking at one thing costs you another. Strategy and sheer speed are mixed together, and that thrills experts and scares off newcomers.
Not turn-based strategy: Both are about choices. Turn-based gives you time to make the right one. Real time asks how many right ones you can make per minute.
Seen in: Westwood's Dune II (1992) set the pattern, and Blizzard's StarCraft (1998) became a professional sport in South Korea. Its workers really do walk to minerals and carry them back, so where you build changes how fast you earn.
Say it: "RTS with one resource: workers really walk to the mine and back, a soldier costs as much as two workers, and the first enemy attack arrives at three minutes to punish greedy openings."
https://gamecraft.dqnotes.com/genres/rts

### MOBA
Also: Multiplayer online battle arena, ARTS, Dota-like.
Two teams, usually of five, each control a single hero, while waves of computer-run minions march down lanes between the bases. It promises a long, deep team match in which you level up from nothing each game.
Gold and levels start at zero every match, and small early leads snowball. This is why it is so absorbing, and why a bad teammate can feel like a personal disaster.
Not rts: It grew from RTS games, but you do not build a base or an army. You are one unit, and your army marches without you.
Seen in: It grew from Aeon of Strife, a StarCraft custom map, and Defense of the Ancients, a Warcraft III mod from 2003. League of Legends (2009) and Dota 2 (2013) followed. Gold goes mainly to whoever lands the killing blow on a minion, and Dota 2 also lets you deny your own.
Say it: "MOBA rules: three lanes, minion waves every 30 seconds, and the last hit earns gold. A match should run about 30 minutes, with towers that punish anyone who pushes too early."
https://gamecraft.dqnotes.com/genres/moba

### Tower defence
Also: TD.
Enemies walk from one side to the other, and you stop them by placing towers that fire on their own. You never touch the enemies. It promises the joy of watching a plan you made work.
It moves all the skill into the placement. Every tower is a decision about where to spend, and the waves are a test of it, which makes it easy to learn and very hard to put down.
Not rts: An RTS has you commanding an army and attacking. In tower defence you never command anyone: you only build, and it plays out.
Seen in: Desktop Tower Defense (2007), a Flash game by Paul Preece, had enemies path around whatever you built, so you made them walk long mazes. Plants vs. Zombies (2009) moved the idea onto fixed lanes, and took it to a far wider audience.
Say it: "Enemies path around towers, so allow mazing, but refuse any placement that seals the route. Start with two tower types: fast single-target, and a slow that hits an area."
https://gamecraft.dqnotes.com/genres/tower-defence

### City builder
Also: City sim, Management sim.
You lay out roads, homes and work, and a simulated population decides whether to move in. There is usually no win screen. It promises a place of your own that visibly comes alive.
The simulation gives you back your own decisions as traffic, taxes and empty lots. Every fix causes the next problem, so there is always one more thing to improve.
Not sandbox: A sandbox lets you make what you like. A city builder lets you make what you like, and then its systems tell you whether the people who live there are happy.
Seen in: SimCity (1989), by Will Wright, has no way to win, and it shows a small R, C, I meter for residential, commercial and industrial demand. Cities: Skylines (2015) added a detailed traffic simulation that became the main thing to fight.
Say it: "City builder with no win state: homes need a road and jobs, shops need residents, and a tax above 12% slows move-ins. Show demand for all three zones as bars."
https://gamecraft.dqnotes.com/genres/city-builder

### Deckbuilder
Also: Deck-building game, Roguelike deckbuilder.
You start with a small and weak deck of cards, and you improve it as you play, adding and removing cards between hands or fights. It promises the delight of watching a pile of cards become a machine.
Each new card makes the others better or worse, and every card added dilutes the draw. This makes a small choice interesting, and a bad deck is your own doing.
Not collectible card game: In a collectible card game you build the deck before you play, often buying the cards. In a deckbuilder, building the deck is the game.
Seen in: Dominion (2008), by Donald X. Vaccarino, invented the form on the tabletop. Slay the Spire (2019) put it into a roguelike run with three energy a turn, and showed each enemy's intent, so that every turn became a small, fair puzzle.
Say it: "Deckbuilder rules: three energy a turn, five cards drawn, and show what every enemy will do. After each fight offer one of three cards, or let the player skip so that a thin deck is a real plan."
https://gamecraft.dqnotes.com/genres/deckbuilder

### Auto-battler
Also: Auto chess, Autobattler.
You buy units from a shop and place them, then the battle fights itself while you watch. It promises strategy without any need for fast hands: the whole game is what you did before the fight.
Your spare gold is always a question: spend now or save for interest. Three copies of a unit merge into a stronger one, so an ordinary shop becomes a hunt for one particular card.
Not tower defence: A tower defence is against set waves and you can act during them. An auto-battler usually pits you against other players' boards, and you can only watch.
Seen in: Dota Auto Chess, a 2019 Dota 2 mod by Drodo Studio, set the rules: eight players, a random shared shop, and three copies merge into a two-star unit. Teamfight Tactics and Dota Underlords followed the same year.
Say it: "Auto-battler economy: five gold a round, plus one for every ten saved, up to five. Reroll costs two, and three copies merge to two-star, with about double health and damage."
https://gamecraft.dqnotes.com/genres/auto-battler

### Incremental game
Also: Idle game, Clicker.
Numbers go up, you spend them to make the numbers go up faster, and it carries on while you are away. It promises steady and absurd growth, with nothing ever taken away.
Costs grow exponentially and so does income, and so there is always a next purchase just within reach. Prestige then resets you with a multiplier, so you have the fun of going fast all over again.
Not skinner box: A Skinner box is a behaviour experiment and a criticism. An incremental game can use the same pull, and many of the best are knowing jokes about it.
Seen in: Progress Quest (2002) was a joke with no input at all, and Cookie Clicker (2013), by Orteil, made the form famous: each building costs 1.15 times as much as the last one. Universal Paperclips (2017), by Frank Lantz, turned the loop into a story about an AI.
Say it: "Each generator costs its base price times 1.15 to the power of how many you own. Offer a prestige reset once the player reaches a billion, and a permanent 2% per point."
https://gamecraft.dqnotes.com/genres/incremental

### Sandbox
Also: Software toy, Creative mode.
A game that gives you tools and rules but no set goal. It promises that the only limit is your idea, and what you make is why you play.
Players supply their own reasons, so it never runs out of content, and it creates the stories people share. But an empty box scares many players, who need a nudge to start.
Not open world: An open world lets you go anywhere. A sandbox lets you change anything. Plenty of open worlds cannot be changed at all.
Seen in: Minecraft (2011) is the great example: its creative mode gives you every block, with no goal, and people have built whole cities in it. Garry's Mod (2006) did the same with physics props and had no objectives of its own.
Say it: "Ship a sandbox mode: unlimited materials, no enemies and no score. Every material should react with at least two others, so that people find things we did not plan."
https://gamecraft.dqnotes.com/genres/sandbox

### Survival craft
Also: Survival game, Crafting survival.
You arrive with nothing. Hunger and the dark keep draining you, so you gather, craft better tools, and build a shelter. It promises the slow satisfaction of turning a hostile place into your home.
Pressure gives every log a purpose, and every crafted thing visibly solves a problem you felt. Once you are safe, though, the tension often runs out, and the game needs a new threat.
Not sandbox: A sandbox gives you the tools and no reason to use them. Survival craft takes the tools away and gives you a reason.
Seen in: Minecraft's survival mode has you racing to build a shelter before your first night. In Don't Starve (2013), complete darkness itself kills you, so a fire is not a comfort but a need. Valheim (2021) turned the same loop into a Viking co-op.
Say it: "Survival loop: food falls to zero in about four minutes, night comes every ten, and the first campfire takes three wood, so a new player is safe, just barely, on their first night."
https://gamecraft.dqnotes.com/genres/survival-craft

### Open world
Also: Free roam.
One large continuous place you can cross as you wish, doing things in nearly any order. It promises that if you can see it, you can go there.
Freedom costs the designer control over pacing, so the world itself has to pull you along. Too many map icons and the world turns into a list of chores. With too few, it is a long empty walk.
Not linear: A linear game chooses what happens next. An open world lets you choose, and has to make every choice interesting, wherever you go.
Seen in: Grand Theft Auto III (2001) made it a mainstream expectation. Assassin's Creed's viewpoints fill your map with icons once you climb them. The towers in The Legend of Zelda: Breath of the Wild (2017) show the land and no icons, so you go by what catches your eye.
Say it: "No icons on the map at the start. Put a landmark you can see from anywhere every 300 metres, so that the player navigates by looking and sets their own next stop."
https://gamecraft.dqnotes.com/genres/open-world

### Extraction shooter
Also: Extraction game, Tarkov-like.
You enter a dangerous map with your gear, grab what you can, and must reach an exit alive. If you die, you lose it all. It promises a fear that no other shooter gives you: having something to lose.
Every crate asks whether to stay or leave. This is loss aversion built into a loop, which makes the walk to the exit more frightening than any fight.
Not battle royale: In a battle royale you only win by being the last one standing. In an extraction game you can win without firing a shot, by getting out with your bag.
Seen in: Escape from Tarkov, by Battlestate Games, entered beta in 2017 and made the form famous. What you carry into a raid is lost if you die, apart from a small secure container, so a player with a full bag often stops fighting and starts running.
Say it: "Extraction rules: 30-minute raids, two exits that open at set times, and everything you carry is lost on death. The best loot sits furthest from the exits, and more enemies arrive the longer you stay."
https://gamecraft.dqnotes.com/genres/extraction-shooter

### Battle royale
Also: BR, Last man standing.
Dozens of players land on a map with nothing, scavenge weapons, and fight until only one remains, while a shrinking safe zone forces them together. It promises that this match, you might be the one.
Without the zone, everybody would hide and nothing would happen. Being beaten by one of a hundred also stings less than losing a duel, and the slow start gives weaker players a moment of hope.
Not deathmatch: In a deathmatch you respawn, and whoever scores the most kills wins. A battle royale gives you one life, and you can win it with very few kills.
Seen in: It takes its name from the 2000 Japanese film Battle Royale. Brendan Greene's mods for Arma led to PlayerUnknown's Battlegrounds (2017), with 100 players and a blue zone that closes in. Fortnite added its own Battle Royale mode the same year.
Say it: "Battle royale for 30 players: a match of about ten minutes, with a zone that shrinks in five stages. Damage outside it rises each stage, so by the end you cannot sit outside and wait."
https://gamecraft.dqnotes.com/genres/battle-royale

### Walking simulator
Also: Walking sim, Environmental narrative game, First-person exploration game.
You walk through a place and piece together a story from what you find. You cannot fail, and there is little else to do. It promises the attention you would give to a good film, in a place you can stand inside.
With nothing threatening you, you actually look at things, and so a note on a kitchen table can move you. The name began as an insult, and many of the people who make them have taken it up.
Not adventure game: An adventure game has puzzles that hold you up until you solve them. A walking simulator almost never stops you, and it trusts you to keep going.
Seen in: Dear Esther (2012) walks you across an empty island to a voice reading letters. In Gone Home (2013) you return to an empty family house, and learn what happened by reading its notes and opening its drawers. Firewatch (2016) added a voice on the radio.
Say it: "Walking sim rules: no fail state, no timer and no objective marker. Each room holds two or three objects that tell the story, and a player who misses one should still understand the ending."
https://gamecraft.dqnotes.com/genres/walking-simulator

### Visual novel
Also: VN, Dating sim (a popular kind), Sound novel.
A story told in text, over character portraits and backgrounds, with occasional choices that send it down different branches. It promises the depth of a novel, and the sense that you changed how it ended.
It is cheap to make per hour of play and can carry far more story than any 3D game can afford. Branches are the expensive part, because each one is more writing that many readers will never see.
Not walking simulator: A walking simulator tells a story through a place you walk around. A visual novel tells it through words on the screen, with branches you pick.
Seen in: It is an established form in Japan, with games such as Clannad (2004) and Steins;Gate (2009), whose choices are made by whether you answer your phone. Doki Doki Literature Club! (2017) uses the genre's cute look to set up a horror game.
Say it: "Visual novel with three endings and six choices that matter. Track two hidden values, trust and honesty, and show a flowchart after the first ending so people can go back for another route."
https://gamecraft.dqnotes.com/genres/visual-novel

### Gacha game
Also: Gacha, Hero collector, Character collector.
A game whose loop is earning currency through chores and spending it on random draws for characters. The name comes from gachapon, the Japanese capsule-toy machines. It promises the thrill of a rare pull, and a roster to show for it.
It pays for a free game by selling chances, so it is built around rates, pity counters and limited banners. Keeping that honest is hard: the same design can easily become a way to take too much money from those least able to stop.
Not loot box: A loot box is one way to sell a random item, in any type of game. A gacha game is built around that draw: the characters you pull are the game's progress.
Seen in: Puzzle & Dragons (2012) showed how much money the model could make in Japan. Genshin Impact (2020) lists a 0.6% base rate for its top rarity on its banners and guarantees one within 90 pulls, which is the pity system this toy runs.
Say it: "Show the odds on the banner itself: a 0.6% top rate with a hard pity at 90 pulls, shown as a counter. Pity carries over between banners, and we state in the same place what 90 pulls cost in money."
https://gamecraft.dqnotes.com/genres/gacha

## 13. Together

Multiplayer and netcode: what changes when the other player is a person, on a different machine, slightly in the past.

### Netcode
Also: Network code, Online code, Multiplayer architecture.
The code and the rules that keep players on different machines in the same game, when no two of them can ever see the same moment.
It decides how the whole game feels online. Added late, it usually means rewriting the game, because it reaches into how every single thing moves and who gets to decide what happened.
Not lag: Lag is what players feel. Netcode is what the game does about it. Good netcode can make a bad connection feel fine, and bad netcode can make a good one feel awful.
Seen in: Riot's article Peeking Into Valorant's Netcode walks through a full set of these decisions: a server that decides what happened, clients that guess their own movement, and hits judged by rewinding to what the shooter saw.
Say it: "Pick the netcode before we build combat. It is a two-player fighter on 60 frames, so I want peer to peer with rollback, a fixed tick and a game state small enough to save every frame."
https://gamecraft.dqnotes.com/together/netcode

### Latency
Also: Ping, Lag, Round-trip time, RTT.
How long data takes to travel between machines. Ping is the time there and back, in milliseconds, and part of it is simply the distance and the speed of light in glass.
You cannot program it away, only hide it. Everything else in this world is a way of hiding it, so the first thing to know about any netcode is how much latency it was built to survive.
Not input latency: Input latency is the delay on your own machine, from button to screen. Network latency is the delay between machines. In an online game you feel both, added together.
Seen in: Riot set out to give 70% of Valorant's players a ping of 35 ms, and got there by standing up servers around the world and building its own network backbone, Riot Direct.
Say it: "Put servers in Frankfurt, Virginia and Singapore, so that most of our players get under 60 ms of ping. Show the ping number in the corner of the scoreboard."
https://gamecraft.dqnotes.com/together/latency

### Jitter
Also: Packet delay variation, Unstable ping.
How much the delay changes from one packet to the next. Packets sent on a steady beat arrive unevenly, and some overtake each other on the way.
A steady 100 ms can be planned for. A connection that swings between 40 and 160 is much harder, and it often plays worse, because whatever is built on the timing keeps being wrong.
Not latency: Latency is how long the trip takes. Jitter is how much that time changes. A connection can have low ping and bad jitter, and that is what wifi often gives you.
Seen in: Riot's Valorant write-up says buffering smooths out uneven packets but "effectively serves as extra network latency", so its targets are about half a frame of buffer on the server and one on the client.
Say it: "Add a jitter buffer that sits at 2 updates and grows only if more than 1% of packets arrive late. I would rather be 30 ms further behind than have other players stutter."
https://gamecraft.dqnotes.com/together/jitter

### Packet loss
Also: Dropped packets, Loss.
Some of what you send simply never arrives. Each game has to choose: ignore it, wait for it to be sent again, or send things in a way that makes it not matter.
Waiting holds up everything behind the missing packet, and that stall is why action games avoid the internet's reliable protocol, TCP, for movement. Ignoring it leaves machines disagreeing for good.
Not latency: Latency is data arriving late. Loss is data not arriving at all. Resending turns loss into latency, and a big spike of it.
Seen in: Quake III avoids TCP entirely. Each update is a difference from the last update that client confirmed, so whatever got lost is automatically included in the next one.
Say it: "Do not resend inputs. Every packet carries the last eight inputs, so a player can lose one packet in ten and nobody notices."
https://gamecraft.dqnotes.com/together/packet-loss

### Tick rate
Also: Tickrate, Server tick, Update rate, Hertz.
How many times a second the server works out what has happened. Anything that reaches it between ticks has to wait for the next one.
A higher rate means fresher updates and shorter waits, but less time for each tick and more data to send. Doubling it costs real money across thousands of servers.
Not frame rate: Frame rate is how often your machine draws a picture. Tick rate is how often the game's truth is updated. You can draw at 240 FPS and still be seeing a 20 Hz server.
Seen in: Valorant runs its servers at 128 ticks a second and Counter-Strike: Global Offensive's official servers ran at 64, while Apex Legends runs at 20.
Say it: "Run the match server at 60 Hz, which gives 16.6 ms for each tick. Measure worst-case tick time at 16 players, and if it goes past 12 ms, cut work before cutting the rate."
https://gamecraft.dqnotes.com/together/tick-rate

### Server authoritative
Also: Authoritative server, Dedicated server, Client-server.
One machine, the server, decides what really happened. Players' machines send only what was pressed and show what they are told.
It is the only reliable defence against many cheats, because a hacked game can only lie to itself. The costs are a server bill and a round trip on everything, which is why it needs prediction to feel good.
Not client authoritative: In a client-authoritative game, each machine reports what happened to it, and everyone believes it. It is simple, and a cheater needs only to change the report.
Seen in: Riot's rule for Valorant is that "a server must never trust a client's view of the world". The first Diablo is the warning: it lacked those defences, and many of its online characters were altered with trainers and editing tools.
Say it: "Clients send inputs and nothing else. The server moves everyone and decides every hit, and it never accepts a position from a client, even a position that seems reasonable."
https://gamecraft.dqnotes.com/together/server-authoritative

### Peer to peer
Also: P2P, Mesh.
Players' machines talk straight to each other, with no server in the middle. Some designs have one player's machine act as host instead, so the others connect to it.
It is cheap to run and gives the shortest route between two players, which is why two-player fighters like it. It also means nobody neutral decides, upload grows with each player, and everyone can see each other's address.
Not listen server: A listen server is one player's machine acting as the server for the rest, which gives that player an advantage. True peer to peer has no such middle.
Seen in: Call of Duty: Modern Warfare 2 (2009) dropped dedicated servers on PC for its IWNet service, and players minded. A petition asking for them back passed 150,000 signatures within ten days.
Say it: "This is a two-player game, so go peer to peer with rollback. Use a relay only when two players cannot reach each other directly, and never expose an IP address to the opponent in that case."
https://gamecraft.dqnotes.com/together/peer-to-peer

### Lockstep
Also: Deterministic lockstep, Synchronous simulation.
Every machine runs the entire game itself, and only orders cross the wire. Nobody moves on to the next turn until they have everyone's orders for it.
Thousands of units cost no more bandwidth than ten, which is how strategy games worked on dial-up modems. In return, the whole game must be identical on every machine, and the slowest player sets everyone's speed.
Not rollback: Lockstep waits until it knows and never has to take anything back. Rollback carries on without knowing and takes it back when it was wrong.
Seen in: Age of Empires kept whole armies in step over 28.8k modems this way, as Paul Bettner and Mark Terrano described in 1500 Archers on a 28.8: a command issued now runs two turns later, so it can reach everyone first.
Say it: "Use deterministic lockstep with 100 ms turns, running orders two turns after they are given. Hide the delay with an immediate acknowledgement sound and a click marker, and send a checksum every turn."
https://gamecraft.dqnotes.com/together/lockstep

### Desync
Also: Out of sync, Determinism bug, Divergence.
Machines running the same game stop agreeing about it. In a game that sends only inputs, one tiny difference grows until the players are in two different matches.
It is among the hardest bugs in the trade: it can come from a rounding difference on one model of processor, and nothing looks wrong until minutes later. Games detect it by comparing checksums of the state.
Not lag: Lag means you are seeing the truth late. A desync means there is no longer one truth. Waiting fixes lag, and it never fixes a desync.
Seen in: Factorio simulates the whole factory deterministically on every client from the actions received, and in the developers' words, "if it's ever different from the server or any other client, a desync occurs".
Say it: "Compare a checksum of the whole game state every 30 ticks. If they differ, save both states to a report and stop the match. Do not use floats or unsorted containers in the simulation."
https://gamecraft.dqnotes.com/together/desync

### Delay-based netcode
Also: Input delay netcode, Frame delay.
Hold your own inputs back for a few frames, so that the other machine's inputs for the same frame have time to arrive. If they still have not, freeze until they do.
Nothing on screen is ever wrong and the idea is simple. The cost is that the controls change with the connection: play someone far away and your character answers late, and any spike stops the game dead.
Not rollback: Both are ways of waiting for the other player. Delay-based waits before it shows you anything. Rollback shows you straight away, and fixes it later if it has to.
Seen in: Guilty Gear Strive's closed beta still ran on Arc System Works' earlier delay-based connection. Rollback got its first public test in the open beta in 2021, and fighting game pros praised it.
Say it: "If we must ship delay-based, fix input delay at 4 frames across all connections so that the feel is steady, and only add more above 130 ms of ping."
https://gamecraft.dqnotes.com/together/delay-based-netcode

### Rollback
Also: Rollback netcode, GGPO, Rollback networking.
Your own inputs happen at once, and the game guesses that the other player is still doing what they were doing. When their real input arrives and differs, it reloads a saved state and quickly replays every frame since.
Your controls feel the same at 100 ms as they do offline, which is everything in a fighting game. In exchange, the other player sometimes jumps a few frames, and the game has to save and replay its entire state within one frame.
Not client-side prediction: Prediction replays only your own character against a server's word. Rollback restores the entire game state on each machine, and normally has no server at all.
Seen in: Tony Cannon released GGPO in 2006, and Street Fighter III: 3rd Strike Online Edition used it in 2011. Guilty Gear Strive built its own rollback in 2021, and GGPO went open source under the MIT licence in 2019.
Say it: "Use rollback with up to 8 frames of prediction and 2 frames of input delay. Game state must save and restore in under 1 ms, and everything in the simulation must be deterministic."
https://gamecraft.dqnotes.com/together/rollback

### Client-side prediction
Also: Prediction, Input prediction, Local prediction.
With a server in charge, your own character moves the moment you press, on the assumption the server will agree. The server's answer comes a whole trip later.
Without it, a server-authoritative game makes every press wait for a round trip, and it feels like steering through treacle. Hardly any fast online game ships without it.
Not rollback: Prediction here covers your own actions, the ones you already know. Rollback also has to guess the other player's, which you cannot know.
Seen in: QuakeWorld added it in December 1996, as part of a revamp of Quake's network code for playing over the internet.
Say it: "Predict movement and jumping on the client, so they never wait for the server. Do not predict hits, damage or deaths: wait for the server, and show a small effect in the meantime."
https://gamecraft.dqnotes.com/together/client-side-prediction

### Server reconciliation
Also: Rubber-banding (when it goes wrong), Correction, Replay.
When a server update arrives, take its word for that moment in the past, then re-apply every input it has not yet seen. If you predicted correctly, nothing on screen moves.
Updates are always old. Believing them outright drags you back a trip's worth of movement, twenty times a second. That is rubber-banding, and it makes prediction useless.
Not client-side prediction: Prediction moves you now. Reconciliation is how that guess gets checked against the server without you noticing.
Seen in: In Valorant, clients predict their own movement and the server sends corrections when they diverge. Gabriel Gambetta's series Fast-Paced Multiplayer is the usual explanation of the method.
Say it: "Number every input. Each server update says the last number it processed, so drop those and replay the rest. If the replayed position is under 5 px off, blend over 100 ms; otherwise snap."
https://gamecraft.dqnotes.com/together/server-reconciliation

### Entity interpolation
Also: Interpolation, Lerp, Snapshot interpolation.
Show other players slightly in the past, by drawing a smooth line between two updates you already have. They move evenly, and they are always a little older than they look.
Updates come far less often than frames, so other players would otherwise jump from spot to spot. The buffer also rides out one lost update. The price is more delay on everyone you are trying to hit.
Not extrapolation: Interpolation draws between two things that happened, so it is always right and always late. Extrapolation guesses ahead, so it is up to date and sometimes wrong, and it looks bad when people turn.
Seen in: Valve's Source engine defaults to 100 ms of interpolation, chosen so that even if one snapshot is lost there are still two valid ones to draw between.
Say it: "Render other players 100 ms in the past, between server updates at 20 Hz. That gives us two updates to work with, and survives one lost packet with no visible hitch."
https://gamecraft.dqnotes.com/together/entity-interpolation

### Lag compensation
Also: Hit rewind, Favour the shooter, Backwards reconciliation.
When a shot reaches the server, it rewinds everyone to where they were on the shooter's screen when they fired, and judges the hit there.
Without it, you would have to aim ahead of your target by your own ping, and every player would need a different lead. With it, you aim at what you see, and the other player sometimes dies after reaching cover.
Not client-side prediction: Prediction hides the delay on your own actions. Lag compensation hides the delay in what you can see of others.
Seen in: Valorant rewinds its world to what the shooter saw, within limits to prevent abuse. Riot estimates the related peeker's advantage at about 141 ms under baseline conditions, and says zero cannot be reached.
Say it: "Rewind hitboxes on the server by the shooter's ping plus interpolation, capped at 200 ms. Above that, a player has to lead their shots, so one awful connection cannot ruin the match for others."
https://gamecraft.dqnotes.com/together/lag-compensation

### PvP and PvE
Also: Player versus player, Player versus environment.
Whether you are playing against other people, or against the game. They change more than the opponent: a script can be learned, and a person learns you back.
The same ability gets tuned differently for each, because what is fun to do to a boss is often miserable to have done to you. Mixing them without thought is how a PvE player ends up as someone else's entertainment.
Not co-op: PvE is about the opponent, the game. Co-op is about who is beside you. Most co-op is PvE, and team-based PvP has teammates too.
Seen in: World of Warcraft used to split its realms into PvP and PvE, and Battle for Azeroth replaced that split with War Mode, a toggle letting each player choose whether they can be attacked.
Say it: "Separate PvP and PvE numbers for every ability, so that we can halve a stun in PvP without touching raids. Players opt into PvP in town and cannot be attacked otherwise."
https://gamecraft.dqnotes.com/together/pvp-and-pve

### Co-op
Also: Cooperative play, Co-operative, Team play.
Players on the same side, against the game. Good co-op makes them need each other, not just share a screen or a map.
Without some reason to rely on one another, co-op is several people playing a single-player game in the same room. A rule that makes you stop and help is what turns them into a team.
Not multiplayer: Multiplayer is any game with more than one player. Co-op is a kind of it, in which they are on one side.
Seen in: In Left 4 Dead, a survivor out of health can only be helped by another survivor, and a Hunter holds its victim down until someone pushes it off. It Takes Two has no single-player option at all.
Say it: "Downed players bleed out over 30 s and only a teammate can revive them, in 3 s. A revive leaves the helper open, so it must be a real decision in the middle of a fight."
https://gamecraft.dqnotes.com/together/co-op

### Split screen
Also: Couch co-op, Local multiplayer, Shared screen.
Players sit in one room, on one machine. Either they each get a slice of the television, or they share one view that has to keep all of them in it.
No netcode, no lag, and you can see your friend's face. But the game has to draw itself several times over, and each player gets less of the screen, so it is an expensive thing to add late.
Not online co-op: Online co-op is people on separate machines, with all the problems in this world. Couch co-op is one machine, and has none of them.
Seen in: Mario Kart 64 divided the television between four players, and some reviewers complained that its quadrants were too small to see. It Takes Two uses a split view even when you play it online.
Say it: "Four-player split screen, running at 30 FPS. Cap the draw distance at 60% and turn off shadows on the other players' views so that we stay inside the frame budget."
https://gamecraft.dqnotes.com/together/split-screen

### Asymmetric multiplayer
Also: One versus many, 1v4, Asymmetric PvP.
The two sides play by different rules, with different abilities and different goals, such as one strong hunter against a team of weak survivors.
Each side gets a fantasy that a mirror match cannot offer. It is also hard to balance, because no symmetry protects you: you can only tune it, measure who wins, and tune it again.
Not balance: A symmetrical game is balanced by default, since both sides are the same. An asymmetric one is not, and small changes can move it a long way.
Seen in: Dead by Daylight has one Killer against four Survivors, who must repair five of seven generators to power the exit gates, while the Killer hunts them down.
Say it: "One hunter against four runners, and I want the hunter to win 50% of matches at every skill level. Give the hunter 115% of runner speed to start, and report win rate by rank each week."
https://gamecraft.dqnotes.com/together/asymmetric-multiplayer

### Emotes
Also: Limited communication, Ping system, Quick chat, Pings.
Giving players a small, fixed way to communicate: a wave, a canned phrase, a mark on the map, a single note. You can say only what the designers let you say.
Open chat invites abuse, and many players avoid voice chat altogether. A small vocabulary keeps strangers useful to each other, and because abuse needs words, it removes most of that as well.
Not chat: Chat can say anything, including the worst things. Emotes can say only what was put there, which is the limit and the whole point.
Seen in: Journey's strangers cannot speak or type, and have only a musical note, because the team feared text or voice would bring biases along. Apex Legends' ping system was refined by play-testing for a month without voice chat.
Say it: "No voice or text with strangers. Add a ping wheel with four marks: go here, enemy, loot, and danger. Each ping says its meaning aloud, in the character's own voice."
https://gamecraft.dqnotes.com/together/emotes

### Griefing and toxicity
Also: Griefer, Griefing, Toxicity, Trolling, Harassment, Toxic players.
Griefing is using the game's own rules to spoil it for others, such as blocking a door or killing a teammate. Toxicity is abuse aimed at other people, usually in chat.
Both drive players away, and the people who leave are the ones you most want to keep. Reports and bans come after the harm, while design can remove the means of doing it in the first place.
Not cheating: A griefer is often playing within the rules, and that is what makes it hard to punish. A cheater is breaking them.
Seen in: Journey's players cannot hinder each other at all. EverQuest let players fight only if they had switched PvP on. League of Legends, long known for toxicity, tried a Tribunal in which players reviewed each other's reports.
Say it: "Team members pass through each other, and friendly fire does no damage. Nothing a teammate can do should be able to stop your progress. Then report and mute are the second line of defence, not the first."
https://gamecraft.dqnotes.com/together/griefing

### Spectating
Also: Spectator mode, Observer mode, Stream delay, Stream sniping.
Letting people watch a match in progress, usually from a free camera with information that the players do not have. Most such feeds run on a delay.
Watching is how players learn and how esports exist. The delay matters, since a live feed lets a friend, or the opponent, see what the player cannot. Using a stream to find a player is called stream sniping.
Not replay: A replay is watching a recording after it is over. Spectating is watching it as it happens, even if you are a little behind.
Seen in: League of Legends makes spectators of matchmade games on Summoner's Rift wait three minutes, so what they see cannot help the players.
Say it: "Add spectating on a two-minute delay, and a free camera with both teams' information. Show each player's view on demand, and let casters pause the feed without pausing the match."
https://gamecraft.dqnotes.com/together/spectating

### Matchmaking
Also: Matchmaker, Queue, SBMM, Skill-based matchmaking.
The system that picks who plays whom, trading how long you wait against how fair the game will be, and how good the connection is.
A strict search gives close games and long queues, and a loose one gives the opposite. Fewer players makes both worse, so a shrinking game loses people faster and faster, unless the rules are changed.
Not server browser: With a server browser, players choose a game themselves. With matchmaking, the system chooses for them, and most players never see how.
Seen in: Halo 2 replaced manual lobbies with playlists: you picked a type of match, and the game chose the map and mode, found other players and ranked your skill. Bungie's own testers at first preferred traditional servers.
Say it: "Start by accepting a skill gap of 50 points and widen it by 10% each second, capped at 400. Never put anyone over 120 ms of ping, and show the expected wait."
https://gamecraft.dqnotes.com/together/matchmaking

### Elo and MMR
Also: Elo rating, MMR, Matchmaking rating, Skill rating, TrueSkill, Glicko.
A number estimating skill. Elo compares your rating with your opponent's to work out how likely you were to win, then moves both by how surprised it was. MMR is the same kind of number in a game.
It finds out how good people are just from who beats whom, without ever being told. Winning takes exactly what the loser gives up, so ratings stay comparable over time.
Not rank: MMR is what the system believes about you and is often hidden. Rank is what it shows you. One is a measure, and the other is partly a reward.
Seen in: Arpad Elo's system was adopted by the US Chess Federation in 1960 and by FIDE in 1970, and FIDE uses K factors of 40, 20 and 10. Microsoft's TrueSkill for Xbox Live adds an estimate of how uncertain the rating is.
Say it: "Use Elo with K of 32 for a player's first 30 games, then 16. Start everyone at the pool average, keep the number hidden, and match on it."
https://gamecraft.dqnotes.com/together/elo

### Ranked play
Also: Ranked, Competitive mode, Ladder, Tiers, Seasons, Rank inflation.
A separate mode in which matches count, showing your standing as a named rank, such as Gold or Diamond. Those ranks are usually lines drawn across a rating, with a few extra rules.
It gives people a goal, and a reason to get better for years. Ranks are also a reward, so there is pressure to make climbing feel good, and paying out more than it takes makes every rank mean less.
Not mmr: Rank is the badge, and MMR is the number behind it. They can drift apart, which is why you can be placed with players from a different rank.
Seen in: League of Legends has ten tiers from Iron to Challenger. It added Iron and Grandmaster for the 2019 season and Emerald in 2023.
Say it: "Six tiers on hidden MMR, with the top tier at the top 1%. A win and a loss at equal rating must move you by the same amount, and we soft-reset by 20% each season."
https://gamecraft.dqnotes.com/together/ranked-play

### Smurfing
Also: Smurf, Alt account, Boosting (its cousin).
A skilled player on a new account, so they get matched with beginners. The system is doing exactly what it should, and it takes many one-sided games to find out.
New players meet the worst opponent possible at the worst time, and many of them leave. Punishing it is hard, because a new account with good results could just as well be a talented newcomer.
Not boosting: A smurf is a good player on a low account, for their own sake. Boosting is a good player on somebody else's account, to raise that person's rank.
Say it: "Give new accounts a high K so a good player climbs in 10 games, not 60. Weigh performance as well as wins for the first 20 games, and do not rate accounts that have no verified phone number."
https://gamecraft.dqnotes.com/together/smurfing

### Anti-cheat
Also: Cheat detection, Kernel anti-cheat, VAC, Bans, Ban wave.
Everything done to find and stop cheaters: software that looks for cheats on your machine, statistics on the server, and other players reviewing suspicious matches.
One cheater spoils the game for everyone else in the match. Every method has a cost, though, and the worst is banning an honest player, so most games wait for strong evidence and then ban in waves.
Not server authoritative: Authority removes whole kinds of cheating, because the server does not believe clients. Anti-cheat catches what is left, such as aim help, which looks exactly like a real player's inputs.
Seen in: Valve Anti-Cheat first shipped with Counter-Strike in 2002, and Valve presented VACNet, its machine-learning detector, at GDC in 2018. In 2023 a graphics driver update from AMD got honest Counter-Strike 2 players banned, and Valve pledged to reverse the bans.
Say it: "Server-side stats first: flag anyone over three standard deviations on headshot rate across 1,000 shots, and send them to human review. Ban in weekly waves, so that cheat makers cannot tell which build was caught."
https://gamecraft.dqnotes.com/together/anti-cheat

## 14. The machine

Loops, clocks, collisions, searches and shaders: the words for what the computer is actually doing while you play.

### Game engine
Also: Engine.
The reusable software under a game: the parts that draw, play sound, read input, simulate physics and load files. The game itself is the rules and content built on top.
Nobody can afford to rewrite all of that for every game. Picking an engine also picks what is easy and what is a fight, so it is really a design decision.
Not framework: A framework gives you pieces and you write the loop. An engine runs the loop and calls you, and it usually comes with an editor and an asset pipeline.
Seen in: id Software kept Doom's code apart from its data, and licensed the Doom engine out: Heretic, Hexen and Strife are all different games on the same one.
Say it: "Use Godot for this. It is a 2D game with few characters, so we do not need Unreal's renderer, and the small download matters more than any feature we would give up."
https://gamecraft.dqnotes.com/machine/game-engine

### Game loop
Also: Main loop, Update loop.
The circle at the centre of every game: read input, update the world, draw the picture, then do it again, usually dozens of times a second. One turn of it makes one frame.
A normal program waits for you. A game cannot, because the world has to keep moving whether you press anything or not. Everything about a game's timing follows from where in this circle a thing happens.
Not gameplay loop: The gameplay loop is what the player repeats: fight, loot, upgrade. The game loop is what the computer repeats, sixty times a second.
Seen in: id released Doom's source code, and its D_DoomLoop function is the shape in the open: start the tic, process events, run the game, update the sound, display.
Say it: "Read input at the start of the loop and apply it in the same frame, so a press never waits a whole extra frame to take effect."
https://gamecraft.dqnotes.com/machine/game-loop

### Frame budget
Also: Frame time, 16.7 ms, Performance budget.
The time one turn of the loop is allowed to take. At 60 frames a second that is 16.7 milliseconds, shared between everything: game logic, physics, AI, animation and drawing.
Run over by a single millisecond and, with vsync on, the frame misses its slot and the game drops to 30 for that moment. Budgets let a team argue about who gets the time before it runs out.
Not frame rate: Frame rate counts frames per second. Frame budget is the same fact from the other side, and it is the useful one, because milliseconds add up and frames per second do not.
Seen in: Fighting games such as Street Fighter lock to 60 frames a second, and every move is timed in those frames, so all the work of a frame has to fit inside 16.7 ms.
Say it: "We are at 22 ms in combat. Give rendering 9 ms, physics 2, AI 2 and game logic 3, and profile whichever goes over until we are under 16.7."
https://gamecraft.dqnotes.com/machine/frame-budget

### Delta time
Also: dt, Frame-rate independence, Time step.
The time the last frame took. Multiply every movement by it, and a character covers the same ground per second however fast the machine runs.
Without it, speed is tied to frame rate: the game plays in slow motion on a weak computer and at double speed on a fast screen. It is one multiplication, and old games that skipped it are often unplayable on new hardware.
Not fixed timestep: Delta time feeds the real, varying length of each frame into your maths. A fixed timestep runs the maths in equal slices whatever the frames do.
Seen in: Dark Souls II tied weapon durability to the frame rate, so weapons wore out much faster at 60 frames a second on PC than at the consoles' 30. Patches in 2015 fixed it, about a year after players reported it.
Say it: "This speed is per frame, so it runs twice as fast on a 120 Hz screen. Make it 240 px per second and multiply by delta time."
https://gamecraft.dqnotes.com/machine/delta-time

### Fixed timestep
Also: Fixed update, Accumulator, Physics tick.
Frames vary, so bank the time they take, and spend it on game logic in equal slices such as a 120th of a second. The picture can run at any rate while the rules always step the same.
Physics maths gives different answers for different slice lengths, so with a varying step the same jump reaches a different height on each machine. Equal slices also make a replay or a networked game come out exactly the same.
Not delta time: Delta time accepts that every step differs and scales for it. A fixed step refuses to differ. Most engines use both: fixed for physics, delta for visuals.
Seen in: Doom runs its game logic at a fixed 35 tics a second. Quake III Arena is the warning: certain frame rates gave slightly higher jumps, and its pmove_fixed setting chops movement into fixed steps, in code commented "to prevent framerate dependent behavior".
Say it: "Run physics on a fixed 120 Hz step with an accumulator, and draw at whatever rate the display runs. Cap it at five steps a frame so a long stall cannot snowball."
https://gamecraft.dqnotes.com/machine/fixed-timestep

### Collision detection
Also: Broad phase, Narrow phase, Spatial partitioning, AABB.
Working out which things are touching. Done in two passes: a broad phase that quickly throws out everything that is nowhere near, and a narrow phase that does the exact test on what is left.
Testing every object against every other grows with the square of the count: 100 objects need 4,950 tests, and 1,000 need nearly 500,000. Without a broad phase, collisions alone will eat the frame.
Not collision response: Detection answers whether two things are touching. Response decides what happens next: bounce, stop, take damage.
Seen in: Doom cuts each level into a blockmap, a grid of 128 by 128 unit cells, so a moving thing only checks what is listed in the cells around it.
Say it: "We have 800 enemies and are testing every pair. Put them in a uniform grid with 64 px cells and only test neighbouring cells."
https://gamecraft.dqnotes.com/machine/collision-detection

### Tunnelling
Also: Bullet through paper, Clipping through walls.
A fast object moves so far in one physics step that it starts on one side of a wall and ends on the other, and the game never sees them touch. Collision was only checked at the two ends.
It is why things fall through floors when the frame rate dips. It only happens with fast or small things, so it survives testing and then shows up on someone's slow laptop.
Not clipping: Clipping is any visible overlap, often harmless, such as a sword through a cape. Tunnelling is a real collision being missed entirely.
Seen in: Super Mario 64 moves Mario in four quarter steps each frame, checking for surfaces at every one: sub-stepping, the simplest fix for this problem.
Say it: "Bullets move 25 px a step and the wall is 12 thick, so they can tunnel. Use a swept test, a raycast from the old position to the new, for anything that fast."
https://gamecraft.dqnotes.com/machine/tunnelling

### Physics engine
Also: Rigid body simulation, Solver, Havok, Box2D, PhysX.
The part of an engine that moves things under forces and keeps them from passing through each other. It solves joints and contacts approximately, by nudging them in several passes each step.
It gives you behaviour nobody had to animate: things stack, topple and swing. It also gives you things nobody asked for, so any object that matters to the design needs testing against it.
Not character controller: Most platformers do not put their hero in the physics engine at all. A character controller is hand-written movement that only asks physics what it is touching, because real physics feels terrible to steer.
Seen in: Half-Life 2 integrated the Havok physics engine into Valve's Source engine, and built its gravity gun and its physics puzzles on it.
Say it: "Crates and barrels go in the physics engine at 60 Hz with 8 solver passes. The player stays on a custom controller so physics can never push them through a wall."
https://gamecraft.dqnotes.com/machine/physics-engine

### Raycast
Also: Ray test, Line trace, Hitscan (for weapons).
Asking what a straight line from here in that direction hits first, and how far away it is. It is the most common question in a game engine.
Line of sight, instant-hit guns, checking for ground under your feet and clicking on things are all one question. It is cheap, so you can ask it hundreds of times a frame.
Not ray tracing: A raycast is a single query about the game world, used for gameplay. Ray tracing fires millions of rays to work out the colour of each pixel, used for lighting.
Seen in: Wolfenstein 3D draws its whole picture this way: one ray for each column of the screen, and the distance it returns sets how tall that strip of wall is.
Say it: "For the guard's sight, cast one ray from his eyes to the player's head and one to their chest. If either arrives without hitting a wall, he can see them."
https://gamecraft.dqnotes.com/machine/raycast

### Tween
Also: Easing, In-betweening, Interpolation curve.
Moving a value from A to B over a set time along a curve, so you only state the start, the end and the feel. The word comes from animators drawing the frames in between.
Almost everything that moves without physics is a tween: menus, health bars, score pops, doors. The curve is what makes it feel snappy or heavy, and a linear tween looks like a machine.
Not physics: A tween knows exactly where it will end up and when, so it cannot be pushed off course. Physics only knows the next step, and it can.
Seen in: Flash put the word in front of a generation of animators: set two keyframes, ask for a motion tween, and it fills in the rest.
Say it: "Tween the health bar down over 250 ms with an ease-out, and have a white ghost bar follow it after a 400 ms delay so the player sees how much they just lost."
https://gamecraft.dqnotes.com/machine/tween

### Entity component system
Also: ECS, Data-oriented design.
A way of organising a game. An entity is only an id, components are plain data attached to it, and systems are loops that run over everything holding the components they care about.
Class hierarchies break when a design wants a flying, poisoned, explosive chest. Here you just add the parts. Keeping similar data together in memory also runs much faster.
Not component pattern: Unity's classic GameObject with attached scripts has components, but each one holds its own logic. In a true ECS the components are data only, and the logic lives in systems.
Seen in: Blizzard's Timothy Ford explained at GDC 2017 how Overwatch is built on an entity component system, and how the team used it to keep the complexity of its gameplay and netcode in check.
Say it: "Do not add a PoisonedEnemy class. Add a Poisoned component with damage and duration, and a system that ticks everything holding one. Then barrels and players can be poisoned too."
https://gamecraft.dqnotes.com/machine/entity-component-system

### Scripting
Also: Scripting language, Gameplay scripting, Lua.
Writing game behaviour in a simpler language that the engine reads and runs, while the fast, difficult parts stay in the engine.
Designers can change a boss without rebuilding the game or breaking the renderer, and a mistake costs an error message instead of a crash. It also runs slower, so heavy loops stay in the engine.
Not visual scripting: Scripting is usually typed text. Visual scripting, such as Unreal's Blueprints, is the same idea drawn as boxes and wires.
Seen in: World of Warcraft runs its interface add-ons in Lua, which is why players can replace most of its interface with add-ons they write themselves.
Say it: "Keep pathfinding in the engine, and put every boss's phases in a Lua script, so a designer can change the timings and try them without a programmer."
https://gamecraft.dqnotes.com/machine/scripting

### Hot reload
Also: Live coding, Hot swap.
Changing code or data while the game keeps running, so the new version takes effect without restarting or losing your place.
Feel gets good through hundreds of tiny adjustments. If each one takes a minute of recompiling and walking back to the same spot, you make far fewer of them, and the game is worse for it.
Not patch: A patch updates a shipped game for players, and usually needs a restart. Hot reload is a development tool, for one person trying things at their desk.
Seen in: Naughty Dog's GOAL language, written by Andy Gavin for Jak and Daxter, let a programmer edit a function, recompile it and insert it into the running game without restarting.
Say it: "Let me change jump values while the game runs. Watch the tuning file, reload it on save, and keep the player where they are."
https://gamecraft.dqnotes.com/machine/hot-reload

### Serialisation
Also: Saving state, Save data, Marshalling.
Turning a live game's state into a flat stream of bytes that can go in a file or over a network, and back again. A save file is serialisation, and so is a network message.
Forget one field and the load looks fine while something is quietly wrong. Change the format later and everyone's old saves can break, unless you stored a version number from the start.
Not save point: A save point is the design decision about when the player may save. Serialisation is the technical work of writing the game down at that moment.
Seen in: Minecraft stores its worlds in NBT, its own tree format for named binary data: level.dat for the world, one file per player, and chunks inside region files.
Say it: "Write a version number at the top of every save, and on load, upgrade old versions step by step. Add a test that saves and loads a full game, then compares every field."
https://gamecraft.dqnotes.com/machine/serialisation

### Object pooling
Also: Object pool, Free list, Garbage collection spikes.
Keeping a stock of finished objects, such as bullets, and reusing them rather than creating new ones and throwing old ones away.
In languages with a garbage collector, thousands of throwaway objects eventually force a cleanup, and it can stop the game for a visible moment. A pool makes that stop unnecessary.
Not caching: A cache keeps a result so you need not work it out again. A pool keeps an empty object so you need not allocate it again.
Seen in: Robert Nystrom's book Game Programming Patterns gives this a chapter of its own, with a particle system as its example.
Say it: "Bullets stutter the frame every few seconds. Pool them: allocate 500 at the start of the level, reuse them, and make sure a reused bullet resets every field."
https://gamecraft.dqnotes.com/machine/object-pooling

### State machine
Also: Finite state machine, FSM.
A thing is in exactly one of a list of states, and a table says which event moves it to which state. An enemy might have Patrol, Chase and Flee, joined by events like "saw the player".
It stops the impossible, like attacking while dead, and it makes behaviour readable. It gets out of hand as states multiply, because every new state needs arrows to and from all the others.
Not behaviour tree: A state machine remembers where it is and waits for an event. A behaviour tree remembers almost nothing, and asks again from the top each time.
Seen in: Each ghost in Pac-Man is always in one of three modes, chase, scatter or frightened, and the mode decides where it heads. Eating an energizer is the event that sends them all to frightened.
Say it: "Give the player a state machine with Idle, Run, Jump, Fall and Dodge. Jumping is only legal from Idle and Run, and that alone fixes the double-jump bug."
https://gamecraft.dqnotes.com/machine/state-machine

### Behaviour tree
Also: BT, Selector and sequence.
AI written as a tree of tests and actions, walked from the top. A selector tries its children until one works; a sequence runs them until one fails. So the top is the most urgent.
Adding a behaviour means adding one branch, with no arrows to rewire. Designers can read it, and an enemy reacts to changes straight away because it asks the questions again.
Not utility ai: A behaviour tree has a fixed order of priorities written by a person. Utility AI gives each option a score from the current situation, and takes the highest.
Seen in: Damian Isla's GDC 2005 talk on Halo 2 described its AI as a tree-like graph of about 50 behaviours chosen by priority. It remains one of the best-known accounts of the approach.
Say it: "Add a branch above Patrol: if the guard heard a noise, walk there and look around for two seconds. Do not touch anything else."
https://gamecraft.dqnotes.com/machine/behaviour-tree

### Utility AI
Also: Utility system, Needs-based AI, Scoring AI.
The character gives every action it could take a score based on the situation, and does the one with the highest. How hungry am I, how far is the food, how much would it help.
Behaviour comes out natural and hard to predict, with no list of rules to get stuck in. It is hard to direct, though, because the only way to change what a character does is to tune numbers.
Not goap: Utility AI chooses the best single next action. GOAP plans a whole chain of actions towards a goal.
Seen in: The Sims matches needs against objects: a fridge advertises that it can help with hunger, and each sim picks whichever option does most for how it feels right now.
Say it: "Score each action as need squared times a weight, minus a penalty for distance, so a very hungry villager crosses the map to eat, but a slightly hungry one snacks nearby."
https://gamecraft.dqnotes.com/machine/utility-ai

### Goal-oriented action planning
Also: GOAP, AI planner.
You give the character goals and a list of actions, each with what it needs and what it causes. A search then finds the cheapest chain of actions that reaches the goal.
A character can solve a problem no one scripted, and adding a new action makes old plans better. The cost is that planning takes time, and you can't easily tell what it will decide.
Not behaviour tree: A behaviour tree is the plan, written by hand in advance. GOAP writes the plan at run time, from the rules.
Seen in: Jeff Orkin built F.E.A.R.'s soldiers this way, described in his GDC 2006 talk Three States and a Plan: a state machine of only three states, with A* planning the actions.
Say it: "Do not script the soldier's reload. Give him a Shoot action that needs a loaded weapon, and a Reload action that causes one, and let the planner work it out."
https://gamecraft.dqnotes.com/machine/goap

### Pathfinding
Also: A*, A star, Dijkstra's algorithm.
Finding a route through a map. A* is the standard: always explore the most promising spot, judged by distance so far plus a guess at what is left, which finds a shortest route quickly.
Characters that walk into walls are the most visible AI failure there is. A* is cheap for one character and expensive for a thousand, so crowds need tricks such as sharing routes.
Not steering: Pathfinding plans the whole route ahead. Steering decides the next small movement, such as dodging someone in the way. Most games need both.
Seen in: Factorio's biters use A*. Its 2019 Friday Facts #317 explains how a second, coarser search across map chunks now provides the estimate, so they can route round whole lakes in a few ticks.
Say it: "Use A* on an 8-way grid with diagonal distance as the estimate, and share one path between enemies going to the same place. Limit it to 20 searches a frame."
https://gamecraft.dqnotes.com/machine/pathfinding

### Navigation mesh
Also: Navmesh, Funnel algorithm.
A set of flat polygons covering every place a character can walk. Search runs over the polygons rather than thousands of squares, and the route is then pulled tight round corners.
A grid wastes memory on open space, and paths on it turn in stiff right angles. A navmesh describes any shape of floor, including slopes and stairs, with far fewer nodes.
Not collision mesh: A collision mesh says what is solid. A navmesh says where a character is allowed to stand, usually shrunk in from walls by the character's radius.
Seen in: Mikko Mononen's open-source Recast builds navmeshes automatically from level geometry, and Unreal and Godot both use it. He also published the "simple stupid funnel" this toy uses.
Say it: "Bake a navmesh with a 40 cm agent radius and mark doors as links, so guards take doors rather than trying to walk through the wall beside them."
https://gamecraft.dqnotes.com/machine/navmesh

### Steering and flocking
Also: Boids, Steering behaviours, Separation, alignment, cohesion.
Each creature follows simple local rules for its next movement. Craig Reynolds showed in 1987 that three are enough for a flock: keep apart, match your neighbours' heading, drift to their centre.
You cannot hand-animate a thousand birds, and no one thing has to be in charge. It looks alive because it is not planned. Its weakness is that a flock can't solve a maze, which is pathfinding's job.
Not pathfinding: Flocking only knows about the neighbours in reach. Pathfinding knows the map. Crowds often use a path for where to go, and steering for how to get along on the way.
Seen in: Reynolds' boids reached feature films in Batman Returns in 1992, as its swarms of bats and its army of penguins.
Say it: "For the swarm, use boids with separation 1.5, alignment 1 and cohesion 1, and a seek force towards the player, so they surround you without any shared plan."
https://gamecraft.dqnotes.com/machine/flocking

### AI director
Also: Director, Dramatic pacing, Adaptive pacing.
A system above the enemies that watches how the players are doing, and decides when and where threats appear, to make the tension rise and fall.
Fixed encounters are the same on every play, and a steady stream of enemies goes numb. A director can give a group a rest when they need one, and that is what makes the next attack land.
Not difficulty adjustment: Adjusting difficulty aims to keep the game beatable. A director aims at pacing: it can leave strong players alone for a while because they need the quiet too.
Seen in: Left 4 Dead's director estimates each survivor's intensity, then cycles through Build Up, Sustain Peak, Peak Fade and Relax, with Relax lasting 30 to 45 seconds with few threats, in Michael Booth's account.
Say it: "Track intensity from damage taken and enemies nearby. Above 75%, stop spawning for 30 seconds or until they move on, and do not spawn anything in their view."
https://gamecraft.dqnotes.com/machine/ai-director

### Seed
Also: Random seed, World seed, Pseudo-random number generator.
The one number a random generator starts from. Computers fake randomness with maths, so the same seed always gives the same sequence, and so the same world.
A whole world can be shared as a few digits, or rebuilt rather than stored. It makes bugs reproducible, too. All of it breaks the moment anything else, such as the clock, sneaks into the sums.
Not random: A seed does not make things more random. It makes randomness repeatable, and that is what you want it for.
Seen in: In Minecraft, anyone who enters the same seed in the same version gets the same world. No Man's Sky builds its 18 quintillion planets from a 64-bit seed, storing almost none of them.
Say it: "Generate the level from a seed, and print the seed on the death screen, so a player can send me the exact level the bug happened on."
https://gamecraft.dqnotes.com/machine/seed

### Procedural generation
Also: Procgen, PCG, Roguelike levels.
Content made by an algorithm rather than by hand, at build time or as you play. The craft is in the rules around the randomness, since pure randomness rarely gives you a good level.
It buys replayability and scale that no team could build by hand. The danger is that everything looks different and feels the same, and that it can make things that are impossible to finish.
Not random: Random is a dice roll. Procedural generation is a set of rules that uses dice rolls and guarantees what must hold: there is an exit, and you can reach it.
Seen in: Spelunky builds each level from a 4 by 4 grid of rooms: it walks a route from top to bottom first, then fills in rooms from templates that keep that route open.
Say it: "Generate the route first, then fill the rooms, so every level can be finished without bombs. Add a check that walks it, and throw away any level that fails."
https://gamecraft.dqnotes.com/machine/procedural-generation

### Perlin noise
Also: Gradient noise, Fractal noise, Octaves, fBm.
Random numbers that change smoothly, made by Ken Perlin. Add several layers, called octaves, each twice as fine and half as strong, and you get something that looks like hills and rock.
Plain randomness looks like television static. Smooth noise has the large shapes and small details natural things have, and it needs no storage: any point can be worked out on demand.
Not white noise: In white noise each value ignores its neighbours, so it looks like static. In Perlin noise neighbours are close in value, so it looks like terrain.
Seen in: Minecraft's terrain uses it: Notch has described it as interpolated 3D Perlin noise offset along the height axis. Perlin himself won a 1997 Academy technical award for the technique.
Say it: "Use Perlin noise for the terrain height, 5 octaves at a gain of 0.5. Add a second, slower layer for biomes, so forests and deserts come in large connected patches."
https://gamecraft.dqnotes.com/machine/perlin-noise

### Wave function collapse
Also: WFC, Model synthesis.
Fill a grid with tiles that have rules about what may sit next to them. Each cell starts as anything; pick the most certain, decide it, and let the rules narrow its neighbours. Repeat.
You describe what fits together, not how to build it, and a small tileset gives endless varied results that always obey the rules. If a cell runs out of options you have to back up or start again.
Not perlin noise: Noise gives you smooth values and knows nothing about the things you place. Wave function collapse knows only about what can sit next to what.
Seen in: Maxim Gumin published the algorithm in 2016. Oskar Stålberg used it in Bad North, and in a different way in Townscaper, where it works out which pieces fit the shape you have just clicked.
Say it: "Generate the dungeon with wave function collapse: 30 tiles with socket rules on each edge. Restart on a contradiction, and bias the weights towards corridors."
https://gamecraft.dqnotes.com/machine/wave-function-collapse

### Shader
Also: Fragment shader, Pixel shader, Vertex shader, GLSL, HLSL.
A small program that runs on the graphics card, usually once for every pixel or corner of a model, in parallel. It decides position and colour, and that decides how everything looks.
A shader changes the whole look of a game without changing a single model. Each one also runs millions of times a frame, so one expensive line can use up the frame budget on its own.
Not texture: A texture is a picture stored in memory. A shader is the code that decides what to do with it.
Seen in: Return of the Obra Dinn renders its scene in greys, then a post-processing shader compares each pixel with a tiling dither pattern to get its 1-bit look, as Lucas Pope set out in his dev log.
Say it: "Write a toon shader for the characters: three flat light bands and a dark rim, so they read clearly against the painted backgrounds."
https://gamecraft.dqnotes.com/machine/shader

### Draw call
Also: Batching, Instancing, Sprite atlas.
One request from the game to the graphics card to draw a batch of triangles. Each one costs CPU time for setting things up, whatever the size of the thing being drawn.
A thousand small objects drawn one at a time can be far slower than one huge model. The fix is batching: put things that share a material and texture into one call.
Not triangle count: Triangles are mostly the graphics card's work, and draw calls are mostly the CPU's. A low-poly game can still crawl if every object is its own draw call.
Say it: "We are at 6,000 draw calls because each tree has its own material. Share one material and an atlas, and use GPU instancing. I want it under 1,000."
https://gamecraft.dqnotes.com/machine/draw-call

### Level of detail
Also: LOD, Pop-in, Nanite.
Using a simpler version of a model the further away it is. Done well, you never notice. Done badly, you see things change shape as you approach, which is called pop-in.
A far tree that is ten pixels wide does not need 50,000 triangles. Level of detail lets a game draw a city at the cost of a street. Someone has to make those lower versions, by hand or by tool.
Not culling: Culling decides whether to draw something at all. Level of detail decides how much of it to draw once you have.
Seen in: Fortnite moved to Unreal Engine 5.1's Nanite in Chapter 4, in which detail is chosen by the engine as small clusters of triangles, rather than as whole hand-made versions of each model.
Say it: "Add three LODs to every tree: full detail to 20 m, a quarter of the triangles to 60 m, then a flat billboard. Fade between them over 2 m so they do not pop."
https://gamecraft.dqnotes.com/machine/level-of-detail

### Culling
Also: Frustum culling, Occlusion culling, Potentially visible set.
Not drawing what you can't see. Frustum culling skips things outside the camera's view. Occlusion culling skips things hidden behind something else, which is harder to work out.
Most of a level is behind you or behind a wall at any moment. Sending it to the graphics card wastes the frame on pictures that get painted over or never appear.
Not level of detail: Culling removes things you can't see. Level of detail simplifies things you can.
Seen in: Quake works out occlusion ahead of time: its map tool stores, for each region of a level, the set of regions that could possibly be visible from it, and then it applies frustum culling to those.
Say it: "Frustum cull everything per object, and bake occlusion for the interiors. When we enter a building, the whole street behind the wall should stop drawing."
https://gamecraft.dqnotes.com/machine/culling

### Dynamic resolution
Also: Dynamic resolution scaling, DRS, Resolution scaling.
The game watches its own frame time and lowers the number of pixels it renders when things get heavy, then scales the picture up to fill the screen.
Players notice a stutter long before they notice a softer picture, so trading pixels for time is usually a good deal. It only helps if drawing pixels is what is slow; it can't fix an overloaded CPU.
Not upscaling: Upscaling, such as DLSS or FSR, is how a smaller image gets back to full size. Dynamic resolution is the decision about how small to render in the first place. They are often used together.
Seen in: Doom (2016) changes its horizontal resolution on PS4 and Xbox One to protect 60 frames a second, as id Software confirmed. Its own anti-aliasing is built to hide the changes.
Say it: "Hold 60 with dynamic resolution: keep GPU time under 15 ms by dropping as low as 70%, and recover slowly so the sharpness does not visibly pump."
https://gamecraft.dqnotes.com/machine/dynamic-resolution

## 15. Making it

From grey boxes to gold: how a game actually gets built, tested, cut and shipped.

### Graybox
Also: Blockout, Whitebox, Greybox.
A playable version of a level or game made only of plain untextured shapes. The hero of this site, Noun, begins as one: a square with the physics values the code calls GRAYBOX.
Layout problems are cheap to fix while a level is boxes and expensive after artists have dressed it. Grey also stops testers commenting on the art, so they tell you about the space.
Not prototype: A prototype asks whether an idea is any good, and can be made of anything, even paper. A graybox is a real level in the real engine, without its art.
Seen in: Valve's Source engine comes with plain "dev" grid textures in orange and grey, marked so you can read distances off them, for blocking out a map before any art exists.
Say it: "Graybox all three arenas this week, with no textures. I want to play them and move the cover around before an artist touches anything."
https://gamecraft.dqnotes.com/making/graybox

### Prototype
Also: Proof of concept.
A quick, throwaway version built to answer one question, such as "is this fun?" or "can the engine do this?". You expect to bin it.
An idea on paper costs nothing and proves nothing. A few days of ugly work can save years spent building something that was never going to be fun.
Not vertical slice: A prototype answers one question as cheaply as it can, and looks terrible. A vertical slice shows the finished quality of the whole game, and costs far more.
Seen in: Nintendo first tried out Breath of the Wild's physics and chemistry in a flat 2D prototype drawn like the original Zelda, and showed it at GDC in 2017. Fire, wind and metal were already interacting.
Say it: "Before we plan anything, spend three days prototyping the grapple with grey cubes. I only want to know whether swinging is fun. Throw the code away afterwards."
https://gamecraft.dqnotes.com/making/prototype

### Find the fun
Also: Following the fun.
Prototyping until you discover what is actually enjoyable in an idea, and then building towards that rather than towards the original plan.
The fun is often not where the plan said it would be. Teams who protect the plan can spend years polishing the part nobody enjoys.
Not polish: Polish makes a thing that is already good feel better. Finding the fun is working out which thing that is, and it has to come first.
Seen in: Grand Theft Auto began at DMA Design as Race 'n' Chase. As its developers tell it, a bug made police cars ram the player, testers enjoyed that far more than the racing, and the team built the game around it.
Say it: "Nobody in the playtest cared about the racing, but everyone laughed when the cops crashed into them. Stop tuning lap times and make a prototype that is only police chases."
https://gamecraft.dqnotes.com/making/find-the-fun

### Game jam
Also: Jam.
An event where people make a whole small game in a fixed short time, often 48 hours, usually to a theme. Ludum Dare has run since 2002 and the Global Game Jam since 2009.
A hard limit forces you to find the smallest version of an idea and finish it. Finishing is a skill, and most people practise it far too rarely.
Not hackathon: A hackathon usually builds a product demo, often judged by sponsors. A jam is about making and finishing a game, and most have no prize worth the name.
Seen in: Baba Is You started at Nordic Game Jam in 2017, and Superhot started in the 7 Day FPS jam in 2013. Both ideas were small enough to test in days, and both became full commercial games.
Say it: "Let's jam it this weekend. Two people, 48 hours, one mechanic, and we throw out anything that is not playable by Saturday night."
https://gamecraft.dqnotes.com/making/game-jam

### Game design document
Also: GDD, Design doc.
The written description of what a game is and how it works: its rules, systems, content and goals. On a good team it is a living reference that gets changed, not a contract.
Everyone on a team needs one place to check what was decided. Written too early and too long, though, it describes a game nobody has played yet, and nobody reads it.
Not one-pager: A design document holds the detail of how things work. A one-pager sells or summarises the idea, and fits on a single sheet.
Seen in: The Doom Bible, written by Tom Hall at id Software in 1992, planned a story and characters in detail. Most of it was thrown away as the game became what John Carmack and the team found was fun to play.
Say it: "Keep the design doc to how each system works and why, with one page per system. Update it the same day a playtest changes something, or delete that page."
https://gamecraft.dqnotes.com/making/game-design-document

### One-pager
Also: One-page design, Pitch sheet.
A single page that says what a game or a system is: what the player does, who it is for and why it is exciting. If it won't fit on one page, you probably haven't decided yet.
People read one page. They will also pin it to a wall and use it. It forces you to choose what matters.
Not game design document: A one-pager is short enough that everybody reads it. A design document is long enough to answer questions, so people only look things up in it.
Seen in: Stone Librande, a designer on Spore and Diablo III, gave a talk at GDC in 2010 on one-page designs: single illustrated sheets, drawn like posters and pinned up for a team, as a replacement for long documents.
Say it: "Write me a one-pager on the fishing system before we schedule it: what the player does, one drawing of the screen, and the one reason it makes the game better."
https://gamecraft.dqnotes.com/making/one-pager

### Vertical slice
Also: Slice.
A short part of the game finished to final quality in every discipline at once: art, sound, code, design and interface. It is one thin slice through the whole cake.
It proves that you can reach the quality you promised and tells you what a minute of the real game costs to make, so the rest can be estimated. It is also what publishers usually ask to see.
Not horizontal slice: A horizontal slice covers the whole game at low quality, so you can check its length and flow. A vertical slice covers a sliver of it at full quality, so you can check it is good.
Seen in: A slice can also make a promise a game fails to keep. Aliens: Colonial Marines was shown at E3 in 2011 with a demo that looked far better than the game that shipped in 2013, and Sega and Gearbox were sued over the difference.
Say it: "For the funding milestone, build a ten-minute vertical slice: one level, final art and audio, one enemy type, one boss. Log the hours, because that is how we will budget the other eleven levels."
https://gamecraft.dqnotes.com/making/vertical-slice

### Pre-production
Also: Preprod.
The phase in which a small team works out what the game is and how to make it, with prototypes, tools and plans, before a large team is hired to build it.
A question costs a small team a week. The same question found later costs a full team a month, because all of them are waiting for the answer.
Not production: Pre-production decides what to make, and it is allowed to throw work away. Production makes it, and throwing work away is now very expensive.
Seen in: Kotaku's 2019 report on BioWare's Anthem described nearly seven years of development in which key decisions kept being put off, and said that most of the game that shipped was built in the last 18 months or so.
Say it: "We stay in pre-production until there is a fun prototype, a vertical slice and a costed plan. Fifteen people until then. No hiring for content until those three exist."
https://gamecraft.dqnotes.com/making/pre-production

### Production
Also: Full production.
The main phase of making the game, when the design is settled and the full team builds the levels, art, code and sound. It is mostly a matter of keeping work flowing.
This is where most of the money goes. Output depends on the slowest step, so adding people to any other step only makes a bigger pile in front of that one.
Not producer: Production is a phase of the work. A producer is the person who keeps it on schedule, and they work in every phase.
Say it: "Art is delivering twelve levels a month and animation can only take four. Do not hire more artists. Move two people to animation and let's see where the pile moves."
https://gamecraft.dqnotes.com/making/production

### Milestone
Also: Deliverable, Milestone build.
An agreed date when a defined piece of work must be delivered and checked. With a publisher, passing one is commonly how the next part of the money gets paid.
Without checkpoints, a project can look fine right up until the end. Milestones force an honest look at the real game on a known day, while there is still time to change course.
Not deadline: A deadline is any date something is due. A milestone is a checkpoint with an agreed definition of done, and for a funded studio it often has money attached.
Say it: "Milestone 4 is 1 March. It means three levels playable start to finish on a console dev kit with no crashes. If that is not true on the day, we report it as a miss."
https://gamecraft.dqnotes.com/making/milestone

### Alpha
Also: Feature complete, Feature freeze.
The point when every feature is in the game and can be played, however buggy and however ugly. After it, the rule is that no new features go in. Each studio draws this line slightly differently.
Every new feature brings new bugs. Bugs can only fall to zero once features stop arriving, so a game with no alpha date has no real ship date either.
Not beta: At alpha all the features exist, but content and quality are still rough. At beta all the content exists too, and the only remaining work is fixing bugs.
Seen in: Minecraft made the word public. Its alpha began in June 2010, and people could already buy the game while major systems were still being added. Inside studios, alpha usually means something much stricter.
Say it: "Alpha is 1 October. Anything that is not playable in the build by then gets cut, however good it is. After that date we only fix what is already in."
https://gamecraft.dqnotes.com/making/alpha

### Beta
Also: Content complete, Open beta, Closed beta.
The stage when all features and content are in and the work is finishing and fixing. A public beta opens this to outside players, usually to test servers and balance.
You find out how a game stands up to a crowd only by having one. A beta weekend gets that answer at a time when failure costs little.
Not demo: A demo is a finished sample made to sell the game. A public beta is unfinished, and is there to test the game. Plenty of public betas are also marketing, and players can tell.
Seen in: Diablo III ran an open beta weekend in April 2012, and on launch day that May still met a login rush large enough that "Error 37" became a joke among players. A beta tests the size of crowd it gets.
Say it: "Run a closed beta with 5,000 players in May, then an open weekend in June, with server capacity set at half our launch estimate so we hit the limit on purpose."
https://gamecraft.dqnotes.com/making/beta

### Gold master
Also: Going gold, Release candidate, RTM.
The final approved build, the one that goes to the disc plant or the storefront. "Going gold" is named after the gold master discs that used to be sent for pressing.
It is the moment that can no longer be undone for anyone who buys a disc. Everything after it can only reach players as a patch.
Not launch: Going gold is when the build is finished and handed over. Launch is when players get it, and it may be weeks later.
Seen in: CD Projekt Red announced that Cyberpunk 2077 had gone gold on 5 October 2020. Three weeks later it delayed the game from 19 November to 10 December, to get its day-one update right across nine platforms.
Say it: "Release candidate 3 goes gold on Friday if it clears the test pass with no severity-one bugs. Anything we find after that goes in the day-one patch."
https://gamecraft.dqnotes.com/making/gold-master

### Crunch
Also: Crunch time, Death march, Overtime.
Long stretches of forced overtime to hit a date, sometimes for months and often unpaid. It is common in the games industry, it damages people, and it is usually the result of a planning failure.
It is sold as dedication. The older research on working hours says that output per hour falls with long hours, until the extra hours buy nothing and the error rate climbs. People burn out and leave, and they take their experience with them.
Not passion: Choosing to stay late on your own project is one thing. Being expected to, with your job or your credit at stake, is crunch, and calling it passion is how it gets accepted.
Seen in: The "EA Spouse" post in 2004 described mandatory days from 9am to 10pm, seven days a week, at Electronic Arts. In 2018 Dan Houser said "100-hour weeks" on Red Dead Redemption 2, then clarified that it meant a few senior writers for three weeks. In 2020 CD Projekt Red made six-day weeks mandatory on Cyberpunk 2077, having said earlier that crunch there would not be compulsory.
Say it: "We are not solving this with overtime. We are three months over, so cut the second biome and the co-op mode now, and keep everyone at 40 hours."
https://gamecraft.dqnotes.com/making/crunch

### Scope
Also: Project scope.
How much game you are trying to make: how many features, levels, characters and hours. It is the one project variable you fully control, and beginners nearly always set it too high.
Each thing you add has to work with everything already there, so costs compound, and the last features cost far more than their estimates. Most projects that die, die of this.
Not budget: Budget is what you can spend. Scope is what you are trying to buy with it. When they do not match, one of them has to change.
Seen in: Portal was kept to a few hours long and sold inside The Orange Box in 2007. A short game built around one idea let a small team take that idea all the way to the end.
Say it: "This is out of scope. We have a two-person team and nine months, so it is one character, one city, and four hours of play. Co-op can be a sequel."
https://gamecraft.dqnotes.com/making/scope

### Scope creep
Also: Mission creep.
The project slowly growing during development, through a long series of small additions that each seemed reasonable, without the deadline or the budget moving to match.
No one decides to make the game twice as big. It happens one yes at a time, and nobody notices the date has become impossible until it is close.
Not feature creep: Scope creep is about the whole project growing against its time and money. Feature creep is about the game itself getting more and more stuff, whether it needs it or not.
Seen in: Star Citizen's 2012 crowdfunding campaign kept adding stretch goals as money came in. Ten years on, with more than 500 million dollars raised, the game was still in alpha.
Say it: "Every new request goes on the list with its cost in weeks, and for each one that gets in, something of the same size comes out. The date does not move."
https://gamecraft.dqnotes.com/making/scope-creep

### Feature creep
Also: Featuritis, Kitchen sink design.
The game collecting more and more features, because each sounded good, until they crowd each other out and the original idea is hard to find.
Every pair of features has to be tested together, so the testing grows far faster than the feature list. Players also have to learn all of it, and a game with everything often feels about nothing.
Not depth: Depth comes from a few rules that combine in many interesting ways. Feature creep adds new rules that mostly do not.
Seen in: Duke Nukem Forever was announced in 1997 and released in 2011. 3D Realms moved engine early on, from Quake II to Unreal, and kept reworking the game to include what newer games were doing.
Say it: "We do not need crafting, or fishing. Name one player problem each of these solves. If nobody can, they come out, because each one adds to every test pass."
https://gamecraft.dqnotes.com/making/feature-creep

### Kill your darlings
Also: Cutting.
Cutting something you love because the game is better or more finishable without it. The advice comes from writing: Arthur Quiller-Couch told students to "murder your darlings" in a 1914 lecture.
The things you are most attached to are the ones you are worst at judging, particularly if they took months. Time already spent is gone either way.
Not sunk cost: Sunk cost is the trap: keeping something because of what it has already cost. Killing your darlings is the way out of it.
Seen in: Blizzard spent about seven years on a massive online game called Titan, then cancelled it. The team that remained took some of what it had into a far smaller idea, and that became Overwatch.
Say it: "I love the grappling hook too, but it took fifteen weeks and scored lowest in the playtest. Cut it. We can use the time on the boss fights people actually mention."
https://gamecraft.dqnotes.com/making/kill-your-darlings

### Playtest
Also: User testing, Hallway test.
Watching other people play your game to see what really happens. You say nothing and take notes, and you do not explain or help.
You know where everything is and what everything means, so you can no longer see your own game. Five minutes of watching a stranger can overturn a month of debate.
Not qa: A playtest asks whether the game is fun and understandable. QA asks whether it is broken. A bug-free game can fail a playtest, and the reverse.
Seen in: Ken Birdwell's 1999 article on Half-Life's development describes how Valve ran tests: testers played while team members sat by, took notes and were not allowed to help, and it changed how the game was designed.
Say it: "Get three people who have never seen the game. Sit behind them, say nothing for 20 minutes, and write down every time they stop or look confused."
https://gamecraft.dqnotes.com/making/playtest

### Think-aloud
Also: Think-aloud protocol.
A playtest in which you ask players to say what they are thinking as they play. It comes from psychology and from usability research, where Clayton Lewis applied it at IBM in the early 1980s.
Footage shows you what a player did but not why. Only their own words show the wrong idea they had in their head, and that is often the thing you can fix.
Not survey: A survey asks people afterwards what they remember and think they felt. Thinking aloud catches the belief at the moment it steers what they do.
Say it: "Ask them to keep talking as they play: what they think that thing is, what they are trying to do. If they go quiet, just ask "what are you thinking?" and never answer their questions."
https://gamecraft.dqnotes.com/making/think-aloud

### Telemetry
Also: Analytics, Game analytics, Instrumentation.
Data the game records and sends back about what players do: where they quit, how long a level takes, which weapons they use. It should be disclosed and it should not identify people needlessly.
It shows what happens across thousands of players, which no playtest can. It tells you where a problem is, but not why it is happening.
Not playtest: A playtest gives you the why from a handful of people. Telemetry gives you the what from all of them. You usually need both.
Seen in: Microsoft's user researchers built a system called TRUE, used on games including Halo 2, that logged events together with video clips of the moment, so a spike in the numbers could be watched as well as counted.
Say it: "Log every level start, death and quit with a timestamp and the position. I want a funnel from the tutorial to the credits, and I want to know the step where we lose the most people."
https://gamecraft.dqnotes.com/making/telemetry

### Heatmap
Also: Death map, Kill map.
A map of a level coloured by how often something happened at each spot: where players died, walked, stopped or looked. It is telemetry you can see at a glance.
A count of 400 deaths tells you little. Four hundred deaths at the same corner tells you exactly what to fix, and a second map tells you whether you fixed it.
Not minimap: A minimap is for the player, while they play. A heatmap is for the team, and is built from many players' sessions.
Seen in: Wired's 2007 piece on Halo 3 described Microsoft's test lab making maps of where players died so that Bungie could spot unfair spots. Valve went further and published death maps from Half-Life 2: Episode Two.
Say it: "Pull a death heatmap for the bridge level from the last 2,000 sessions. If it is one hot corner, move the turret back five metres, then pull the map again next week."
https://gamecraft.dqnotes.com/making/heatmap

### Bug triage
Also: Triage, Bug scrub.
Deciding which bugs to fix first, and which never to fix, using how bad each one is (severity) and how often players meet it (frequency).
There is always more wrong than there is time. Without triage, the scariest-sounding bugs get fixed, while a mild thing that annoys everybody ships.
Not severity: Severity is only how bad it is when it happens. Priority, which is what triage decides, also counts how often it happens and what it costs to fix.
Say it: "Rank by severity times frequency. The save-corruption bug is severity one at 0.1% of players, so it ranks below the stutter hitting half of them, unless the fix is a one-liner."
https://gamecraft.dqnotes.com/making/bug-triage

### Technical debt
Also: Tech debt.
The future cost of a quick fix or a messy shortcut in the code. Like a loan, it buys you speed now and charges interest on all later work until it is paid. Ward Cunningham coined the comparison in 1992.
A little debt on purpose is a good deal when you need to get something out. But ignored debt slows every later feature, until the team spends its time working around its own past.
Not bug: A bug is something wrong that a player can see. Technical debt can work perfectly and still make the next change take twice as long.
Seen in: Bethesda's engine has long tied parts of the game's speed to the frame rate. In Fallout 76 in 2018, players found that they moved faster when looking at the ground, where frames came quicker, and Bethesda had to patch it.
Say it: "We can hard-code the three enemy types to make the demo, but put the debt on the list. Spend one day in five after the demo pulling that into a proper data table."
https://gamecraft.dqnotes.com/making/technical-debt

### QA
Also: Quality assurance, Game testing, Testing.
The work of finding what is broken before players do, by testing every system, edge and combination on purpose and writing bugs up so they can be reproduced. It is skilled work and often underpaid.
Players all walk the same few roads, so they find the same few bugs. A tester walks into every corner, and a bug report that says exactly how to reproduce a bug is what makes it fixable.
Not playtest: QA asks "is anything broken?" and tries to break it. A playtest asks "is this good?" and tries not to interfere.
Seen in: Rare described at GDC in 2019 how it built automated tests into Sea of Thieves, so that gameplay features were checked by machine after every change and human testers could spend their time on what machines can't judge.
Say it: "Test the save system by pulling the power during a save, filling the disk, and loading a save from the previous build. Report each one with the exact steps and the build number."
https://gamecraft.dqnotes.com/making/qa

### Localisation
Also: L10n, Translation, Internationalisation, i18n.
Adapting a game for other languages and places: translation, but also longer text, right-to-left reading, fonts, voice acting, date formats and culture. Internationalisation is the engineering that makes it possible.
Many players will never see your game in English. Retrofitting a game whose text is hard-coded and whose boxes fit only English is slow and expensive, and it shows.
Not translation: Translation changes the words. Localisation changes whatever else has to change as well, such as the jokes, the layouts and sometimes the content.
Seen in: The European Mega Drive release of Zero Wing in 1991 had the opening line "All your base are belong to us". It became one of the internet's first big jokes, and a lasting example of what a cheap translation looks like.
Say it: "No text in code: everything goes in string tables from day one. Let every text box grow by at least 40% for German, and test the menus right to left before anyone builds more of them."
https://gamecraft.dqnotes.com/making/localisation

### Certification
Also: Cert, TRC, XR, Lotcheck, Submission.
The check a console maker runs before a game can be released on its machine. Sony calls its list TRC, Microsoft XR and Nintendo Lotcheck. They test things like save handling, button names and what happens when a controller is unplugged.
A failed submission costs you a fix and another full review, and the launch date does not move for you. Teams that have never done it often don't allow time for it.
Not age rating: Certification is the platform checking your game works by its rules. An age rating is a separate body such as PEGI or the ESRB judging its content.
Seen in: Xbox rules generally ask for the same features on Series S and Series X. Baldur's Gate 3's split-screen would not run acceptably on Series S, which held up the Xbox version until Microsoft agreed an exception in 2023.
Say it: "Book two full submission cycles before the launch date, and run our own TRC pass before the first one. Test unplugging the controller and pulling the disk in every screen."
https://gamecraft.dqnotes.com/making/certification

### Day-one patch
Also: Day-zero patch, Launch patch.
An update that is ready when a game launches, holding the fixes made in the weeks after the final build went off to be manufactured or certified.
A game goes on being fixed after the build is frozen, and the patch is how that work reaches players. The catch is that the disc alone remains the old game, for anybody offline and for ever.
Not unfinished game: A day-one patch usually shows that a gap exists between gold and launch, and the team used it. It does not by itself mean the game was unfinished, though it can hide that.
Seen in: Halo: The Master Chief Collection launched in 2014 with a day-one patch of about 20 GB, and its matchmaking was still badly broken at launch. A patch is no guarantee.
Say it: "Gold is six weeks before launch, so plan for a day-one patch and test the game both with it and without it, since disc-only players will get that version."
https://gamecraft.dqnotes.com/making/day-one-patch

### Post-mortem
Also: Postmortem, Retrospective.
A look back after a project, or a failure, at what went right, what went wrong and what to change next time. The aim is to fix how you work, and it is pointless if it becomes a search for someone to blame.
Teams repeat what they do not write down, and the lessons leave when the people do. Published post-mortems are also how the industry teaches itself.
Not patch notes: Patch notes tell players what changed in the game. A post-mortem tells the team, and sometimes the trade, what happened in the making of it.
Seen in: Game Developer magazine and its website Gamasutra ran post-mortems for years in a fixed format: five things that went right and five that went wrong, from the developers who made the game. Many are still online and worth reading.
Say it: "Within two weeks of launch, hold a post-mortem: five things that went right, five that went wrong, and one owner and one change for each. It is about the process, so no names."
https://gamecraft.dqnotes.com/making/post-mortem

### Art bible
Also: Style guide, Visual bible.
A shared document that sets how the game looks and how to get that look: palette, shapes, materials, proportions and line, with examples of what is right and wrong.
A large team of good artists will drift apart, each one drawing their own idea of the game. A bible lets dozens of people, and outside studios, make things that look as if one hand made them.
Not concept art: Concept art imagines what a thing might look like. An art bible states the rules every finished thing has to follow.
Say it: "Write the art bible before outsourcing: a palette of 12 colours, how thick lines get, corner shapes, three examples each of right and wrong, and how scale compares to the player."
https://gamecraft.dqnotes.com/making/art-bible

## 16. Selling it

Prices, stores, retention and monetisation, with the arithmetic on show and the tricks to refuse named as tricks.

### Premium
Also: Pay once, Buy to play, B2P.
The player pays once, up front, and owns the whole game. It is the oldest model there is.
The deal is clear and your design has nothing to fight: every decision can serve the player, since you were paid before they started. You also get just one sale per person, so you have to find enough people.
Not free-to-play: A premium game earns at the door. A free-to-play game earns inside the game, so it needs people to stay and spend.
Seen in: Hollow Knight sold for $15 with no paid extras of any kind, and its later content updates were free.
Say it: "We are premium at $20 with no store in the game. Our budget has to pay back on about 50,000 copies after Steam's cut and discounts, so let's check that against what similar games really sold."
https://gamecraft.dqnotes.com/selling/premium

### Free-to-play
Also: F2P, Freemium.
The game costs nothing to download and earns money by selling things inside it. Most players never pay, so the few who do carry the rest.
It takes away the barrier to trying a game, so far more people will. It also means your income depends on people staying and spending, and that pressure will reach into every design decision if you let it. It can be done honestly, and it takes discipline.
Not pay-to-win: Free-to-play is about when you pay. Pay-to-win is about what you pay for. A free game can sell only hats.
Seen in: Fortnite earns its money from free players who choose to buy cosmetics, which shows a free game does not have to sell advantage.
Say it: "We are free-to-play and we sell only skins. Nothing in the store changes a match. If that means a smaller ARPDAU, I accept that, and we plan the budget for it."
https://gamecraft.dqnotes.com/selling/free-to-play

### Platform fee
Also: Store cut, Revenue share, Rev share, 30%.
The share of each sale the store keeps. Steam's standard cut is 30%, falling to 25% and then 20% on revenue above set amounts.
You budget from what is left, not from the sticker price, and this is the biggest cost most people forget. At November 2026: Steam takes 30% up to $10 million of a game's revenue, 25% to $50 million and 20% above that, with tiers dating from late 2018. Epic takes nothing on a product's first $1 million a year, then 12%. Apple's rates now differ by region, cut in 2026 to 25% in China and 26% in the EU from a standard 30%, so check the page for where you sell.
Not royalty: A store fee pays for distribution and comes out of each sale. An engine or publisher royalty is a separate cost on top of that.
Seen in: Steam also charges a $100 fee per game through Steam Direct, which you earn back once it has made $1,000 in adjusted gross revenue.
Say it: "At a $20 price, plan on $14 a copy before tax, refunds and regional pricing. Run the numbers again at a 30% average discount, since many copies will not sell at full price."
https://gamecraft.dqnotes.com/selling/platform-fee

### Store page
Also: Capsule art, Capsule, Key art, Steam page.
The page where someone decides whether to buy or wishlist. Capsule art is its small banner image, which is what most people see first, in a list among others.
Most people will only ever see the capsule, so if it does not read at thumbnail size, the rest of the page does not get a look. On Steam, as of November 2026, the small capsule is 462 by 174 pixels and the header capsule 920 by 430, sizes in force since August 2024. Pay for capsule art even if you pay for nothing else.
Not trailer: The capsule gets the click and the trailer gets the wishlist. They do different jobs and you need them both.
Seen in: Balatro's capsule is a joker card and the name in big letters, and it reads even when tiny.
Say it: "Redo the capsule: one character, the title in big letters across a third of the image, and no tagline. Check it at 120 pixels wide next to ten competitors before we approve it."
https://gamecraft.dqnotes.com/selling/store-page

### The hook
Also: Elevator pitch, USP, Unique selling point.
The one thing that makes someone who has never heard of your game want it, said in a sentence. It needs to be something they can imagine doing and have not seen before.
If you cannot say it, nobody can repeat it to a friend. A long list of features is what you write when you do not have one. Find out early, since a weak hook is far cheaper to fix before you build the game than after.
Not genre: A genre says what shelf you go on. A hook says why you are the one to take down. "Roguelike deckbuilder" is a shelf. "A poker roguelike" begins to be a hook.
Seen in: Balatro calls itself a poker roguelike. Papers, Please: you are a border guard stamping passports in a police state. Both explain themselves in one line and leave you curious.
Say it: "Before we build more levels, write the hook on a card. Show it to ten strangers, and if they can't tell you back what you do in the game, we fix the card or the game."
https://gamecraft.dqnotes.com/selling/the-hook

### Wishlist
Also: Wishlists, Wishlist count.
A saved list of games a player wants on a store. When a game on it launches or goes on sale at 20% or more, Steam emails them.
It is a free way to reach interested people before and after launch, and a rough sign of interest you can watch. Treat it as a signal and not a promise: rules of thumb about how many wishlists turn into sales vary widely, so do not plan your budget on a borrowed ratio. Check what your own similar launches did.
Not pre-order: A pre-order is money up front. A wishlist is a free bookmark, which is why far more people will do it and why not all of them buy.
Seen in: Steam's documentation says any discount of 20% or more triggers wishlist emails, with a 1 to 2 week cooldown between them. Checked November 2026.
Say it: "Put the store page up a year before launch, and make our only call to action "wishlist". I want the count weekly and I want to know what moved it."
https://gamecraft.dqnotes.com/selling/wishlist

### Demo
Also: Free trial, Slice, Steam Next Fest demo.
A free piece of your game for people to try. On Steam, Next Fest is an event where many games release demos at once for visitors to browse.
A good demo sells what a trailer can only claim. A bad one gives people a reason to skip you, so it needs to be polished and to contain your best idea early. Keep it short and end it at a high point, with the wishlist button right there.
Not vertical slice: A vertical slice is an internal proof that the game works. A demo is for the public, and it needs far more polish on far less content.
Seen in: Steam Next Fest runs several times a year and many indie launches build their whole wishlist campaign around it.
Say it: "End the demo right after the first boss, on the reveal of the second weapon. Put the wishlist button on the end screen. Twenty minutes, nothing longer."
https://gamecraft.dqnotes.com/selling/demo

### Early access
Also: EA, Steam Early Access, Paid beta.
Selling a game before it is finished, and updating it in public. Steam started its programme in March 2013.
It pays for development and gets you real testers. The cost is that you only launch once, and many people will have formed an opinion from the unfinished version. Promise only what you can pay for: buyers are paying for what exists today, and many early access games are never finished.
Not crowdfunding: Crowdfunding sells a promise, often before there is any game. Early access sells a game you can play today.
Seen in: Hades was in early access from December 2018 before leaving in September 2020. Baldur's Gate 3 spent nearly three years there, from October 2020 to August 2023, and player feedback changed it a lot.
Say it: "Only go to early access once a full run is fun on its own. Publish a roadmap with dates we would be happy to miss, and sell it on what is in the build today, not on what is coming."
https://gamecraft.dqnotes.com/selling/early-access

### Discounting
Also: Sale, Launch discount, Steam sale.
Selling for less for a while to reach people who would not pay full price. Expect many of your copies to sell at a discount, and plan for it.
It finds buyers you would otherwise never reach, with no cost per copy. Done too deep and too often, it teaches your fans to wait. Steam's rules, checked November 2026: discounts run from 10% to 95%, a launch discount cannot exceed 40%, there must be 30 days between discounts outside seasonal sales, and 20% or more triggers wishlist emails.
Not price cut: A discount is temporary and ends on a date. A price cut is permanent, and players who bought last week will notice.
Say it: "Launch at 10% off for two weeks, then no sale for three months. Never go past 50% in the first year. Those who paid full price at launch should not feel foolish a month later."
https://gamecraft.dqnotes.com/selling/discounting

### Publisher
Also: Advance, Recoup, Publishing deal.
A company that pays for or markets your game in return for a share of the money. Often they pay an advance, and you get no share of sales until your share has paid it back, which is called recouping.
You are selling a share of your game for money now and for work you might not be able to do yourself. Read what they will actually do, who owns the IP, and when the money stops. Terms vary widely, so ask a lawyer and other developers before you sign.
Not investor: An investor buys part of your company. A publisher typically buys part of one game's revenue, and often does the marketing.
Seen in: Devolver Digital published Hotline Miami and Cult of the Lamb. Team Cherry self-published Hollow Knight. Both routes can work.
Say it: "We will take a $200k advance at 40%, but we keep the IP, the share drops once they have recouped twice over, and the contract lists exactly what marketing they will fund."
https://gamecraft.dqnotes.com/selling/publisher

### Streamability
Also: Watchable, Twitch-friendly, Clip-worthy.
How much fun a game is to watch someone else play: whether it makes moments people would clip and share, and whether viewers can follow it.
Streamers can give you more attention than any ad budget, but they can also make a player feel they have already seen it. Linear story games have that problem. It is why games with surprise and other people in them travel so well.
Not virality: Virality means people spreading something. Streamability is one way that can happen, where watching makes you want to play.
Seen in: Among Us came out in 2018 and took off in 2020, once streamers started playing it together. Lethal Company spread the same way in 2023: friends failing together is funny to watch.
Say it: "Add a streamer mode that hides lobby codes and removes licensed music, and make sure a viewer can tell from the screen who is winning. Our big moments need to read in a clip with no context."
https://gamecraft.dqnotes.com/selling/streamability

### Review bombing
Also: Review bomb, Off-topic reviews.
A flood of negative user reviews in a short time, often about something other than the game itself: a policy, a statement, a store deal.
The store label can move in days, and it directly hits sales. Sometimes it is an abusive pile-on, and sometimes it is the only pressure players can apply, so read what they are saying before you decide which. Steam has flagged and excluded off-topic bombs from scores since March 2019. In 2025 it added language-specific scores, partly because players in different regions can have very different experiences.
Not bad reviews: Bad reviews are people telling you about the game. A bomb is mostly a protest, and may be from people who haven't played it, though they can be right to be angry.
Seen in: Helldivers 2 got a wave of negative Steam reviews in May 2024 when Sony said PC players would need a PlayStation Network account, even in countries where that service was not available. Sony dropped the requirement within days.
Say it: "If a bomb starts, the first post says what happened and what we will do. No blaming players, no asking fans to counter-review. If they are right, we change it."
https://gamecraft.dqnotes.com/selling/review-bombing

### DLC
Also: Downloadable content, Expansion, Add-on.
Extra content sold after the base game. It can be a large expansion or a small item, and players judge the two very differently.
It earns from people who already love your game, without needing to find new ones. It works when it feels like more, and turns sour when it feels like something was cut from the game you already bought.
Not microtransaction: DLC is usually a fixed piece of content you buy once. A microtransaction is a small repeatable purchase, often of currency or a consumable.
Seen in: The Witcher 3's Blood and Wine is roughly a game in its own right, and players felt they got more than they paid for. Oblivion's $2.50 horse armour in 2006 became the standard joke about the opposite.
Say it: "No day-one DLC from content we had finished before launch. Plan one expansion for a year out, with at least ten hours and a new region, and price it at half the base game."
https://gamecraft.dqnotes.com/selling/dlc

### Cosmetics
Also: Skins, Vanity items, Hats.
Paid items that change how you look and nothing about how you play.
This is the cleanest thing to sell inside a game: nobody who skips it loses out, and match results stay fair. It works best in games where other people can see you, and can still be sold with tricks of its own, such as loot boxes and limited-time pressure.
Not pay-to-win: Cosmetics change looks. Pay-to-win changes results. The test: would a skilled player without it be at a disadvantage?
Seen in: Fortnite's skins do nothing for your chances in a match and they made it one of the highest-earning games ever. Team Fortress 2's hats made a game about shooting into a hat economy.
Say it: "Everything we sell is cosmetic, and every cosmetic has a price in money that you see before you buy, with no boxes and no timer."
https://gamecraft.dqnotes.com/selling/cosmetics

### Pay-to-win
Also: P2W, Pay for advantage.
Players can buy an advantage over other players. Money beats skill, at least some of the time.
Refuse this in anything competitive. It breaks the thing the players came for, and those without money either leave or pay. Once they leave, the people who paid have nobody to beat. Even at small edges people will notice and complain, and the name will stick to you.
Not pay-to-skip: Paying to skip a grind saves time without beating anyone. It is milder, but be honest: you may have made that grind long in order to sell the skip.
Seen in: Star Wars Battlefront II (2017) tied hero unlocks and upgrades to its loot and paid currency. The defence that this gave players "a sense of pride and accomplishment" became the most downvoted comment in Reddit's history, and EA turned the purchases off before launch.
Say it: "No item sold for money can change damage, health, speed or matchmaking, in any mode. If a designer wants to sell one, they have to write down who loses to the person who bought it."
https://gamecraft.dqnotes.com/selling/pay-to-win

### Premium currency
Also: Hard currency, Gems, Virtual currency, Paid currency.
Money you buy with real money and then spend in the game, as gems or coins. It sits between your wallet and the price, on purpose.
Refuse the tricks in this. Odd pack sizes strand you with leftovers, and a price in gems is hard to turn back into dollars, particularly for children. In January 2025 HoYoverse agreed to pay the FTC $20 million over Genshin Impact, including complaints about multi-step currencies that make spending hard to track. It disputed much of this, and the settlement requires that it sells loot boxes directly for dollars in the US.
Not soft currency: Soft currency is earned by playing, and it is a normal part of an in-game economy. Premium currency is bought, and the design question is what it hides.
Seen in: Most free-to-play games use it. It is common enough that a price tag in plain money stands out as the exception.
Say it: "Show the real price on every item, right next to the gems. Sell packs in amounts that match what things cost, and if a player has leftovers, let them cash out or buy something small."
https://gamecraft.dqnotes.com/selling/premium-currency

### Energy system
Also: Lives, Stamina, Timers, Wait or pay.
You can only play so much before the game makes you wait, with a meter that refills over time or for money.
It splits play into short sessions and brings people back, and it can stop binge-burnout, which is a real benefit. But it is a wall built so you can sell the door. If your game needs to stop people playing to earn money, ask if you are making a game or a tollbooth. Respect people who choose to wait.
Not daily limit: A daily cap with no way to pay past it is a pacing decision. An energy system with a refill for sale is a sales tool, and it is the sale that changes what it is.
Seen in: Candy Crush Saga gives five lives, one back every 30 minutes, with refills for sale in gold bars. Failing a level costs a life.
Say it: "Remove the energy meter from the campaign. If we need to pace it, end the session with a story beat and a full meter, and do not sell any way past it."
https://gamecraft.dqnotes.com/selling/energy-system

### Battle pass
Also: Season pass, Reward track.
A track of rewards you unlock by playing during a season, usually with a free line and a paid one. Dota 2's 2013 Compendium was the forerunner, and Fortnite made it the norm from late 2017.
It is a fairer deal than a loot box, since you know what you are getting and it has a known price. Its catch is the deadline, which turns a leisure activity into something like a work schedule, especially once a player has paid. Selling tiers near the end charges them again for the hours they ran out of. Make it hard to fall behind, or drop the deadline.
Not season pass (old sense): A season pass used to mean buying all of a game's planned DLC in advance. A battle pass is a reward track you work through by playing.
Seen in: Halo Infinite's battle passes do not expire, so you can keep working through one later. It shows that the deadline is a choice and the model works without it.
Say it: "Make the pass finishable in 40 hours over ten weeks at most, with no paid tier skips in the last two weeks. Once you own a pass it does not expire."
https://gamecraft.dqnotes.com/selling/battle-pass

### Loot box
Also: Crate, Random item box, Surprise mechanics.
A paid item with a random reward inside, so you pay for a chance at what you want. It runs on the same schedule that keeps people at slot machines, with a game wrapped around it.
Refuse to sell these. Belgium's Gaming Commission found in April 2018 that paid boxes in FIFA 18, Counter-Strike: Global Offensive and Overwatch broke its gambling law. Elsewhere the rules vary and keep changing. Children can buy them. And you have no way of telling a hobbyist from someone who is losing the rent, because the cost has no ceiling.
Not random drop: A random drop you earn by playing costs you time. A loot box costs you money, and that is the line that matters to regulators and should matter to you.
Seen in: Overwatch's original loot boxes held only cosmetics, and even those were found to break Belgian law, so being cosmetic did not save them. Blizzard later dropped paid boxes altogether in Overwatch 2.
Say it: "No paid random rewards. If we use a box as a surprise, it is only earned by play. Anything sold for money shows exactly what you get before you pay."
https://gamecraft.dqnotes.com/selling/loot-box

### Gacha
Also: Banner, Pulls, Wishes, Pity timer, Soft pity.
A loot box system built into the core of a game: you pay for pulls on a banner to win characters at very low odds. The word comes from Japanese capsule-toy machines called gachapon.
A pity timer caps the worst case and it is a real kindness, but the result is often that a character costs hundreds of dollars, sold one pull at a time. In Genshin Impact a five-star has a 0.6% base rate, 1.6% counting the guarantee, certain by wish 90, and a 50/50 on whether it is the banner character (about 55/45 since version 5.0). The odds are shown. The total cost, which is what matters, is left for you to add up.
Not loot box: A loot box is often a side purchase. In a gacha game, the pull is what the whole business is built on, and the characters are the content.
Seen in: Genshin Impact is the best-known example today. Its pity counter carries over between banners of the same type, which gives you a reason not to stop.
Say it: "If we must ship pulls, show the running real-money total and the most it could cost on the banner itself. Pity at 50 pulls, no 50/50, and no selling to under-18s."
https://gamecraft.dqnotes.com/selling/gacha

### Whale
Also: High spender, Big spender, VIP, Dolphins and minnows.
Trade slang for the small number of players who spend far more than anyone else. It comes from casinos, where it means a high roller.
Question the word. It is a way of not thinking about who these people are: some are rich, some are lonely, some have a gambling problem, and some are children using a parent's card. If your income depends on them, you have a duty to find out which, and a spending limit is one way to act on it. How concentrated spending is differs by game, and most published figures come from vendors, so do not borrow one: measure your own.
Not superfan: A superfan loves your game. A whale spends a lot on it. They can be the same person, but the word only measures the second.
Say it: "Stop calling them whales in our dashboards: call them top spenders. Add a monthly spending cap, on by default, and a human who contacts anyone spending over $500 in a month to ask if they are fine."
https://gamecraft.dqnotes.com/selling/whale

### Live ops
Also: Live operations, Live service, Games as a service, GaaS.
Running a game after launch as an ongoing service: events, new content, balance changes, seasons, all planned on a calendar.
It keeps a community alive for years, and it never ends. You are committing a team and a budget indefinitely, and a live game that stops being fed dies in public. Before starting, ask who is on it in year three and what happens to players' purchases when you turn off the servers.
Not patching: Patching fixes what you shipped. Live ops is a planned stream of new reasons to come back.
Seen in: Fortnite runs seasons and one-off events such as the 2020 Travis Scott concert. Concord, a 2024 live-service shooter, was shut down two weeks after launch, and Sony refunded buyers.
Say it: "Plan 12 weeks of events we can make with the team we have, reusing systems and not new art. If we can't sustain that for two years, we shouldn't make it a live game at all."
https://gamecraft.dqnotes.com/selling/live-ops

### Soft launch
Also: Beta market, Test market.
Releasing a mobile or live game quietly in a few countries to measure how people behave and fix it before the big release.
It finds problems when they are still cheap, but a small sample lies. Below a certain size, a healthy game can look dead and a dead one healthy, so work out how many installs you need before you run it. Be ready to act on the answer, including cancelling.
Not early access: Early access sells an unfinished game to fans who know it is unfinished. A soft launch is a quiet test in limited markets, mostly to measure.
Seen in: Supercell tests in soft launch and is known for cancelling games there if the numbers do not hold up, as it did with Rush Wars.
Say it: "Soft launch in two English-speaking countries, with enough installs to measure day 7 to within two points. Write down the kill bar before we start, and don't move it afterwards."
https://gamecraft.dqnotes.com/selling/soft-launch

### Retention curve
Also: Day-1 retention, D1, D7, D30, Churn, Cohort retention.
The share of players who come back after a given number of days, starting from those who first played on the same day. Day-1 retention is the share who return the next day. Churn is the opposite: those who leave.
It tells you whether people want your game, more honestly than downloads do. Curves differ a lot by genre and platform, and any "good" number you hear will be a vendor's average at some date, so compare your game with its own last version. Retention is also not the same as enjoyment. You can raise it with streaks and loss aversion and be left with people who feel trapped.
Not engagement: Retention is whether they come back at all. Engagement is how much they do once they're here.
Say it: "Day 1 dropped 5 points after the new tutorial. Check the first ten minutes for where people quit, and fix that before we touch anything later in the game."
https://gamecraft.dqnotes.com/selling/retention-curve

### DAU and MAU
Also: Daily active users, Monthly active users, Stickiness, DAU/MAU.
How many different people played on a given day, and in the last 30 days. Dividing the first by the second gives a rough idea of how often people come back.
They are the headline numbers, and they are easy to flatter: ad-bought installs lift both, and MAU falls slowly. Ask how "active" is defined, since each company picks its own. A growing DAU can hide a shrinking game, one with worse retention and more ad spend.
Not installs: Installs count how many people ever tried it. DAU counts who played today. A game with ten million installs can have nobody playing.
Say it: "Stop reporting total installs at the monthly review. Show DAU split by cohort, so we can see if new players are staying or if ad spend is holding the line up."
https://gamecraft.dqnotes.com/selling/dau-and-mau

### ARPU and ARPDAU
Also: Average revenue per user, Average revenue per daily active user, ARPPU.
Total revenue divided by players, over a period. ARPDAU divides one day's revenue by that day's players. ARPPU counts paying players only.
It lets you compare days and test changes, and that is about all. An average of mostly zeros and a few huge numbers describes nobody: if 97 in 100 pay nothing, the "average player" does not exist. Raising it can mean you have found a better offer, or that you have pushed harder on the people least able to stop.
Not lifetime value: ARPU is a rate for a period. Lifetime value is what a player brings in over their whole time with the game.
Say it: "ARPDAU went up 12% after the new offer. Split it by spender band before we celebrate: if it's all from the top 1%, I want to look at what that offer does to them."
https://gamecraft.dqnotes.com/selling/arpu

### Lifetime value
Also: LTV, LTV/CAC, Payback.
The money you expect from a player over all the time they play. If it's more than you pay to acquire one, you can buy growth with advertising.
It drives what you can spend on ads and whether the game can exist at all. It is also a forecast: it has to guess the long tail from short data, and small changes in the guess swing the answer, so an honest LTV comes with a range. It is a poor reason to design anything, because anyone chasing it starts asking how to extract more.
Not revenue: Revenue is what came in. Lifetime value is a prediction about what will come in from a player you have only just met.
Say it: "At a $2 install cost and $1.60 predicted LTV, we lose money on every install. Pausing ads until day 7 improves, and don't let anyone fix this by raising prices on current players."
https://gamecraft.dqnotes.com/selling/lifetime-value