The Tutorial and the Art of Not Explaining

A tutorial is a confession. Whatever a game chooses to explain is the thing it suspects its own design cannot communicate on its own, and the length of the explanation is a measure of the gap. This is why the tutorial is the most diagnostic twenty minutes in any game, and why so many of them read like a hostage note.

The good news is that the craft here is old, well understood, and repeatedly demonstrated. The bad news is that it keeps getting rediscovered, because the economics of a big release reward the pop-up and punish the room.

The manual was the tutorial, and the load was the reading time

I want to start with the thing that is genuinely gone, because it explains a lot of what came after. Games used to ship with paper — a booklet, often substantial, occasionally excellent. Elite in 1984 came with a manual and a novella, The Dark Wheel, written by Robert Holdstock, a real science fiction author with a real career. The manual taught you docking. The novella taught you what the game was for.

The delivery mechanism was accidental genius: the game took several minutes to load, so you sat there with the booklet because there was nothing else to do. The tape drive enforced a reading period. Whatever else you want to say about the patience a tape load taught, it built a slot in the experience where instruction fit naturally, and it cost the developer nothing in memory — which mattered enormously when the entire universe had to fit in 32K.

Copy protection then weaponised the arrangement. The code wheel, the “type the third word on page 12” prompt, LucasArts’ Dial-A-Pirate disc for The Secret of Monkey Island: publishers made the manual load-bearing so it couldn’t be left in the shop. Cynical, and it worked, and it meant that for about a decade the average player actually opened the documentation.

Then loads got fast, boxes got thin, digital distribution killed the booklet entirely, and everything the manual used to carry had to move inside the game. That is the whole origin of the modern tutorial. It’s a paper artefact that got absorbed into the software and never found the right organ.

World 1-1, and teaching by consequence

The canonical answer to the problem predates the problem. Super Mario Bros. opens on a screen with almost nothing in it, and every element is placed to extract a specific proof from you.

Mario stands at the left, facing right, alone. A Goomba walks toward you: the first moving object is also the first threat, and it teaches that contact hurts, at a walking pace slow enough to survive the lesson. Jump on it and you learn the verb. Above sits a row of blocks including a question mark, and hitting it drops a mushroom. Here is the part that is genuinely clever — most players hit that block from the left side, so the mushroom emerges moving right, hits the pipe ahead, and bounces back toward Mario. A player who thinks the mushroom is an enemy and tries to flee is chased down by the reward. The game removes your ability to get the lesson wrong.

Nintendo taught the entire vocabulary in about fifteen seconds, with no text, in a screen the player could not fail. The technique has a name in the literature — guided discovery — and its defining property is that the player produces the knowledge, so the player owns it. Being told a thing puts it in short-term memory. Doing a thing under mild consequence puts it somewhere better.

Mega Man X in 1993 ran the same play with more nerve. The opening stage drops you into a pit whose walls you can only escape by wall-jumping, a move the game never mentions. Flail at the wall and you slide down it; the slide is a hint delivered in animation. The stage ends with a fight against Vile you are designed to lose — the health bar drains, your shots do nothing, and Zero arrives to rescue you. The game teaches you its own ceiling by putting you under it. A text box saying “some enemies are too strong for now” carries the same information and none of the meaning.

Half-Life put the tutorial in a separate room, then made the game teach anyway

Valve’s solution in 1998 is the one people misremember. Half-Life has a Hazard Course. It is a discrete training level, reached from the menu, with a voice explaining ladders and long jumps — an old-fashioned tutorial, sitting in a box marked tutorial, entirely optional.

The reason it works is that Valve then built the actual game to not need it. The tram ride and the argument it makes hand you a camera and nothing else for several minutes, and Black Mesa spends its first hour showing you scientists opening doors you’ll later open, security guards shooting things you’ll later shoot, and a resonance cascade that reframes every corridor you just walked. The instruction is a rehearsal of the world.

FromSoftware found the cheapest version of the same idea. The Undead Asylum that opens Dark Souls teaches its controls through orange messages scrawled on the floor — the same message system players use on each other for the next forty hours, seeded here by the developers. The tutorial and the multiplayer are one feature. You learn to roll by reading a stranger’s note, which is precisely the habit the rest of the game needs you to have, and by the time you reach the level design that folds back on itself you have already been trained to trust the floor. Nothing is explained twice, and nothing is explained from outside the fiction.

Portal, from the team Valve hired out of DigiPen after Narbacular Drop, took that to its endpoint: the first half of the game is a tutorial and the tutorial is the entire product. Every test chamber introduces one property of the portal and then demands it back. GLaDOS narrates because a testing facility has a reason to narrate — the explaining voice is a character with motives, so the exposition does double duty as characterisation, and when she starts lying the tutorial becomes the plot.

Tunic, Baba Is You, and handing back the manual

The most interesting recent work has gone one further and made the absence of explanation into a mechanic.

Tunic’s central object is an instruction booklet, scattered through the world as collectible pages, drawn in the visual language of a late-80s console manual — including the folded-in-half creases and the sun-faded colour. It’s largely written in an invented script you cannot read. What you can read is the layout: diagrams, arrows, a circled item, a hand-scrawled annotation. Andrew Shouldice built a game where the manual is the game, and where finding out how to play is the reward loop rather than the entry fee. Anyone who grew up with a machine whose games shipped with paper knows exactly what feeling is being harvested.

Baba Is You does something colder and, I think, harder. Arvi Teikari’s puzzle game states its rules as physical blocks on the level — BABA IS YOU, WALL IS STOP, FLAG IS WIN — and you win by pushing those words around until the rules say something else. There is no tutorial because the tutorial is unnecessary and also impossible: any explanation would have to explain itself in the same grammar the game is teaching. Teikari’s answer is to open with levels so simple the grammar is the only thing in them. The player derives the rule system from three words in a field, and the whole rest of the game is that derivation compounding.

Chants of Sennaar plays a related trick with an entire invented language, turning comprehension itself into the empathy machine: you’re issued a notebook, you guess at glyphs, the game silently confirms when your guess holds up across contexts. The design lesson underneath all three is identical — withholding information is only generous when the player has a reliable way to earn it.

What the withholding actually costs

I want to be honest about the failure mode, because “don’t explain anything” has become a piece of received wisdom among people who have mistaken obscurity for depth.

Guided discovery requires the designer to control the environment completely. Mario 1-1 works because Nintendo could guarantee the Goomba’s arrival time and the mushroom’s bounce. In an open world, in a systems-heavy sim, in anything with an inventory, the designer cannot guarantee the player’s position, order or attention — and a lesson delivered to a player facing the wrong way is no lesson. This is exactly why immersive sims tend to sprawl into text, and why so many otherwise elegant games collapse into a wall of tooltips the moment the player gets freedom. Freedom and teaching pull against each other; that’s a real constraint rather than a failure of nerve.

The other cost is exclusion. A game that teaches only through consequence is a game that teaches only players who can survive the consequence. If your grammar lesson is a pit, you’ve filtered out everyone who can’t climb — and unless climbing is the point, you’ve paid for elegance with an audience.

So the test I’d apply is narrow. Ask what the game explained, and then ask whether the room could have said it instead. Every pop-up that survives that question earns its place. Most don’t, and the ones that don’t are usually describing a system the designer already suspected was arbitrary.

The tutorial is where a game tells you what it doesn’t trust about itself. Read it that way and it becomes the most useful thing in the box — which, once, is exactly what it was.