Teaching Through Play
The opening that hands the player the rules without ever stating them, so the first thing they learn is that the game can be read.
A good puzzle game does not begin by explaining itself. It begins by arranging the world so the first rule is impossible to miss and impossible to misread, and then stands back so the player infers it. Nothing is announced. A block sits where a jump would clear it; a coloured line runs from a dot to an edge; two words on the floor say what is true. The player acts and the world answers back, so a rule the game never spoke is now the player’s own.
This is teaching by demonstration rather than instruction, and it is one of the oldest problems in the craft. A text tutorial tells the player what to do before they have any reason to care. It breaks the spell, and the player forgets it within minutes. The alternative is to stage the world as a lesson: present the mechanic where it can be observed safely, tried, failed without cost, and grasped visibly. The trick lives entirely in the staging. The correct action has to be obvious in hindsight but not trivial in advance: too plain and the player feels patronised, too hidden and they leave the opening having learned nothing and begin the rest of the game with a broken model.
The reward for getting it right is not just comprehension. A rule the player inferred is a rule they own, and ownership is what every later puzzle trades on. This article is about how that ownership is built. A game shows its mechanics and settles the conventions a solver will rely on, then paces the reveal of new rules so the curve teaches rather than overwhelms.
The Mechanism (How it works)
Show the mechanic, do not state it
The core move is to introduce a new mechanic in a context where the player can watch it work before anything is at stake. The canonical case is Super Mario Bros. World 1-1 (1985), studied to exhaustion because it teaches an entire control vocabulary in its first thirty seconds with no text at all. The camera scrolls right, so the player learns to go right; a Goomba approaches in open ground, so the player learns that contact can kill; a glowing block sits overhead, so the player learns to jump into ceilings; the mushroom that emerges teaches that power-ups exist. The level layout is the tutorial in itself. The load-bearing detail is the safe margin of empty ground to the left of the first Goomba: the designer hands the player time to watch before any reaction is forced. It works like a field-guide margin note saying stop here and look at this.
Portal (Valve, 2007) industrialised the technique into the single-purpose room. Each opening chamber is stripped of anything that would distract from the one mechanic it teaches. The first chamber covers looking around. A later one introduces the cube with its pressure plate. A third is built around the redirected energy pellet. The mechanic is demonstrated by the environment before the player is asked to apply it, and only once it is owned does the next chamber combine it with something already learned. The pattern is learnable and portable. Isolate the mechanic and strip out the noise while keeping the solution path visible, then combine. After Portal, folding the tutorial into the first hour of play became the default for the genre.
The principle has a name borrowed from prose craft, show, don’t tell, and it depends on affordances the player can read. A door has a frame and a handle; a ledge shows a lip; a pressure plate sits in a depression under a different texture. The puzzle-game version of this is the close cousin of the live escape room’s problem, where props and spatial layout have to teach what a tooltip cannot, and of interactive fiction’s diegetic design, where the world’s own affordances replace a menu. The same instinct, that the world should be legible without a UI layer narrating it, runs under all three.
The test that a staged lesson has landed is ownership, not comprehension: not whether the player can recite the rule but whether they can act on it. The real check is a fresh pair of hands on the opening with the sound off and your mouth shut. If they go for the right action unprompted, the convention landed. The same diagnostic runs in reverse, and that is why the urge to add a tooltip is worth distrusting: it usually signals a staging problem, not a teaching one, and the fix is to arrange the situation so the rule is the only coherent inference rather than to narrate it.
Build a convention, then never break it
Showing a mechanic once is not enough; the player has to be able to trust that what they learned still holds. A convention, in puzzle design, is the implicit agreement that a given shape carries a given meaning and that a verb is available here. It also fixes which move counts as an action and which does not. Most strong puzzle games establish their conventions silently, by placing the player in a situation where the convention is the only coherent reading of what just happened, then widening the context as fluency grows. In The Witness (Thekla, 2016), walking into a panel activates it and walking into a wall does not; the game never says so, but the opening corridor contains exactly one active panel and no wall a player would plausibly push, so the convention establishes itself in the first half-minute.
The discipline this demands is absolute consistency, because a broken convention is worse than no convention. If a bright outline means this object is you in the first level, it must mean that in every level; if unusual capitals signal a hidden message once, the solver will hunt for hidden messages every time they appear. That is the same trust contract that governs Cluing & Fairness: the calibration opening of a puzzle is teaching the solver which signals to read closely, and a game that quietly changes what a signal means destroys the reading the opening built. Convention establishment is fairness applied to the rules themselves rather than to a single clue.
A practical corollary is that legibility and convention are not separable tasks. A convention cannot be built for a symbol the player cannot reliably tell apart from other symbols, so icon design amounts to convention design. The Witness keeps its handful of rule symbols visually distinct because the player meets each one first in isolation and must form a stable reading before it appears in combination.
Conventions can be inherited, borrowed, or built from nothing
Where a convention comes from changes how much teaching the opening has to do. A sokoban-style game can start layering novelty immediately, because four decades of push-the-box games have already handed the player a complete base vocabulary; treat genre familiarity as a design resource. The risk is that an inherited convention may be wrong for the new game and strand the player who arrives with the wrong reference frame, which is the failure mode of a puzzle game that quietly assumes platformer physics.
The opposite extreme is the convention with no precedent at all. Baba Is You (Hempuli, 2019) teaches its entire premise inside its first level: the words BABA IS YOU and FLAG IS WIN sit on the board as pushable tiles, the player walks Baba into the flag and wins, and then, brushing a word block, watches a rule change. No prior game supplies a reference point, so the convention has to be built entirely from within the game’s own first moments. The design solution is to make the consequence immediately visible and the rule spatially legible: the rule is literally written on the floor, and understanding the tutorial means understanding the game. Between these poles sits the borrowed convention, such as Blue Prince (Dogubomb, 2025) lifting the draft-three-pick-one menu from tabletop deckbuilders. The move is the borrowing, but because not every player arrives with that fluency, the game places the draft at the first available action, which settles the convention unambiguously whether or not the player has seen it before.
The reveal has a curve, and the curve is the curriculum
A game does not teach one rule; it teaches a sequence, and the curriculum is the trajectory along which it introduces mechanics, complicates them, then finally folds them together. The quality of a puzzle game often reduces to whether this curve is well-paced. Too steep and the player is lost; too shallow and they are bored; a badly-timed combination leaves them confused without being stimulated. The curve is also where the payoff lives, because the deepest moments come not from a new rule but from an old rule applied in a configuration that contradicts the player’s internalised shortcut: the rule did not change, their model of it was too narrow. That collision is the engine of The Aha Moment, and it is why a teaching curve and a satisfaction curve are the same curve.
The shapes this curve can take vary with the game’s structure, and they trade control of the player’s knowledge for freedom of movement.
| Curve shape | Example | How it teaches | What it trades |
|---|---|---|---|
| Strict linear | Portal’s test chambers | One lesson per room, fixed order; the designer knows exactly what the player knows | Maximum control, minimum exploration |
| Self-teaching archipelago | The Witness’s zones | Each zone teaches one symbolic language from first principles; any order | Freedom, at the cost of guaranteed prior knowledge |
| Expanding vocabulary | Baba Is You | Each world adds one or two words, then combines them with all prior ones | Surprise, at risk of combinatorial overload |
| Knowledge as the mechanic | Outer Wilds | Every tool is available from minute one; only the player’s understanding evolves | The whole curve lives in the player’s head, none in the systems |
The self-teaching archipelago is the costly one to build, because without a fixed sequence the designer cannot guarantee what a player brings to any given zone, so each zone has to teach itself from scratch. The strict linear path is the cheapest to control and the easiest to over-pad, which is where the mid-game collapse lurks. Once the introductory phase ends, the designer can either keep adding mechanics and overwhelm a player who has not yet internalised the current set, or extend existing mechanics into ever more elaborate configurations until the sequence feels like busywork. The reliable escape is to change the mode of engagement rather than add a rule, shifting where the puzzle-solving happens, as The Witness does when the solution path leaves the panel and is traced in the surrounding world. How that curve maps onto a whole event, rather than a single mechanic, is the work of Clue Sequencing & Chains: the opening sequence is the tutorial, and the difficulty curve is its lesson plan.
When the tutorial is absent on purpose
A distinct tradition withholds the lesson entirely and treats discovering the rule as the first puzzle. Tunic (2022) hides how to open its own manual; La-Mulana never explains that the whip ruins certain tiles; Animal Well keeps verbs hidden for hours, inferrable only from environmental anomalies. Here the player is not being taught but tested: the convention is that the world contains tools you do not know you have, and the task is to notice a discrepancy and experiment until a new verb emerges. This is the same instinct as the doubled-indirect clue in A Taxonomy of Puzzles, where the solver must first work out what kind of thing they are looking for. The line to police is the one between a withheld lesson and an unteachable one: a discovery puzzle is fair only when the world supplies a noticeable anomaly and the experiment that resolves it is reachable.
The hunt teaches in-band, through the difficulty curve
Move from a single-player puzzle game to a puzzle hunt and the teaching problem reappears in a sharper form, because a hunt rarely teaches its conventions anywhere but in the puzzles themselves. The defining convention is that a puzzle will not declare itself, so working out what kind of puzzle it is forms part of the puzzle. So the first literacy a newcomer must build is pattern-recognition: the reflex that a list of clues probably wants solving and a wall of pictures probably wants identifying, tempered by the knowledge that hunts deliberately subvert these expectations. The skill that separates hunts from ordinary puzzles, and the hardest thing for first-timers to learn, is extraction, which pushes a solved puzzle a step further to pull out a final word. Most often this means indexing, where you take the Nth letter of an answer. New solvers are drilled on the reflex to check, every time they open a puzzle, whether the first letters spell something. This is Extraction, Indexing & Meta-Puzzles seen from the newcomer’s side: the literacy a hunt must teach before its metas become solvable.
The teaching is implicit by design, and that is why a hunt’s pedagogy is progressive disclosure through the difficulty curve. Early puzzles are easy and also expose one mechanic cleanly, so the solver accumulates the extraction, indexing, and flavour-reading vocabulary they will need when later puzzles combine and disguise those same moves. Flavour text doubles as the sanctioned hint channel here, the surface where a mention of transmissions hints at Morse or a mention of blindness hints at Braille, so the prose that looks decorative is the onramp’s teaching surface. Because the implicit curriculum has a steep first step, the community has built explicit scaffolding around it. Gentle archived events such as Puzzled Pint are repeatedly recommended as a first hunt. The genre even teaches its own social conventions, such as using ROT13 to share graduated hints without spoiling, since the cipher is its own inverse. The whole apparatus is the hunt-scale answer to the same question Portal answers in a room: how do you hand a newcomer the rules without a rulebook.
The Witness never prints a rule. Its opening garden teaches the base convention, draw a continuous path from the start dot to the endpoint, through a handful of panels that admit only one interpretation, and only once that is owned does each later panel add a single element to the grammar: a coloured dot, a pair of stars, a tetromino, a sun. The player infers each new rule from the behaviour of correct and incorrect solutions, so the puzzles themselves are the documentation; a solver who has cleared a hundred panels can read symbols that would be opaque to anyone who skipped the sequence. Jonathan Blow has described this as building a language with the player. In that account each puzzle is a sentence that extends what the previous ones established, and the design goal is a string of miniature epiphanies thrown off by the act of learning itself. The late game then weaponises the convention it built. Some puzzles look like standard panels but require perceiving something outside the panel entirely, so that the base reading of the puzzle is on the panel is deliberately broken, and the break itself is the puzzle.
The two failures sit at opposite ends and both are quiet. Convention overload is introducing five new conventions in the first two minutes. Each is graspable on its own, but the load of tracking five at once means the player emerges with approximate models of all of them and solid command of none. At the other end is the convention you assume the player already holds from a genre they have in fact not played; it strands the newcomer with no reference frame and no way to recover. The most dangerous version of both is invisible failure. The misunderstanding looks identical to progress, so the player carries a broken model forward until a puzzle becomes impossible minutes later. By that point the failure is too far from its cause to diagnose. The tell is the same one that betrays an unfair clue, the playtest note that says they should have realised. A convention the player was never given is not a lesson, it is a wall.
Where this fits
The bigger picture. Teaching through play is where a puzzle’s trust contract is first signed. Before a solver can reason about any clue they have to know how the game is read. Which marks are signals? Which actions count? And what does a given symbol mean? The opening sequence answers all of that without a word. Every later mechanism in this family assumes the answer is already in hand. It is the same floor that Cluing & Fairness stands on, seen one level down. Fairness promises that an attentive solver can reach the answer from the materials. Teaching through play is what makes those materials legible enough for attention to have somewhere to land. Get the opening wrong and no amount of fair cluing later can recover a player who never learned to read the game. Get it right and the rules become the player’s own, the only form in which a rule is worth anything.
Next steps. Once the conventions are established, the craft turns to pacing their reveal across a whole event. That work is Clue Sequencing & Chains. It also turns to engineering the collision between an owned rule and a configuration that stretches it, the subject of The Aha Moment. To decide which conventions a given kind of puzzle most needs to teach up front, start from A Taxonomy of Puzzles. In live and mixed-ability play, where the opening calibration puzzle has to teach the group’s conventions before the challenge proper begins, that onboarding is part of designing for Mixed-Ability Groups & Universal Design. And when teaching by demonstration is not enough and a solver is still stuck for reasons no convention will fix, the graceful exit is Hint Systems & Recovery.
Referenced by
- Baba Is You
- Clue Sequencing & Chains
- IF Mechanics & Procedural Rhetoric
- Mixed-Ability Groups & Universal Design
- The Aha Moment
- The Witness
- Time Loop Patterns