The Verb of Pushing — What Sokoban Invented by Removing "Pull"

Introduction

The moment you push a Sokoban crate flush against a wall, your hand stops. That crate can never be pulled back. Sokoban, released in 1982 by Hiroyuki Imabayashi for the PC-8801 and FM-7, has only two verbs—walk and push—yet that one-way asymmetry alone has kept it difficult for over four decades. I have spent recent months taking thinking-puzzle verbs apart one at a time, and none is as unassuming, or as deeply rooted in design's skeleton, as pushing.

This essay rereads why the push verb has survived so long, through the lens of deadlocks. How did Sokoban's constraint—no pulling—pass down into three utterly different descendants: Sokobond's molecular bonding, Baba Is You's rule-rewriting, and Patrick's Parabox's nested recursion? And where would I place this asymmetry if I built a pushing puzzle next? Read it as part of the Single-Verb Design series, an attempt to take pushing itself apart.

A Game Built on Two Verbs — What Sokoban Cut Away

A Sokoban player can only walk and push the single crate directly ahead—not through a wall, not through another crate, and, crucially, never backward. That last constraint is the design's core. Many early puzzle games assumed you could simply retry after a mistake; Sokoban instead changes the board irreversibly with every move, forcing you to check, before pushing, how that single move reshapes the whole board and whether a dead end waits beyond it.

This was no accident. Had Imabayashi allowed pulling, a failed push could be undone on the spot, reducing the puzzle to brute-force labor. Push-only, no-undo is what introduced the concept of a deadlock into the state space—a configuration you can back yourself into with no path to the solution—and knowing that possibility exists is what forces you to read ahead. Within the lineage of subtractive design, Sokoban is among the earliest and purest examples.

What's striking is how invisible this asymmetry is. The screen shows only crates, a person, and walls; the rule that you cannot pull is written nowhere on screen. You learn its weight only from the manual, or by backing yourself into a deadlock once. Pushing looks, as an affordance, barely different from walking, yet its design role is entirely different—and that gap is part of why Sokoban is still studied today.

The Invention of the Deadlock — What Irreversibility Demands

A deadlock is frightening because it doesn't announce itself the instant it happens. Push a crate into a corner and the board looks unbroken; it's simply that no path to the goal remains from that configuration. You learn a deadlock only after noticing it, by restarting the whole level. Building that invisible kind of failure into the design is what turns Sokoban from mere transport labor into a thinking puzzle. I discussed how to make failure legible in Legible Failure; pushing puzzles, by contrast, often build difficulty by hiding failure instead.

This irreversibility generates difficulty differently from combinatorial explosion. A walking-only maze offers choices equal to its paths, but in a pushing puzzle the arrangement of crates itself constitutes state, branching with every move—and most branches lead straight to a deadlock, leaving only a sliver actually explorable. Good pushing-puzzle designers arrange the board so most of those branches read as obviously bad moves early on; letting players push freely and only later discover the damage is unkind, while placing cues that give some foresight the instant you push is the craft.

That's why pushing puzzles trace a distinctive learning curve. The first levels are simple enough that a deadlock cannot occur, quietly instilling the premise that a push cannot be undone without ever making you fail. Only once you're comfortable do small deadlocks appear, and only then does foresight become a required skill. Skip that staging and players hit dead ends for no clear reason, learn nothing but unfairness, and quit. The difficulty of pushing lies not in the verb's complexity but in how you teach the weight each move carries.

Three Heirs — Keeping the Push Grammar, Adding Exactly One Thing

At least three works inherit Sokoban's push-only, no-pull grammar wholesale while producing entirely different difficulty. Sokobond, released in 2013 by Alan Hazelden and Harry Lee, swaps crates for atoms. Push two atoms adjacent and they bond, after which the resulting molecule moves as one unit; irreversibility shifts from the crate's position to the molecule's shape. Early levels keep bonds permanent, but later ones add tools to remove or rearrange them, letting chemistry wear a convincing skin over a simple push.

Baba Is You, released in 2019 by Arvi "Hempuli" Teikari, is stranger still: what you push are not cargo but the word blocks that compose the rules themselves. Push blocks like BABA IS YOU or WALL IS STOP into a new arrangement and the game's rules rewrite on the spot. The push action never changes, yet what you can push next expands or shrinks as a consequence—not just board state but the grammar governing the board becomes the object of pushing. I touched on this turn in Inside Arvi Teikari's Philosophy; it's arguably one endpoint of the push verb.

Patrick's Parabox, released on Steam in March 2022 by Patrick Traynor (winner of IndieCade 2019's Developers Choice Award and IGF 2020's Excellence in Design), nests the pushed box itself. Push open one box and another board unfolds inside it—sometimes containing the very player pushing it. What you push chains from box to box across dimensions. That recursive grammar is the subject of Recursion as a Puzzle Grammar, yet at its root sits the same minimal grammar as Sokoban: push, and never pull.

What these three share is that none moves the verb away from pushing. Sokobond changes the object to atoms, Baba Is You to rules, Patrick's Parabox to space itself, but the action a player actually performs stays "push whatever is adjacent." Changing what a verb acts on, rather than adding verbs, is—I'd argue—the most economical way to grow difficulty within the Single-Verb Design theme.

Pushing vs. Rolling — Neighboring Verbs, Different Grammars

As I wrote in The Verb of Rolling, rolling puzzles deepen the state space by adding orientation to position. Pushing looks similar but works differently. A rolling object carries hidden posture as state, yet its position itself is often recoverable—roll back onto the same tile and position, at least, returns. A pushed object's position is everything, and irreversibility is built into that change from the start. Rolling creates difficulty by adding state; pushing creates it by removing reversibility. Two visually similar verbs reach difficulty by opposite means.

The difference shows in how failure feels. Deadlocked in a rolling puzzle, you usually sense you got the orientation wrong and expect to recover from the same spot. Deadlocked in a pushing puzzle, you must face the plain fact that you cannot go back. The former is a technical failure; the latter is closer to a failure of judgment. Pushing feels unusually "heavy" among thinking-puzzle verbs precisely because that irreversibility of decision never lets go.

Closing

Pushing looks barely different from walking, yet a single constraint—no pulling—produced the concept of a deadlock and has kept the verb at the skeleton of thinking puzzles for over four decades. Sokobond swaps the object for atoms, Baba Is You for rules, Patrick's Parabox for space itself, but no one has let go of the minimal grammar: push, never pull. Changing the object rather than adding verbs is among the longest-lived moves in subtractive design's lineage.

If I built a pushing puzzle next, I would start with levels that instill the weight of this verb without ever letting the player experience a deadlock, then decide what to substitute as the object of pushing—atoms, words, space, or something I haven't thought of yet. So I leave you a question: if you could grant the thing you push exactly one property that can never be undone, what would you make irreversible?