Legible Failure — Making the Dead End Readable in Puzzles

Introduction

I want to reopen the question of what 'failure' means in a puzzle game. In an action game, failure is death: a life lost, a fade to black, a retry screen. It is loud and unmistakable. In a thinking puzzle, failure is quieter and far more ambiguous. Most often it arrives not as death but as the dead end. The board is still there, the character still alive, input still accepted, and yet the position can no longer be solved. Sokoban, released by Thinking Rabbit for the PC-8801 in December 1982, carried this strange 'alive but unsolvable' failure from the genre's very start.

The trouble with the dead end is that it is not announced the instant it happens. Death is legible. A dead end is often noticed only when it is far too late. The player realizes, minutes later, that the fatal move was pushing a crate against a wall three turns ago. What I want to examine is not whether failure should be punished or forgiven; I treated that in The Ethics of Undo. The question here is earlier: can the player read that they have failed at all? This is a problem of legibility.

The Sokoban Dead End — Irreversibility Born of the Push

The Sokoban dead end is born of a single verb. All you can do is push a crate; you can never pull it, and only one at a time. That constraint stamps irreversibility onto the board. Push a crate into a wall corner and it can never move again. If that square is not its goal, the puzzle becomes unsolvable in that instant. As I wrote in the lineage of subtractive design, narrowing the verb deepens structure rather than impoverishing it. But the same subtraction also produces the dead end as a byproduct.

What is striking is that early Sokoban had no undo. A player who noticed the dead end could only restart the level from scratch. The Sokoban dead end was thus irreversibility itself. This looks like punishment, but as design it was a device that forced thought. Each turn it asks: is this push recoverable? The habit of reading before pushing arises only because the move cannot be taken back. The dead end was not a flaw but a pressure that drew out caution.

Yet a legibility problem hides here. An experienced player senses the dead end the instant a crate enters a corner. A beginner cannot see that the board is already lost. They keep pushing for minutes, still believing a solution exists. Sokoban is called harsh on newcomers less because of raw difficulty than because of this invisibility of the dead end. The failure is right there, but it cannot be read. This is the central problem later puzzles keep wrestling with.

The Invisible Dead End — The Worst Failure Is One You Cannot Notice

The most damaging dead end of all is the one the player cannot notice. Stephen's Sausage Roll, Stephen Lavelle's April 2016 work, can be read as a game that takes this problem head-on. Behind the simple goal of grilling all six faces of a sausage, the board often slips quietly into a state where you can still move but can no longer solve. The angle of a fork, one roll of a sausage, turns the position irreversible without your noticing. The player keeps searching an unsolvable board, certain it can be solved.

2015's Snakebird carries the same kind of difficulty. The lengthening body blocks its own retreat, and countless layouts become dead ends the moment you grab a fruit. Here too the failure is silent: the snake does not die, the screen gives no warning. The vertical wall I discussed in shaping the learning curve was a wall where no solution is found. A dead end's wall differs in kind: you keep climbing a wall whose solution no longer exists, without knowing it.

From a designer's view, this invisible dead end produces a double loss. First, the player melts time into an unsolvable position and accumulates futility. Second, and worse, they can no longer tell whether their thinking is wrong or the board is already dead. That inability betrays the puzzle's promise: think and you will solve. This is why hard-edged games like Stephen's Sausage Roll and Snakebird make restarting instant. To reset at no cost the moment you sense the dead end is the minimum honesty a puzzle owes its player.

Designing Out the Dead End — The Choice of The Witness and COCOON

Instead of agonizing over the legibility of the dead end, a designer can simply refuse to create one. Jonathan Blow's The Witness, from 2016, is the prime example. A line puzzle can be redrawn any number of times. A wrong line is never committed; failure stays a provisional, undoable state. The only thing the player loses is time; the board itself never turns irreversible. Here failure is a 'mistake,' not a 'dead end.' A mistake can always be redrawn. This focuses attention on observation and inference.

Geometric Interactive's COCOON, from September 2023, reaches the same point from another direction. This game of carrying world-containing orbs is built so the player almost never falls into a dead end. Designer Jeppe Carlsen arranges the board so a wrong move never loses anything permanently. Despite the nesting complexity that belongs to the lineage of meta-puzzles, COCOON produces almost no irreversible failure. By removing the dead end, it lets players experiment in safety. Where failure is no punishment, exploration itself becomes play.

Yet removing the dead end entirely has a cost. The tension Sokoban's irreversibility forced, reading before you push, thins out in a redrawable world. As the cost of failure approaches zero, players try before they think. This is a virtue that lowers the barrier and, at once, a danger that lowers the density of thought. The Witness and COCOON hold together because they replace the lost bite with other difficulty: observational resolution, spatial reasoning. Designing out the dead end is not flight into easiness but a relocation of where difficulty lives.

Tools That Make Failure Readable — Undo, Restart, and Telegraphing

There are also tools that keep the dead end but raise its legibility. The most widespread is undo. Baba Is You, Arvi Teikari's March 2019 game, lets you rewrite the rules themselves, so players frequently destroy a needed element and dead-end. Its unlimited undo is the safety device that rewinds that instantly. Undo erases failure and, at the same time, is a teaching device: it lets you trace back which single move was fatal. Each step back, the player follows causality in reverse and reads the cause of the dead end.

Patrick Traynor's Patrick's Parabox, from March 2022, makes the shape of the dead end counterintuitive through boxes nested inside boxes. Here too unlimited undo and instant restart are the precondition for reading a complex dead end. The point is that undo and restart are not mere rescue but devices for making failure visible. Because it can be undone, the player safely peers over the edge of the dead end and learns what was irreversible. The experiment that was terrifying in unforgiving Sokoban becomes a means of learning in a world that can be undone.

The other tool is telegraphing: visual and structural cues embedded in the board that foreshadow a dead end before it happens. A good puzzle gives the player a small unease just before the fatal move. The sense that 'pushing this crate cannot be undone' is conveyed not by an explicit warning but by the shape of the board itself. The works gathered in The Design of Undo have, in the end, competed over this one point: how to make failure readable. Undo is after-the-fact legibility; telegraphing is legibility beforehand. How you blend the two decides the feel of the puzzle.

Closing

Whether to punish or forgive failure is often discussed at the entrance to design. But there is a more fundamental question before it: can the player read that they have failed? Sokoban left the irreversible dead end; Stephen's Sausage Roll and Snakebird took on the difficulty of the invisible dead end; The Witness and COCOON erased the dead end from design itself. Baba Is You and Patrick's Parabox used undo as a device of visibility and turned failure into learning. Each work chooses its own position on the single axis of how legible failure is.

If I made a puzzle next, I would begin from this question: in this game, can the player notice they have dead-ended? If not, is that hard-edged difficulty or merely unkindness? If yes, do I show it after the fact through undo, or foreshadow it beforehand through the shape of the board? You can erase failure entirely, but then you must decide at once where the bite goes instead. Failure is not a flaw to hide but a subject of design. I leave the reader a question: the last time you dead-ended in a puzzle, when and how did you notice? That manner of noticing is the very place a designer should carve most carefully.