The Undo-Button Problem: When to Allow It

Look, the undo button is the single most contested design decision in puzzle games and I have been thinking about it for years. Some games allow unlimited undo. Some allow one undo per move. Some allow none at all. And some allow a fixed number of undos per puzzle. Each choice changes the puzzles fundamentally, and I want to think through why.

This matters because the wrong undo decision can ruin an otherwise-good puzzle game. I have abandoned several puzzles on this site that I would have rated higher if the undo system had been thought through. And I have over-rated a few that turned out to have undo systems that mask design flaws elsewhere. The undo button is doing more work than it gets credit for.

What undo actually does

At its most literal, undo reverts a players last move. The board state goes back to what it was before they acted. They can then make a different move.

But what undo actually does is change the relationship between thinking and acting. With unlimited undo, you can try moves and reverse them with no cost. Trial-and-error becomes a valid strategy. Without undo, you have to think through the consequences of each move before committing. Trial-and-error is punished by progress loss.

This is a real design choice. The two modes produce different play patterns. The puzzles that work in one mode often do not work in the other.

When unlimited undo is right

For long, complex puzzles where the solution space is large but each move is reversible in principle, unlimited undo is the right call. Sudoku is the canonical example. You can place a digit, realise it conflicts with a constraint several rows away, and undo to try a different digit. The puzzle is the same with or without undo; undo just removes the manual restart penalty for backtracking.

Number 2048 on this site allows one undo per move, which is a compromise between unlimited and none. The compromise makes sense because 2048 has random elements (the spawn after each move). Unlimited undo would let you re-roll the spawn until you get the tile you want, which breaks the puzzle. One undo per move lets you take back a clear mistake without breaking the random element.

For traditional logic puzzles with no random elements (Sudoku, nonograms, lights-out, picross), unlimited undo is appropriate. The puzzles structure is what it is, and undo just removes incidental friction.

When undo is wrong

For puzzles where the difficulty is constructed around commitment, undo wrecks the design.

Spider Solitaire is the example I want to push back on. The four-suit variant on this site is rated three stars from me, partly because the difficulty hinges on the cost of bad moves. With unlimited undo, the variant becomes solvable through patience alone. With no undo, the variant becomes a luck-check against the deal. Either extreme ruins what should be a strategic decision game.

The compromise (limited-undo, say five undos per game) would respect the design intent. The browser implementation gives unlimited, which makes the game less interesting than it should be. I have rated it accordingly.

For pure-execution puzzles like sliding-tile games, undo is also wrong. The 15 Puzzle on this site is reviewed at 2.5 stars partly because the algorithm to solve it is well-known and there is no decision to make. With unlimited undo, the puzzle becomes execute the algorithm with infinite retries, which removes the small skill there is.

The hint-versus-undo distinction

A related design choice that often gets confused with undo is the hint system. Hints reveal information about the solution. Undo reverses moves. They are not the same thing, even though both modes reduce difficulty.

Hint systems are usually better than undo systems for puzzles where the difficulty comes from finding the solution rather than executing it. A Sudoku player who needs help is better served by a hint that reveals one cell than by being told to undo. Undo does not help if you do not know what to do differently. A hint actually gives you information.

Best puzzle implementations on this site use limited hints rather than unlimited undo. Tangram Master allows three hints per puzzle. Nonogram puzzles in our catalogue typically allow one mistake-check before the puzzle becomes unsolvable in pure-logic mode. Both are stricter than unlimited undo and more respectful of the puzzles design.

The mobile context

I want to note the mobile-specific dimension. Touch input is less precise than mouse-and-keyboard. A misclick on a touchscreen is more common than a misclick with a mouse. The undo button is, in part, a misclick-recovery mechanism on mobile.

For mobile puzzle games, some form of undo is therefore necessary even when the desktop version would not need it. The question is whether the undo is bounded (one undo per move, accommodates misclicks) or unbounded (unlimited, accommodates strategic exploration).

I think bounded undo is the right answer for most mobile puzzles. It accommodates the input-imprecision reality without breaking the puzzles design intent. Most browser puzzle developers ship unbounded undo because that is the path of least resistance, not because it serves the puzzle well.

The competitive dimension

For puzzle games with leaderboards or daily challenges, undo decisions matter for fairness across players. If your version of the puzzle has unlimited undo and another players version has none, their score and your score are not comparable.

This is a real problem for cross-player ranking systems. The puzzle community has converged on a few standards. Speedrun categories often disallow undo entirely. Casual leaderboards usually allow unbounded undo because that matches the casual player session. Daily challenges often allow one undo per move with the rest disabled.

Standards are sensible but not universal. When you see a leaderboard, check the undo rules. The top scores often reflect undo policies you might not share.

What this tells you about a puzzle game

Undo system is a tell about how the developer thinks about their puzzle. A puzzle with unlimited undo is one where the developer expects players to explore solutions through trial-and-error. A puzzle with no undo is one where the developer expects players to commit and pay for mistakes. A puzzle with bounded undo is one where the developer is trying to balance both.

None of these is wrong by itself. The wrong choice is one that doesnt match the puzzles actual structure. A logic puzzle with no undo is frustrating for the wrong reasons. An execution puzzle with unlimited undo is empty for the wrong reasons. A puzzle with bounded undo that is also luck-heavy is the worst of all worlds because the bounded undo runs out before luck can balance.

I check the undo system in the first few minutes of every puzzle game review. The system tells me how to evaluate the rest of the games design.

Best browser puzzle games respect the undo question and answer it thoughtfully. Minesweeper Classic has no undo, which is correct for a logic puzzle where the deductive process is the whole game. Sudoku Master has unlimited undo, which is correct for a constraint-satisfaction puzzle where backtracking is the standard solving technique. Number 2048 has one-undo-per-move, which is the right answer for a puzzle with random elements.

When you read puzzle reviews on this site, the undo system is usually mentioned because the system matters more than it gets credit for. Look for it.