When Minesweeper forces you to guess

Strategy · about 9 minutes to read

Most people who quit Minesweeper quit for the same reason. They play carefully, they clear three-quarters of the board without a single careless click, and then they arrive at a position with two closed cells, one mine left, and nothing to tell them apart. They click, they lose, and they conclude that they must have missed something earlier.

Usually they did not. Minesweeper contains positions where no amount of reasoning produces a safe move, and reaching one of them is not a mistake. But those positions are much rarer than beginners think, because most of what feels like a guess is actually a deduction nobody showed them how to make. This guide covers the deductions first and the guessing second, in that order, because that is the order in which they matter.

A number is a constraint, not a hint

When you open a cell and it shows a number, that number counts the mines in the eight cells touching it — the four orthogonal neighbors plus the four diagonals. Along an edge a cell has only five neighbors, and in a corner only three, which is why edges and corners behave differently from the middle of the board. A blank opened cell is a zero: it says that none of its neighbors is a mine, which is why blanks cascade open into large empty regions.

The useful way to think about this is that each number is a small equation. A 3 sitting next to five closed cells says: exactly three of these five are mines. It says nothing at all about any other cell on the board. Every number you can see is one of these equations, and the board as a whole is a system of them that you solve locally, a few overlapping cells at a time. You never need to look at the whole grid at once. You need to look at two or three numbers that share neighbors.

The two rules that clear most of a board

Almost every safe move in Minesweeper comes from one of two observations, and both are immediate once you see the number as an equation.

1. If a cell's number equals the number of flags already around it, every other closed neighbor is safe. The equation is satisfied. Whatever mines that number was counting, you have already found all of them, so the remaining closed neighbors cannot contain any.
2. If a cell's number equals its count of closed neighbors, all of them are mines. A 3 with exactly three closed neighbors has no room left to distribute anything. Flag all three.

These two rules alone will take you a long way. If you are losing boards early, you are almost certainly not applying them exhaustively — it is very easy to satisfy a number, move on, and forget to look back at the neighbors it just cleared. A good habit after every flag is to re-check every numbered cell touching that flag, because a flag can complete several equations at once.

Subtracting overlapping constraints

This is the single most valuable idea in the game, and it is the one that separates players who guess often from players who almost never do.

Two numbered cells that sit near each other usually share some of their closed neighbors. When they do, you can subtract one equation from the other, and the difference tells you something about the cells they do not share. Stated plainly: if constraint A covers a set of cells that is entirely contained inside the set covered by constraint B, then B's count minus A's count is the number of mines in the cells that belong to B alone.

A concrete case. Suppose two closed cells, call them X and Y, are the only closed neighbors of a cell showing 1. A second cell nearby shows 2, and its closed neighbors are X, Y and Z. The 1 tells you there is exactly one mine among X and Y. The 2 tells you there are exactly two mines among X, Y and Z. Subtract: two minus one is one, and the only cell in the second set that is not in the first set is Z. So Z is a mine, with certainty, even though you have no idea which of X or Y is the other one.

The same subtraction runs the other way. If the second cell showed 1 instead of 2, then there is exactly one mine among X, Y and Z, and you already know that one mine lives in X or Y. One minus one is zero, so Z is safe. You still do not know anything about X and Y individually, and you do not need to — Z is a free click, and opening it will very often produce the number that resolves X and Y as well.

Two things make this practical. First, you are looking for containment: a small set of closed cells fully inside a larger one. That happens constantly along the frontier between opened and closed territory, especially where two numbers are diagonal neighbors or two apart in a row. Second, you do not have to resolve everything. Finding one certain mine or one certain safe cell is enough, because each new piece of information reshapes the constraints around it.

Once subtraction is a reflex, the frontier stops looking like a wall of numbers and starts looking like a chain of overlapping sets, most of which cancel against each other.

The named patterns, and why they exist

Experienced players talk about the 1-1, the 1-2 and the 1-2-1 as if they were shapes. They are not shapes worth memorizing on their own; they are the subtraction above, applied to the most common geometry on the board — a straight row of numbers with a straight row of closed cells beside it. If you understand where they come from you will also recognize them when they appear rotated, mirrored, or slightly deformed, which is most of the time.

The 1-1 along a wall

Picture a horizontal run of opened cells with a row of closed cells directly below them, and imagine the board's left edge cutting the run off. The leftmost two opened cells both show 1. Because the wall is there, the first 1 has only two closed neighbors: the cell below it and the cell below-right. The second 1 has three: those same two, plus one more to the right.

That is exactly the containment case. One mine in the first pair, one mine in a set that includes the first pair, so the extra cell on the right is safe. The wall is doing the real work here — without it the first 1 would have a third closed neighbor on its left, the containment would break, and the deduction would evaporate. That is why this pattern is described as happening along a wall, and why the same three cells in open field tell you nothing.

The 1-2

Same corridor, same wall, but the second number is a 2. Now the larger set needs two mines and the contained pair supplies one of them, so the extra cell on the far side of the 2 must be the second mine. Flag it. The rule of thumb players repeat — that a 1-2 flags the cell beyond the 2 — is just this subtraction with the arithmetic already done.

1-2-1 and 1-2-2-1

These appear when a corridor is closed off at both ends, so the closed cells are bounded and the numbers along the corridor constrain each other from both directions.

In a 1-2-1, three closed cells sit below the three numbers. The left 1 sees the first two cells; the 2 sees all three; the right 1 sees the last two. Suppose the middle cell were a mine. Then the left 1 is satisfied, so the first cell is safe; the right 1 is satisfied too, so the third cell is safe; and the corridor holds exactly one mine. But the 2 demands two. The assumption fails, so the middle cell is safe, and each 1 must get its mine from the outside cell it touches. Mines at the ends, a safe click in the middle.

The 1-2-2-1 works the same way with four closed cells and comes out inverted: the two mines sit under the 2s and the cells under the 1s are safe. Work it through once with a pencil and you will not need to memorize it — the constraint arithmetic gives you the answer faster than recall does, and it keeps working when the corridor is vertical or diagonal instead.

Positions where there is genuinely no safe move

Now the honest part. Some boards end in a position that cannot be solved, and this is a property of the position, not a gap in your reasoning.

The classic case is the last two cells. Suppose you have flagged every mine but one, and two closed cells remain. Both are neighbors of the same numbered cell, and that number is the only constraint touching either of them. Every equation on the board is consistent with the mine being in the first cell, and every equation is equally consistent with it being in the second. There is no chain of subtractions that separates them, because the two cells are interchangeable as far as the constraint system is concerned.

This can be established rigorously. A solver can enumerate every arrangement of mines consistent with all the visible numbers and the mine counter, and if a cell contains a mine in some of those arrangements and not in others, no logic can decide it. When every remaining closed cell is like that, the correct move is to guess. It is not a failure state and it is not a bug; it is the game. What you can control is how often you arrive there and which cell you click when you do.

Choosing the least bad guess

When you must guess, guess deliberately. The goal is the cell with the lowest chance of holding a mine, and there are three things worth doing before you click.

Work out the baseline. Divide the mines still unaccounted for by the number of closed cells still on the board. If the counter says twelve and forty cells are closed, the average closed cell is about a thirty percent risk. That number is not the answer, but it is the yardstick: any candidate you can show is riskier than the baseline should be avoided, and a cell in a region far from every number carries roughly the baseline risk.

Read the local constraints. A closed cell touching a 1 that has three closed neighbors is about a one-in-three risk. The same cell touching a 3 with four closed neighbors is a three-in-four risk and should be nowhere near your shortlist. Comparing local ratios like this is rough, because overlapping constraints can shift the true odds in ways simple division misses, but it reliably separates the terrible guesses from the tolerable ones.

Prefer corners and edges. This is a heuristic rather than a theorem, and it is worth being clear about that. A corner cell has three neighbors instead of eight, so whatever it turns out to be constrains a much smaller region, and the number it reveals is far easier to resolve completely. Edge guesses have a similar advantage. On top of that, a guess into a large closed region can cascade open a big area and hand you a whole new frontier of solvable numbers, whereas a guess wedged into a two-cell pocket buys you almost nothing even when it survives. There are positions where a central guess is measurably better; the point is that corners and edges are a good default when the probabilities look similar.

Playing in an order that avoids coin flips

Much of the guessing that beginners endure at the end of a board was created earlier, by the order in which they opened it. A few habits help.

Open space first. Large blank cascades are pure profit: they cost nothing, they reveal many numbers at once, and every number they reveal is a new constraint you can subtract against. Chasing a half-resolved cluster while the middle of the board is untouched is the opposite trade.

Resolve the middle before the edges are all that is left. Central regions are constrained from every direction, so they are usually the easiest to finish; edges and corners have fewer neighbors and therefore fewer constraints, which is exactly why an unresolved corner is such a common home for a genuine fifty-fifty. Leaving the constrained part for later and the unconstrained part for last is backwards.

Do not leave isolated pockets. A closed cell surrounded by finished numbers has nothing left to tell you about it, and if it survives to the endgame it becomes a forced guess. Whenever you satisfy a number, check whether you have just cut a cell off from every remaining source of information, and if so, deal with it while there is still context around it.

The first click

In our version of the game the first click is always safe, because the mines are placed after you click rather than before. That is why an opening click often blows open a large region — the generator can position mines around your choice. It is a common convention in Minesweeper implementations and a sensible one, since losing on move one is not an interesting outcome, but it is a convention rather than a rule of the game. Other versions guarantee only that the first click is not a mine, and some older ones guarantee nothing at all, so it is worth checking rather than assuming.

The rest of the board is generated at that moment and does not change afterward. Nothing you do moves a mine, and nothing rewards you for clicking quickly. Minesweeper punishes exactly one thing: clicking before you have finished reading the numbers you already have.