Three kinds of "guesses"
Not every guess is the same. Before reaching for probability, sort the situation into one of three buckets:
- False guesses. The deduction exists — you just haven't found it. 1-2 patterns, subset deductions, and mine-counter math frequently bail out positions that look impossible.
- Asymmetric guesses. A region with three cells and one mine is not the same as a region with two cells and one mine. Cells in the larger region are individually safer (1-in-3 vs 1-in-2).
- True 50/50s. Two cells, exactly one mine, no other constraints touching either cell. These exist but are rarer than they look.
Spend ten seconds eliminating the first two before treating a situation as a true coin flip.
The mine-counter math
The mine counter at the top of the board is the most underused signal in minesweeper. Late game, it pins down global probability on every unrevealed cell that isn't constrained by a neighbour.
Suppose: 30 unrevealed cells remain, 8 mines left. Cells next to a satisfied number sit at 0% mine probability. Cells with no number neighbours float at 8/30 ≈ 27%. The boundary-constrained cells inherit local probabilities from their numbers.
Here's the punch: when a boundary number forces a regional probability higher or lower than 27%, the rest of the unconstrained cells shift in the opposite direction. If 3 cells are forced to hold all 8 of the remaining mines (impossible — but try a smaller example), the rest are forced safe. Real positions are softer, but the principle scales: tight constraints leak probability to loose constraints.
Per-region probability
When two regions look equally bad, count cells and mines per region and compare ratios.
Example
Mid-game on intermediate. The unrevealed boundary splits into two disconnected groups:
- Region A: 4 cells, constrained to hold exactly 1 mine. Per-cell probability: 25%.
- Region B: 3 cells, constrained to hold exactly 1 mine. Per-cell probability: 33%.
If you must guess, pick a cell from A. The math is 25% vs 33% — roughly 8 percentage points safer per click. Over a hundred forced-guess situations that adds up to many extra wins.
Some positions have many regions. Compute each ratio quickly and pick the lowest. Don't let the visual closeness of cells bias you — distance on the board doesn't change probability.
The 1-1 boundary 50/50
The classic forced-guess shape. Two unrevealed cells along a wall, with a 1 next to them satisfied by a known mine already counted, and another 1 elsewhere also satisfied. Neither cell has additional constraints. Mine count says exactly one is a mine.
Locally this is 50/50. But globally, the mine counter often rebalances it:
- If the rest of the board has plenty of unrevealed cells with no neighbour constraints, those cells eat some of the residual mine count. The 50/50 shifts to perhaps 40/60 or 60/40 depending on which side has more "room" elsewhere.
- If the rest of the board is nearly fully resolved, the local 50/50 is genuine.
Always check the mine counter before resigning to a true coin flip. The visual shape is often misleading.
Information value — pick the click that teaches you the most
When two cells are roughly equal in safety, prefer the one that gives more information when revealed.
What "more information" means
- A cell deep in an unrevealed region is likely to reveal a 0 (or low number) and trigger a flood-fill. High information yield.
- A cell at the boundary edge usually reveals a single number with no flood-fill. Low information yield.
- A cell adjacent to multiple existing constraints will tighten several deductions at once when its value is known. Medium-to-high information yield.
The trade-off
Sometimes a 30% mine cell with high information yield beats a 25% mine cell with low yield. The 5% extra risk buys you information that may resolve the entire endgame. In races especially, information-rich clicks compound across the cascade.
Conditional probability
When you have a constraint that affects multiple regions, compute conditional probabilities rather than treating regions as independent.
Example
Two regions A and B share a single constraint cell. A has 3 cells, B has 4 cells. The constraint says "exactly one mine across A+B's six exclusive cells plus the one shared cell".
You can't just compute A's per-cell probability as 1/7 (one mine spread across all seven), because the constraint interacts with other constraints on A and B individually. Layer the constraints: first work out which configurations of mine positions are consistent with all constraints; then weight by the number of valid configurations each cell appears as a mine in.
This is the heart of why minesweeper solvers exist — the bookkeeping gets gnarly fast. For human play, the practical version is: don't compute exactly, just remember that shared constraints make the regions less independent than they look, and prefer guesses that involve cells with the fewest outside connections.
The "delay the guess" tactic
Sometimes a guess that's forced right now becomes avoidable two clicks from now. If you have two regions and only one is forced into a guess, resolve the unforced region first — its result may give you the mine-counter information you need to eliminate the original guess.
On expert this is huge. The board is large enough that a forced guess in one corner can be defused by counter-math once another corner is resolved. Don't pull the trigger prematurely.
When the math says guess, guess
Once you've eliminated false guesses, ranked the asymmetric guesses, considered information value, and the math still says "click a 30% cell or lose the board" — click. Hesitation doesn't lower the probability; it only burns time.
Top players accept 5-10% of expert boards as forced losses. That's not a failure of skill — it's the structure of the game. The skill is in not turning a 30% guess into a 45% guess by panicking.
Practice the math on the daily
The daily challenge is a great venue to drill probability calls. The same board is served to every player, so if you encounter a tricky late-game situation, you can come back tomorrow and replay it slowly. Top players on the leaderboard generally average lower guess rates than mid-tier players — not because their boards are different, but because they're seeing deductions others miss.
Read the expert strategy guide for a fuller picture of how probability fits into the broader speedrun framework.