Sudoku vs. Minesweeper: information entropy in two "logic games"

Playing this site's Zen Garden (sudoku), I never "guess" — every number I place is deduced. But playing Mine Sweep on expert, 15–20% of games require a guess. Both called "logic games" but the underlying structures are different. This piece explains why, from an information entropy perspective.

What's a "complete information game"

Game theory divides games into:

Complete information games: all relevant info is on the table; players can deduce the unique solution. Examples: sudoku, Go (within the board), card solitaire (after the opening).

Incomplete information games: information exists that players can't see; probability is required. Examples: poker (opponent cards unknown), Scrabble (next letter tile unknown), minesweeper (mine positions unknown).

Sudoku and minesweeper differ exactly here.

Sudoku is complete information

Given a valid sudoku puzzle, the visible filled numbers plus the constraints "9x9 board + no repeats in row/column/box" already uniquely determine every empty cell. Theoretically, you could deduce the whole solution from the opening state.

Meaning: sudoku never requires guessing. Any quality sudoku puzzle (including all generated by Zen Garden) has a unique solution. Your job isn't "find an answer," it's "deduce the answer."

The pleasure of this game type:

• Knowing the answer exists, you just haven't found it
• Every step is deterministic; no luck
• Only one difficulty axis: making deduction chains longer (from "obvious next step" to "needs 5 steps of inference")

Minesweeper isn't complete information

This is counterintuitive. Minesweeper's "number hints" look very informative, but actually —

Mine placement is random, and that randomness is hidden from you. When number hints can't logically determine a cell, you fall back on "local probability."

Example: a "3" tile has 6 unrevealed neighbors; 1 mine already flagged. So 2 mines remain among 5 cells. If those 5 cells are logically equivalent (no other number hint disambiguates), you must guess — 40% mine probability.

This happens about 15–20% of the time on expert (18x14, 70 mines). So even perfect play on expert has only ~80% theoretical completion.

(Note: professional minesweeper players developed "global probability" algorithms — going beyond local hints to factor in remaining mine count and unrevealed cell count. This pushes completion to 90%+ but never 100%.)

Sudoku's difficulty is "deduction chain length." Minesweeper's difficulty is "probability of forced guess." Not the same thing.

From an information-entropy view

Information entropy (Shannon 1948) measures uncertainty. Rough estimates —

9x9 sudoku opening (30 filled, 51 empty): each empty cell has 1–9 (9 possibilities), so apparent entropy is high. But because of the "unique solution" constraint, each cell's effective entropy is 0 — it has exactly one correct value.

Minesweeper expert opening: you see only numbers. If you randomly click an unrevealed cell, mine probability is 70/252 ≈ 28%. This is genuine probabilistic uncertainty. Even a mathematician can't deduce "this cell has no mine" from opening info.

That's why minesweeper's "opening strategies" (corner-first / center-first) are actually optimizations over random reveal probability, not logical deductions.

Effects on player psychology

Coworkers playing these games reflect the theory:

Sudoku players, when failing (wrong number, can't finish), feel frustrated but accepting: "I deduced wrong." They come back the next day, fresh logic.

Minesweeper players, when failing (mine hit), if it was "a guess," feel angry: "There was nothing I could do, the game wasn't solvable." This anger comes from unlearnable failure — next time you face the same configuration, you'll still have to guess.

This difference makes sudoku more psychologically durable. Many people play sudoku for 20 years because every failure is "I can get better." Minesweeper's hardcore community is smaller because probabilistic failure burns most players out.

But minesweeper has its advantages

Calling minesweeper "the loser" is unfair. It has two strengths sudoku doesn't:

1. Time pressure. Minesweeper expert demands sub-100-second completion for top players. You're not just "deducing right" — you're "deducing fast." This speed dimension makes minesweeper a competitive game with world records, tournaments, professional players. Sudoku has speed variants too, but slower (top players take 5 minutes per puzzle); the skill isn't as "action-y."

2. Shorter sessions. A hard sudoku takes 15–30 minutes. An expert minesweeper takes 3–5 minutes. Minesweeper fits fragmented time; sudoku needs focused stretches.

The two serve different psychological needs; not a "better/worse" question.

My personal play choice

Under stress, needing "absolute control": sudoku. Every step is logic, no luck. That certainty is medicine for anxiety.

Scattered attention, wanting "short bursts": minesweeper. 3 minutes per round, dying occasionally is fine. That "probability tolerance" lets the brain rest.

Blanked out, no brain: neither. Open 2048 or Snake.

Lesson for designers

Building Zen Garden, I made a key decision: guarantee unique solutions for every puzzle. Looks like a technical detail; it's actually a philosophical choice.

Some sudoku apps speed up generation by sacrificing uniqueness — your solution may not match the "official one," and the system marks you wrong. This experience enrages players because it violates sudoku's core promise: logic = truth.

Our generator does backtracking verification on every hole to confirm uniqueness. It's slower but worth it. Every Zen Garden player can trust the system: if you're stuck, the puzzle isn't broken — you just haven't found the step.

That trust is the most precious design asset for a logic game.

Closing

Next time someone says "minesweeper is just sudoku, both logic puzzles," you can politely note: their information entropy structures are completely different. One is complete-information deduction; the other is partial-information probability.

Both are fun, but fun in different ways.

Leo is BverGame's co-engineer. This piece uses simplified information-theory concepts; mathematical rigor is limited. For deeper understanding, see Shannon's 1948 original paper A Mathematical Theory of Communication. Minesweeper's "unsolvable position" phenomenon was analyzed by community mathematicians in the 1990s; see minesweeper.info.