No Guessing Allowed: How Certain, Contour, Lattice and Lumen Prove Every Puzzle Can Be Solved by Logic
Every minesweeper player knows the moment. The board is nearly clear, two cells are left, one of them hides the last mine, and the numbers around them say nothing that separates the two. You tap one. Half the time the game ends. Nothing you could have done would have changed the odds, and that is what makes it feel unfair: you were beaten by the dice, not by the puzzle. This article is about the rule that runs through the four logic games on this site — Certain, Contour, Lattice and Lumen — which is that no puzzle reaches you unless it has been shown that you will never need that tap.
Two different promises
It helps to separate two things that are often run together. The first is that a puzzle has exactly one solution. The second is that the solution can be reached by reasoning, one certain step after another, without ever trying something to see what happens.
The first does not imply the second. A puzzle can have a single answer that is only findable by trial and error: you assume a cell is filled, follow the consequences for twenty moves, hit a contradiction, and back out. Strictly speaking that is still logic, but nobody enjoys doing it in their head, and on a phone it is indistinguishable from guessing. Classic minesweeper fails even the first promise — a randomly placed field often has several arrangements of mines that fit every number on the screen. Hand-made nonograms and Slitherlink puzzles usually keep the first promise and sometimes break the second.
The four games here make both promises, and the second is the one that takes real work.
A solver that is only allowed to reason
The tool that checks a puzzle is a solver, but a deliberately limited one. A general-purpose solver would happily search through possibilities until it found the answer, and it would declare almost any puzzle solvable. That is no use here, because a search is exactly what a person should not have to do. So the checking solver is restricted to the kinds of deduction a person makes: it looks at a clue, or a small group of clues, and asks what must be true. If it can prove a cell is safe, filled, lit or part of the loop, it marks it and starts again. It never assumes anything and never backtracks.
When that solver gets stuck before the puzzle is finished, the puzzle is rejected. Not adjusted in the hope that a player will find something the solver could not — rejected. That one rule is the whole guarantee. If the restricted solver can clear a board from start to finish, then a chain of ordinary deductions exists from the first move to the last, and you can find it too.
What counts as an “ordinary deduction” depends on the game.
Certain: counting mines
In minesweeper the basic rule is counting. A 1 with only one hidden neighbour means that neighbour is a mine. A number whose mines are all already flagged means every other neighbour is safe — the move Certain’s hint describes as “this number already has all its mines”. Most boards need more than that, and the next step is comparing two numbers that share neighbours. If one number needs one mine among cells A and B, and the number beside it needs one mine among A, B and C, then C is safe: the mine both numbers are counting must be in A or B.
A candidate board is generated, the solver plays it from the opening using only rules like these, and the board is kept only if the solver clears it without ever being forced to pick between cells it cannot tell apart. The summary at the end of a game — “Logic needed: basic counting”, or something deeper on harder boards — is the solver reporting which rules it actually had to use.
Lattice: lines against columns
Nonograms are solved one line at a time. The most useful rule is overlap: in a row of fifteen cells with a single clue of 11, wherever the run of eleven sits, it must cover the middle seven cells, so those seven can be filled immediately. Every filled or crossed cell then changes what is possible in the column that crosses it, and the columns feed back into the rows.
A solver that only works line by line — never holding two lines in mind at once — clears a large share of all nonograms, and it is the yardstick for Lattice. Each of the 53 hand-drawn pictures and every practice puzzle has to be completed by repeated line solving alone. A picture that would require testing a guess across several lines is redrawn or dropped. It is also why the game is happy to grey out a clue once its line is satisfied: the state of each line on its own is always enough to tell you where to look next.
Contour: edges and the single loop
Slitherlink has two kinds of rule. The local ones come from the numbers: a 0 crosses out all four of its edges, a 3 in a corner of the grid must use both of its outer edges, two 3s side by side must share the line between them. The global one comes from the loop itself: every dot has either zero or two lines touching it, and the loop must be a single closed path, so you may never close a small loop while other lines are still unconnected.
That global rule is what makes Slitherlink harder to check than the others, because a deduction can depend on lines far apart on the board. Contour’s solver applies both kinds — local patterns and the “don’t close a loop early” rule — and a puzzle is kept only if they are enough to draw the whole loop. That is also why crosses matter so much in the game. A crossed-out edge is a finished deduction, and it is usually what lets the next one happen.
Lumen: who can light this cell?
In Light Up the numbered walls give the first moves: a 4 needs bulbs on all four sides, a 0 rules all four out. After that, the rule that does the most work is a question you ask of a single dark cell: which positions could possibly light it? If only one cell can reach it — every other candidate is blocked by a wall, already lit by something else, or ruled out by a number — the bulb must go there. Lumen’s solver repeats that question, along with the wall counts and the rule that two bulbs may never see each other, and keeps a puzzle only if the board fills from those steps alone.
Where difficulty comes from
Take guessing away and you might expect every puzzle to become easy. It does not, because the depth of reasoning varies enormously. A gentle board can be finished with the simplest rule over and over. A hard one has stretches where none of the simple rules apply anywhere on the board, and the only way forward is a comparison between two clues, or a chain where one deduction unlocks a second that unlocks a third.
That gives a much more honest measure of difficulty than grid size. The checking solver records which rules it had to reach for, and how often, and that is what the difficulty labels are built on. A big board that only needs counting is a long puzzle, not a hard one; a small board that needs a subset argument in the middle is the opposite.
Hints come from the same place
Because the solver works the way a person does, its reasoning can be shown to you. When you ask for a hint, the game looks at your current board and points to a deduction you can make — which clue proves what, and why. That is more useful than revealing a cell. A reveal tells you the answer; a hint tells you the move, and after a few of them you start spotting the same patterns before you need to ask.
What the promise does not cover
“Solvable by logic” is a promise about the puzzle, not about every move you make. A wrong flag in Certain, a stray line in Contour or a misplaced bulb in Lumen can lead you somewhere the puzzle never intended, and the logic you build on top of a mistake is sound but useless. That is what the optional mistake highlighting is for, and why Certain has a Zen mode where a mis-tap simply undoes itself.
It is also not a promise that a puzzle is easy to see. The deduction always exists; finding it can still take a while. That, really, is the point. When you are stuck in one of these four games, you are stuck the way you would be on a crossword clue, not the way you are at a roulette table — and the way out is to look again, not to tap and hope.