GenerateRandomSearch

Futoshiki Generator

Futoshiki gives you a grid where every row and column holds each number once, and almost no numbers to start with. What you get instead are inequality signs between neighbouring cells. The solve is pure deduction from relationships, which makes it a very different exercise from Sudoku even though the underlying object is the same. Every puzzle here is solved exhaustively before you see it, so it has exactly one answer, and every clue has been tested for necessity.

What this generator does

Builds a completed Latin square, derives every inequality between adjacent cells, then removes clues one at a time and keeps a removal only while the puzzle still has exactly one solution. The result is a minimal clue set: take away any single sign and the puzzle becomes ambiguous. A handful of digits are revealed only where the inequalities alone cannot pin the square down.

How to use this tool

  1. Choose a grid size. Five is the classic; four is a gentle introduction.
  2. Press make a puzzle. Signs between cells show which of the two is larger.
  3. Solve it, then reveal the answer to check.
  4. Use a seed if you want a whole class to get the same puzzle.

Understanding the controls

Size
From 4 to 6. The work grows steeply, because every candidate clue removal is tested by solving the puzzle exhaustively without it: a 4 by 4 takes about 50ms and a 6 by 6 about 800ms. Seven is not offered — it did not finish in five minutes when measured, and a size that freezes the tab is worse than one that is missing.
Seed
Any word reproduces the same puzzle exactly. That is the difference between a puzzle you can hand out to thirty people and one you cannot.
Show the answer
Reveals the unique solution. It is the same grid the puzzle was generated from, not a second solve, so it cannot disagree with the clues.

Common use cases

  • A logic starter that needs no arithmetic, so it works across a wide age range
  • Puzzle practice for someone who finds Sudoku has become mechanical
  • A printable set for a cover lesson, generated from a seed so it can be reproduced
  • Demonstrating constraint propagation in a way people can do by hand
  • A reasoning warm-up that takes minutes rather than a coffee break

How this generator works

A random Latin square is built by backtracking over shuffled candidate values. Every inequality between orthogonally adjacent cells is then derived from it. Cells are revealed until the constraint set has exactly one solution — necessary, because a full set of inequalities does not always determine the square on its own. Then clues and givens are removed one at a time, each removal kept only when an exhaustive solver still counts exactly one solution. The solver stops counting at two, because more than one is all the information the decision needs.

Randomness and fairness

The underlying square and the order clues are tried in are both random. Without a seed this uses the browser's cryptographically secure generator; with a seed it uses a small deterministic generator, which is reproducible and therefore explicitly not cryptographically secure. The puzzle you get is random; its having one answer is not.

For how randomness is produced across the whole site, see how Generate Random works.

Assumptions this tool makes

  • Every row and column contains each number from 1 to the grid size exactly once.
  • A puzzle should have exactly one solution, reachable without guessing.

Limitations and good to know

  • Difficulty is not graded. A minimal clue set is not the same as a gentle one, and two puzzles of the same size can differ a lot in how hard they feel.
  • Sizes above 6 are not offered. Verifying necessity means an exhaustive solve per candidate clue, and at 7 by 7 that did not finish in five minutes — raising the cap needs a constraint-propagating solver, not a faster machine.
  • There is no step-by-step solver, so it will not explain the deduction — only confirm the answer.
  • Nothing is stored between visits. Use a seed if you need the same puzzle again.

Common mistakes

Reading a sign as pointing at the larger number
The sign is an ordinary greater-than: the open end faces the larger value. Vertical signs use the same rule read downwards.
Assuming a chain of signs fixes the values
Three cells in increasing order only tells you the order. Which actual numbers they are still depends on the rest of the row and column.
Guessing when the deduction stalls
Every puzzle here is solvable by deduction alone, so a stall means a clue has not been used yet rather than that a guess is needed.

Practical tips

  • Start at the extremes: a cell that must be larger than two neighbours cannot be 1, and one smaller than two cannot be the maximum.
  • Work the longest chain of signs first — it constrains the most cells at once.
  • For a class set, seed with the date and the group name so the puzzle can be regenerated months later.

Privacy and your data

Puzzles are generated entirely in your browser. No seed, puzzle or solution is sent anywhere, and nothing is stored between visits.

Frequently asked questions

Is there always exactly one answer?
Yes, and it is verified rather than assumed. Each puzzle is solved exhaustively during generation, and any clue whose removal would allow a second solution is kept.
Why are some numbers already filled in?
Because inequalities alone do not always determine the square. Cells are revealed only until the puzzle has one answer, and each of those is then tested to see whether it can be taken away again.
How is this different from Sudoku?
Sudoku gives you digits and adds box constraints. Futoshiki gives you relationships and no boxes, so the deduction runs through comparisons rather than through elimination on given values.