← Back to your puzzle
BEHIND THE BOARD

How our puzzles are made and rated.

Every puzzle on Sudoku Dawn is made by the site’s own software and checked by the same logical solver that powers the hint coach. This page explains the three stages: building a complete grid, removing clues without losing the unique answer, and rating the result by the hardest technique it needs.

A complete grid first

Generation starts with a full, valid grid. A seeded search fills the 81 squares one at a time, trying digits in a shuffled order and backing up whenever a row, column or box would repeat a digit, until every square holds a legal value. The seed is what makes the process repeatable: a classic puzzle is seeded by the moment you asked for it, and a daily puzzle is seeded by its date, which is why everyone who opens the same date sees the same board and why that board never changes.

Removing clues without losing the answer

The puzzle is the grid with most of its digits taken away. The generator visits the squares in a shuffled order and tries to blank each one. After every removal it counts the solutions of what remains; if there is still exactly one, the removal stands, and if a second solution has appeared, the digit goes back. It stops when the puzzle reaches the clue count for its preset, or when no more clues can be removed safely, so some boards keep a few more clues than the target. The result always has exactly one solution.

Rated by the hardest step you need

A puzzle’s level is not decided by its clue count but by the reasoning it demands. The logical solver works through the puzzle taking the easiest available technique at every step: naked singles first, then hidden singles, then locked candidates, pairs and triples, then fish, wings, quads and chains. The level is the hardest technique it had to use.

When you choose a level for a classic game, the generator makes candidate puzzles until one rates exactly at that level, so a Hard puzzle really does need a Hard technique somewhere and never needs an Expert one.

  • Easy: naked singles only.
  • Medium: adds hidden singles.
  • Hard: adds locked candidates, naked pairs and triples, and hidden pairs and triples.
  • Expert: adds X-Wings, Swordfish, Jellyfish, XY-Wings, XYZ-Wings, naked quads, and chains of up to three strong links.

Why guessing is never required

A puzzle is only handed out at a level if the solver finished it completely using that level’s techniques. Since the solver never guesses, neither do you have to: there is always a deduction available on the board, even if it is not the one you see first. This is also what lets the hint coach promise an explanation for every step. It runs the same solver on your current board and walks you through the easiest deduction it finds, so a Hard puzzle is explained with Hard techniques and nothing harder.

Why daily levels vary

Daily boards are generated from their date with a fixed clue preset, and their level is measured afterwards rather than chosen. Most days rate Easy or Medium, Hard days are rare, and a few need Expert techniques. Occasionally a daily needs a chain longer than the Expert limit; those puzzles are still labelled Expert, the coach explains the chain in full, and our tests confirm that every daily puzzle through 2030 can be solved and explained by logic alone.

What Expert means here

Difficulty names are not standardised across Sudoku websites and books. Some grade by clue count, some by solving time, some by techniques with different boundaries. On Sudoku Dawn the names describe exactly which techniques our solver needed, as listed above, so a puzzle labelled Expert here may be called Hard, Fiendish or Diabolical elsewhere. If you want to know what a level means, the technique library shows every pattern it can contain.

The coach uses the same solver

The rater and the hint coach share one engine. That is a deliberate design choice: because a puzzle generated at a level was finished by the solver with that level’s techniques, the coach can always find and explain a next move using only those techniques. The worked examples in the technique library were found the same way, by running the solver over real daily and generated puzzles and recording positions where each pattern appears, and automated tests re-derive every example whenever the solver changes.