You solve a Sudoku puzzle by filling a 9x9 grid so every row, column, and 3x3 box contains the digits 1 through 9 exactly once, using only logic, not guessing. Every well-formed puzzle has one unique solution reachable by deduction. Start by scanning for digits that can only fit one cell, then track candidate digits per cell in pencil, and escalate to advanced techniques only when simple scans stall.

Mechanism: why constraint propagation does the work

A Sudoku is a constraint-satisfaction problem, not a math problem. The arithmetic never matters; only the uniqueness rules do. Each cell sits at the intersection of three units: one row, one column, one box. That means every digit you place eliminates a candidate from 20 other cells: 8 in its row, 8 in its column, and 4 remaining in its box. The cascade is the point. A single correct placement can collapse several near-solved units at once.

The core technique is cross-hatching: for each digit 1-9, scan rows and columns to see where that digit is already fixed, then block out those lines within a target box. When eight cells in a box are blocked for a digit, the ninth is forced. This works because the rule is exclusive: if a digit already appears in a row, no cell in that row can hold it again.

Candidates are the second engine. In pencil, list every digit a cell could still legally hold. A cell with one remaining candidate is a naked single. A digit with only one legal cell inside a unit is a hidden single. Both are forced placements, not guesses. Human solvers use pencil marks because visual scanning alone cannot track nine digits across 27 units at once.

Here is the plain-language version: the grid is a machine that eliminates options. Every correct digit you enter makes the remaining cells easier, and difficulty comes from how long elimination takes to force a placement.

Quantified context: why some puzzles stall

A completed Sudoku has 81 cells and 729 possible digit placements (81 cells times 9 digits). A typical published puzzle gives you 25 to 35 givens, meaning roughly 30 percent of cells are pre-filled. The remaining 50-plus cells must be deduced.

Difficulty is not set by given count alone. A puzzle with 30 givens can be harder than one with 25 if the givens create fewer forced placements early. The minimum for a unique solution is 17 givens, a threshold proved by exhaustive computer search; puzzles below that count cannot have a single solution. Most newspaper puzzles sit well above the floor and are solvable with singles and simple candidate work.

Pencil-marking overhead is real: fully marking candidates in a sparse grid can mean writing 3 to 5 digits per empty cell, or 150 to 250 marks. That is why experienced solvers mark only constrained units or use dot notation for compactness.

Timeline table: a typical solving progression

Phase What you do Typical time (easy puzzle) Signal to escalate
1. Cross-hatch Scan each digit across rows, columns, boxes 1-3 min No forced placements found
2. Naked singles Check cells with one remaining candidate 1-2 min Singles exhausted
3. Hidden singles Find digits with one legal cell in a unit 2-4 min No hidden singles remain
4. Candidate marking Pencil all remaining candidates 3-5 min Grid fully marked
5. Intersections Pointing pairs and box-line reduction 3-6 min No eliminations triggered
6. Advanced patterns Naked/hidden pairs, then triples 5-15 min Still stalled
7. Harder chains X-Wing, then XY-Wing and beyond 10-30 min Puzzle solved

This progression is a default, not a law. Some puzzles reward jumping straight to intersections; others resolve entirely in phases 1-3.

Edge cases: when the standard advice fails

Sparse grids mislead scanners. A puzzle with 17-22 givens often has almost no cross-hatch targets, because each digit appears only two or three times. Here, marking candidates early beats scanning, because the forced placements are hidden singles buried in candidate lists, not visible line blocks.

Givens can be decoys. A given that looks helpful may sit in a crowded row and contribute nothing to cross-hatching. Symmetric givens (rotationally paired) are a common design choice but carry no solving advantage.

Symmetry is cosmetic. Puzzle symmetry looks elegant and affects nothing logically. Do not let a symmetric pattern bias where you look.

Digit distribution is uneven by design. Some digits appear as givens more often than others, which makes them easier cross-hatch targets. Scanning the most frequent digit first is a genuinely useful heuristic, not a shortcut.

Comparison table: solving techniques by cost and payoff

Technique Prerequisite What it eliminates Typical appearance
Cross-hatching Some givens per digit Whole cells (forced placement) Nearly all puzzles
Naked single One candidate left That cell All puzzles
Hidden single One legal cell per digit in a unit That cell All puzzles
Pointing pair Candidates aligned in a box Candidates along a line Medium puzzles
Box-line reduction Candidates aligned in a row or column Candidates inside a box Medium puzzles
Naked/hidden pairs Two cells sharing two candidates Those candidates elsewhere Hard puzzles
X-Wing Four cells forming a rectangle One candidate in two lines Hard puzzles
XY-Wing Three cells with linked candidates One candidate in a shared unit Expert puzzles

The ordering matters less than the discipline: exhaust cheap techniques fully before reaching for expensive ones. A solver who checks all singles before any X-Wing will finish faster on average.

Misconceptions: guessing, arithmetic, and difficulty myths

The biggest misunderstanding is that hard puzzles require guessing. They do not. A valid puzzle has exactly one solution, and every cell is reachable by logic, though "logic" may include pattern techniques most solvers never learn. Guessing is a symptom of an exhausted technique list, not a requirement.

Sudoku is not arithmetic. You never add, subtract, or compare values. The digits are symbols; any nine distinct symbols would work identically. Do not look for numerical patterns.

Pencil marks are not optional at higher difficulty. Solvers who refuse to mark candidates hit a wall around medium difficulty, because hidden relationships between cells are invisible without a written candidate map.

Difficulty labels are unreliable across sources. A newspaper "hard" can be easier than a book "medium," because the labels reflect the techniques the publisher expects, not an objective measure. Judge a puzzle by how long your techniques stall, not by its ranking.

One counterintuitive detail professionals rely on: erasing is a warning sign. If you are frequently rubbing out candidate marks, you placed a digit without sufficient proof. Each placement should eliminate candidates permanently; churn suggests a wrong entry upstream. Treat repeated erasing as a red flag and audit your last few placements.

Red-flag warnings worth internalizing:

A puzzle with no forced placements anywhere usually needs pencil marks, not more scanning. A grid with two cells sharing identical single-candidate pairs may hide a fork; check for a wrong earlier entry before advancing. Any puzzle with more than one solution is malformed, and no logic will finish it uniquely. Placings justified only by "it seems to fit" should be avoided until all singles are exhausted. Difficulty rankings from different publishers cannot be compared directly. A stalled grid after full candidate marking means you need a named technique, not more staring.

Practical actions: a repeatable solving routine

Work in a fixed order so you never lose track of what you have exhausted.

Scan the digit with the most givens first. Cross-hatch it across all nine boxes and place every forced cell. Repeat for each remaining digit in descending order of frequency. Sweep for naked singles. Check every empty cell that touches at least three or four filled cells; its candidate list is short and a single often pops out. Sweep for hidden singles unit by unit. For each row, column, and box, ask which digit can legally sit in only one cell there. Mark candidates only when steps 1-3 stall. Mark compactly, using dot positions rather than full digits, and keep marks consistent. Hunt intersections next: pointing pairs first, then box-line reduction. Both are cheap and often crack a stalled grid. Escalate to pairs and triples, then to rectangle patterns like X-Wing. Apply each technique across the whole grid before moving to the next tier. Audit before any hard technique. Verify every placement once; one bad digit poisons the entire candidate map and causes the erasing churn described earlier.

Set concrete stopping thresholds. If five minutes of scanning produces zero placements, switch to marking. If marking plus intersections produces zero eliminations, re-audit the grid before reaching for advanced patterns. On paper, use one pencil weight for givens and another for deductions so an accidental mark is visually obvious.

Limitations: what logic cannot promise

Logic guarantees a solution only when the puzzle is well-formed and unique. Poorly generated puzzles with multiple solutions defeat any technique, because different valid completions will each locally satisfy the rules. If you reach a contradiction with correct deductions, the puzzle itself is likely faulty; stop and check the givens.

Human technique lists are also finite. Extremely hard puzzles may require long forcing chains that are technically logical but exhausting to execute by hand, and most solvers reasonably treat these as outside their toolkit. Speed has a ceiling too: an easy puzzle takes a practiced solver a few minutes, while expert grids can absorb 30 minutes or more regardless of skill. Finally, difficulty is solver-relative; the same grid that stalls one person resolves quickly for another with a different technique vocabulary. None of this makes the method unreliable. It simply marks where deduction ends and puzzle design or human patience takes over.