Kakuro Strategy: Solve Cross Sums Without Guessing (2026)

1. Read the clue triangles first

A Kakuro board is a grid of black and white cells. White cells are the ones you fill; black cells carry the clues. A black cell can hold up to two numbers, and their position tells you which direction they govern: the number in the top-right triangle is the sum of the white run to its right, the number in the bottom-left triangle is the sum of the white run below it. A black cell with only one triangle has a clue in just that direction.

A "run" is a straight line of white cells ended by a black cell (or the board edge). The rule is the same in both directions: the digits in that run are distinct, each between 1 and 9, and they add up to the clue. So before you write anything, trace every run you can see and note which black cell owns it. Most beginners lose time because they start filling a run without checking whether the same black cell also constrains a perpendicular run through the same white cell.

2. Tight clues leave almost no choice

The fastest way to start is to find clues so extreme that only one set of digits can reach them. With three white cells and a clue of 6, the only distinct digits 1–9 that sum to 6 are 1, 2 and 3 — in any order, that run is forced. A clue of 24 in three cells is equally locked: 7, 8 and 9. The two-cell versions are even sharper: a sum of 3 means 1 and 2, a sum of 17 means 8 and 9.

Scan the whole board for these tight clues before you touch a digit. Each one you resolve removes a candidate from every crossing run, and a single forced cell often cascades into the next. Clues near the middle of the range (a 4-cell run summing 20, say) leave dozens of combinations and should be left for last. The puzzle is designed so that the tight clues plus propagation are enough; you should never need to guess.

3. Use the no-repeat rule as a filter

Inside one run no digit appears twice, so a digit you place in one cell is no longer available to any other cell of that same run. That is the constraint that makes Kakuro solvable by logic rather than trial and error.

Work it as elimination. For a run of length 4 with a mid-range sum, list the digits that could still fit given what is already placed in its cells and given which digits are already used by neighbouring runs that share a cell. The intersection of those two sets is your candidate list, and a candidate list of one is a placement. Do not try to hold the whole board in your head; resolve one run, write it down, and let the new information simplify the next. The red and amber marks are there to catch the one time you do not.

4. Propagate through shared cells

Every white cell belongs to exactly two runs — one across, one down — so a digit placed for the horizontal clue must also be compatible with the vertical clue. That shared cell is the bridge, and it is where most of the real solving happens.

The practical habit: whenever you place a digit, immediately look at the perpendicular run through that cell and remove the digit from its candidates. If that removal leaves the perpendicular run with only one way to reach its sum, you have just found another placement for free. Conversely, if a run has a cell whose digit is forced by its other clue, write it in and re-check the first run. Kakuro is not a sequence of independent runs; it is a web of shared cells, and the skill is learning to follow the thread from one forced cell to the next.

5. Spend the three hints late

Each puzzle gives you three hints, and each one reveals a single correct cell from the unique solution. Used early, a hint just hands you a placement you could have found by scanning tight clues; used late, it can break a deadlock that pure elimination has not cracked in two minutes.

The honest order: scan tight clues, propagate, and only when you have genuinely stalled — no cell with a single candidate, no run with one legal completion — spend one hint, then immediately re-scan, because the revealed cell usually re-opens the cascade. Treat the three hints as a budget for the whole puzzle, not as three free moves, and try to finish at least one puzzle per tier with a hint unused. That is the only honest measure of whether the strategy is working.

6. Climb the three tiers

The library ships three puzzles in each of three tiers, and they are ordered by size and density, not by a hidden trick.

• Easy (6×7, 16 white cells): two small blocks of four cells. Good for learning to read the triangles and to spot tight clues; a puzzle here should take a few minutes.
• Medium (7×7): a single connected block with more crossing runs. The propagation chains are longer, so the "place, then check the perpendicular run" habit from section 4 starts to matter.
• Hard (7×10): the longest runs and the fewest tight clues per cell. Expect to spend real time on candidate elimination, and expect the hint budget to feel smaller.

Move up a tier only when you can finish the current one without using all three hints. The mechanics never change — digits 1–9, no repeats, sums to the clue — so difficulty is purely a matter of how long the chains are.

7. The three mistakes that stall a solve

• Filling a run before checking its crossings. A digit that fits the horizontal clue can still break the vertical one. Always confirm the perpendicular run still has a legal completion before committing.
• Ignoring amber. An amber cell means a repeated digit inside a run, which is always illegal. It is not a warning to work around; it is the puzzle telling you that an earlier placement was wrong. Fix the earlier placement, do not paper over it.
• Guessing to unblock. Because every shipped puzzle has exactly one solution, a guess that does not immediately contradict a clue can still be wrong in a way you only discover ten moves later. If you feel the urge to guess, you have missed a tight clue or a forced cell somewhere; re-scan from the corners before you commit.

FAQ

How is this different from the rules on the game page?

The game page explains how to operate the grid — tap a white cell then a digit, or type 1–9; arrow keys move; Backspace clears; Undo steps back one entry; Restart reloads; Hint reveals one correct cell; the arrows move between puzzles; a complete run that sums wrong turns red and a repeated digit turns amber. This guide assumes you know that and plays the layer above it: reading the triangles, finding tight clues, eliminating, propagating, and budgeting hints. Read the rules first, then work the strategy.

Can I really solve every puzzle without guessing?

Yes. Every puzzle in the library was verified, before release, to admit exactly one completion. There are no puzzles that require a guess followed by backtracking; the tight-clue and propagation route from section 2 and section 4 is sufficient for all nine. If you feel stuck, you have missed a forced cell, not hit a wall.

What do the red and amber marks mean?

Red means a run is complete but its digits do not sum to the clue. Amber means a digit appears twice inside one run, which is always illegal. Both are live: they update as you type, so you see a mistake the moment you make it instead of at the end. A cell can be both red and amber if it is part of a run with a wrong sum and also duplicates a digit.

When should I use Hint?

Only after you have genuinely stalled — no cell with a single candidate and no run with one legal completion. Spend one hint, then re-scan immediately, because the revealed cell usually re-opens the propagation chain. See section 5 for the full budgeting rule.

How do the three tiers differ?

By size and density, not by a new rule. Easy is 6×7 with 16 white cells in two small blocks; medium is 7×7 as one connected block; hard is 7×10 with the longest runs and the fewest tight clues per cell. The mechanics — digits 1–9, no repeats, sums to the clue — are identical in all three. Move up a tier when you can finish the current one without using all three hints.

Where can I play it?

Right on this site's Kakuro page. The mechanic — fill white cells so every across and down run sums to its clue with no repeated digit — has no single author, so this build is written from scratch by Playnesta with no byte-reused upstream files: the rules engine (js/engine.js: run extraction, the candidate and sum-feasibility solver behind the hint budget, conflict detection and the solved check), the frozen puzzle library (js/puzzles.js: nine offline-generated puzzles, each proven to admit exactly one completion), the shell (js/main.js), the stylesheet, the analytics bridge and the touch shim are all Playnesta-original (MIT, see LICENSE.txt in the package). During selection (2026-10-16) the certified upstream search came back with no self-hostable interactive base: taylorjg/dlxlib-demos ships no licence, senegrom/GridPuzzle is AGPL-3.0, and billstclair/kakuro-master (MIT, Copyright Bill St. Clair) is a Cordova/PhoneGap project rather than a plain static build, so it is recorded as a mechanism reference only — no bytes are copied and no file under its licence is redistributed. Playnesta Games is the developer and publisher of this web implementation.