Hard Kakuro
Play it here, or print it and work in pen. Deduction gets you a long way and then stops once — and the way past is to follow one digit and see what it does.
The point where reasoning runs out
On a hard grid there comes a moment when every run has been narrowed as far as it can go, every elimination has been made, and nothing places. This is not a mistake and the puzzle is not broken. It is the defining feature of this level.
What gets you past it is picking the square with the fewest candidates left — usually two — writing one of them in, and following the consequences. Either the grid completes, in which case you were right, or it contradicts itself, in which case the other candidate was the answer. Either way you now know.
Solvers call this bifurcation. Some players consider it inelegant and prefer puzzles that never need it, which is a fair position and exactly why the easy and medium pages exist. What it is not is guessing in the ordinary sense: you are not hoping, you are testing, and the test always resolves.
How to branch without ruining the grid
Choose the square with two candidates, not three, and prefer one that sits in short runs. A branch in a corner where the runs are two and three squares long resolves in a handful of steps; a branch in the middle of two long runs can take you halfway across the grid before anything contradicts.
Mark it. In pencil, a small circle round the digit you assumed and a note of the alternative is enough. On screen the difficulty is the same and the answer is the same: know which digits followed from the assumption, because if it fails you have to take all of them back and not one more.
And follow it properly. The commonest way this goes wrong is stopping halfway, deciding it "looks fine", and carrying on. It has to run to a completed grid or to an actual contradiction — two of the same digit in a run, or a run that cannot make its clue.
Only one branch, and that is checked
A puzzle on this page needs exactly one of these. Once you have followed a single digit to its conclusion, the rest of the grid falls to ordinary reasoning — you never have to make an assumption while already inside one.
That is measured, not estimated. The grader runs the full deductive solver, and when it stalls it takes the most constrained square and tries each candidate in turn. If one round of that finishes the job, the puzzle belongs here; if a branch is still stuck and needs a second assumption inside the first, the puzzle goes to the expert page instead.
Nine-by-nine grids, around forty squares to fill. Expect half an hour and a rubber.
Common questions
- Do hard Kakuro puzzles require guessing?
- They require one branch: pick the square with two candidates, follow one, and it either completes the grid or contradicts. That is a test rather than a guess, because the outcome tells you the answer either way.
- Which square should I branch on?
- The one with the fewest candidates, preferably sitting in short runs. A branch among two-square and three-square runs resolves quickly; one in the middle of long runs can run half the grid before it settles.
- What if my branch neither completes nor contradicts?
- Then it has not been followed far enough. It must reach a finished grid or a real contradiction — a repeated digit in a run, or a run that cannot reach its clue.
Check this before you rely on it. This tool is provided free and without warranty, and its results are not professional advice.