The cover says “Extreme Kakuro.” You open to a 15x15 grid and finish it in twelve minutes, because every corner has a 3 or a 17 in two cells that practically fills itself in. Two pages later there’s a little 7x7 marked “Medium.” Forty minutes in, you’re staring at a block of empty cells with no idea where to start.
The labels aren’t lying on purpose. They’re measuring the wrong thing. Big grids take longer; they aren’t necessarily harder. What makes a Kakuro hard is the toughest step the solver is forced to make, and that’s something you choose when you build the puzzle. This article sets out a four-tier grading scale, shows what each level looks like on the page, grades the series 7x7 grid, and then refills the same layout so it needs a deduction one tier harder.
Grid Size vs. Difficulty
Long Isn’t Hard
A 15x15 grid full of easy steps is a lot of easy steps. It takes time, not thought. A 7x7 can force a step that takes real insight, and one such step is enough to make the whole puzzle hard.
So define difficulty this way: the hardest technique the solver must use to finish without guessing. Not the average step, and not the number of steps. The single hardest one.
Inconsistent Labels
It’s tempting to label puzzles by grid size or clue count, because those are easy to measure. The result is a set of puzzles where “Hard” means three different things on three pages. Solvers notice. Grading by technique fixes that, because every label means the same deduction wherever it appears.
A Four-Tier Grading Scale
This is a practical scale for constructors, not an industry standard. Each tier is a kind of deduction, and a puzzle’s grade is the highest tier it requires.
Tier 1: Magic Blocks (Easy)
Some sums have only one possible set of digits. These magic blocks let the solver read digits almost straight off the clue:
| Run length | Magic blocks |
|---|---|
| 2 cells | 3 {1,2}, 4 {1,3}, 16 {7,9}, 17 {8,9} |
| 3 cells | 6 {1,2,3}, 7 {1,2,4}, 23 {6,8,9}, 24 {7,8,9} |
| 4 cells | 10 {1,2,3,4}, 11 {1,2,3,5}, 29 {5,7,8,9}, 30 {6,7,8,9} |
When two magic blocks cross, the shared cell often falls out immediately. A 17 across ({8, 9}) crossing a 16 down ({7, 9}) shares only 9, so that cell is 9.
Not every pair can cross. A 3 in two cells is {1, 2}, and it shares nothing with a 16 or a 17. Put them across each other and the shared cell has no possible digit. That isn’t a hard puzzle; it’s a broken one.
Tier 2: Intersections (Medium)
Most clues have several possible sets. The solver narrows a cell by comparing what its across run allows with what its down run allows, then keeps comparing as digits fill in. That takes pencil marks and attention, but no new ideas.
The series grid’s first move is a Tier 2 step. The 6 down along the right edge is three cells, so it’s {1, 2, 3}, a magic block. The 27 across along the bottom row is four cells, which could be {3, 7, 8, 9}, {4, 6, 8, 9} or {5, 6, 7, 9}. Only 3 appears in both, so the corner cell r6c6 is 3. One magic block, one comparison.
Intersection Logic covers this tier in depth from the solver’s side.
Tier 3: Region Counts, Innies and Outies (Hard)
Pick a block of cells. Add up the across clues that cover it exactly, then add up the down clues that pass through it. If the down runs stick out past the block, or don’t fully cover it, the difference between the two totals is the value of the cells that only one direction covers. Those cells are outies (outside the block) or innies (inside it).
No 45 is involved. In Kakuro you compare across against down. The number 45 only enters the count when a run is all nine cells long, because only a nine-cell run is guaranteed to hold 1 through 9. Mastering the 45 Rule covers the solving side.
Tier 4: Parity and Chains (Expert)
Odd/even counting across connected runs, and short “if this cell is 4, then that one can’t be 2” chains. This is the top of the scale. Anything beyond it requires guessing, and a good puzzle never does. Eliminating Guesswork in Kakuro works through examples.
The grading rule: Tier 1 → Easy, Tier 2 → Medium, Tier 3 → Hard, Tier 4 → Expert.
As a test you run on your own grid, the scale works as a ladder. Stop at the first rung that finishes the puzzle:
What Each Level Looks Like
Easy
Several magic blocks spread around the grid, usually near corners and edges. Each placed digit leads straight to the next one, and the solver rarely needs to hold more than one run in their head.
Medium
A few magic blocks, plus clues with several possible sets. The solver needs intersections and some pencil marks, but never has to step back and count a whole region.
The series grid, as repaired in How to Make Sure Your Kakuro Has Exactly One Solution, is a Medium:
+-------+-------+-------+-------+-------+-------+-------+
|#######| 16\ | 12\ | 10\ | 8\ |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \12 | | | | |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \27 | | | | | 21\ |#######|
+-------+-------+-------+-------+-------+-------+-------+
| \3 | | | 19\8 | | | 6\ |
+-------+-------+-------+-------+-------+-------+-------+
|#######| \8 | | | 17\8 | | |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \16 | | | | |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \27 | | | | |
+-------+-------+-------+-------+-------+-------+-------+
It has three magic blocks: the 3 across in row 3 ({1, 2}), the 17 down in column 4 ({8, 9}) and the 6 down in column 6 ({1, 2, 3}). Intersections finish every cell from there. No region counts, no parity.
Hard
Few magic blocks, and the easy steps run out partway through. The solver gets a foothold, makes some progress, then hits a wall that only a region count gets past.
A common misconception is that a Hard grid must have no magic blocks at all, or that the first digit should need a region count. That usually makes a puzzle nobody can get into. The more satisfying shape is the one in the next section: a few footholds, then a wall.
Pacing the Solve
A good hard puzzle has a shape: foothold → wall → breakthrough → cascade.
The foothold gets the solver onto the page and building momentum. The wall is where the difficulty lives: progress stops, and the usual moves don’t work. The breakthrough is the one deduction that gets past it. The cascade is the reward, as the rest of the grid falls quickly once the wall is down.
One clear wall feels better than several small ones. Three stalls in a row reads as a slog, not a challenge. And after the breakthrough, the finish should be fast. A second wall right after the first one makes the breakthrough feel like it didn’t count.
Compare the two series grids. The Medium never stalls. Every move leads to the next at a steady rate. The Hard, below, places seven cells, stops dead with 17 cells open, and then finishes all 16 remaining cells after one region count. Same layout, a different shape entirely.
Practical Example: Making the Series Grid Harder
The goal: keep the series layout and make it need a Tier 3 step.
- Same layout, so same symmetry
- One magic block only
- Intersections alone must stall
- A region count must restart the solve
- Still exactly one solution
Step 1: Grade the starting grid. The Main grid above solves on Tiers 1 and 2, so it’s Medium.
Step 2: Refill the layout. Changing a few clues on the Main grid isn’t enough, because a clue is just the sum of the digits under it. To change what the solver experiences, you change the digits. Here’s the same layout with a completely new fill:
+-------+-------+-------+-------+-------+-------+-------+
|#######| 24\ | 15\ | 6\ | 14\ |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \17 | | | | |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \13 | | | | | 24\ |#######|
+-------+-------+-------+-------+-------+-------+-------+
| \12 | | | 11\12 | | | 9\ |
+-------+-------+-------+-------+-------+-------+-------+
|#######| \10 | | | 7\5 | | |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \14 | | | | |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \27 | | | | |
+-------+-------+-------+-------+-------+-------+-------+
Only one clue is a magic block now: the 24 down in column 1, which must be {7, 8, 9}.
Step 3: Solve until it stalls. The 24 gives the first foothold. r2c1 is also in the 13 across on row 2, a four-cell run. Four different digits adding to 13 can’t include anything above 7 (13 − 1 − 2 − 3), so r2c1 = 7. Six more cells fall, mostly on the right side: r6c5 = 9, r5c5 = 8, r3c4 = 9, r3c5 = 3, r4c5 = 4 and r4c6 = 1.
Then nothing. Seventeen cells are still open, each with two or three candidates, and no crossing narrows any of them further.
Step 4: Count the region. Take the block covered by the three left-hand across runs: row 1 (17), row 2 (13) and the left run of row 3 (12). Across total: 42.
Now add the down runs that pass through that block: column 1 (24), column 2 (15), the top run of column 3 (6) and the top run of column 4 (14). Down total: 59.
+-------+-------+-------+-------+-------+-------+-------+
|#######| 24\ | 15\ | 6\ | 14\ |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \17 | · | · | · | · |#######|#######|
+-------+-------+-------+-------+-------+-------+-------+
| \13 | 7 | · | · | · | 24\ |#######|
+-------+-------+-------+-------+-------+-------+-------+
| \12 | · | · | 11\12 | 9 OUT | 3 | 9\ |
+-------+-------+-------+-------+-------+-------+-------+
|#######| \10 | OUTIE | | 7\5 | 4 | 1 |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \14 | | | 8 | |
+-------+-------+-------+-------+-------+-------+-------+
|#######|#######| \27 | | | 9 | |
+-------+-------+-------+-------+-------+-------+-------+
The down runs cover the whole block plus two cells that poke out of it: r4c2 (the bottom of the column-2 run) and r3c4 (the bottom of the column-4 run). So those two cells must add up to the difference: 59 − 42 = 17. Two cells making 17 can only be 8 and 9. r3c4 is already 9, so r4c2 = 8.
Step 5: Watch the cascade. With r4c2 = 8, the 10 across on row 4 makes r4c3 = 2. From there the remaining cells fall in a steady run: r3c2, r5c3, r6c3, r2c4, r3c1, r6c4, r6c6, r1c1, r1c4, r5c4, r5c6, r1c2, r1c3, r2c2 and r2c3. That’s 16 cells after the breakthrough, no second wall.
Step 6: Confirm one solution. A forced-only solve that includes the region count completes, and a solution counter (see Build a Kakuro Solver and Generator in Python) finds exactly one answer.
You end up with the same layout and a Hard grade: one clear wall, then a fast finish. Both grids are in the printable series puzzles. Solve them back to back and you’ll feel exactly where the Hard one stops you.
Best Practices
Do:
- ✓ Grade by test-solving and noting the hardest step you needed.
- ✓ Keep difficulty consistent across a set of puzzles.
- ✓ Give hard puzzles one clear breakthrough rather than many small stalls.
Don’t:
- ✗ Confuse a puzzle with two solutions with a hard puzzle. Ambiguous is not difficult.
- ✗ Require guessing at any level.
- ✗ Grade by grid size.
Troubleshooting
A “Hard” puzzle solves too fast. A magic block or a crossing that allows only one digit gives the solver a way around the wall. Find the shortcut in your test solve, change the digits there so its clue has more possible sets, and re-check that the puzzle still has exactly one solution.
Testers give up on an “Expert” puzzle. The breakthrough is too well hidden, or there are several walls in a row. Make the key region easier to spot: a block with only one or two cells poking out is much easier to find than one spread across the grid. Or add a foothold near it so the solver’s attention is already in the right place.
The region count gives an impossible number. A run you counted doesn’t lie completely inside or completely across the region, or you counted a clue twice. List every run you’re counting, mark which cells each one covers, and check that the across runs cover the block exactly.
What to Take Away
Difficulty is the hardest step you force, not the size of the grid. Grade on four tiers: magic blocks, intersections, region counts, parity and chains. Region counts compare across with down, and 45 doesn’t come into it. The best hard puzzles give a foothold, one wall and a fast finish. And to make a grid harder, change its digits, not just its label.
Grade a few puzzles from your shelf by their hardest step, then solve both series puzzles and notice where the Hard one stalls. Build a Kakuro Solver and Generator in Python automates this whole grading process.
The Kakuro Puzzle Design series:
- How to Make a Kakuro Puzzle
- Designing a Kakuro Grid
- How to Make Sure Your Kakuro Has Exactly One Solution
- Build a Kakuro Solver and Generator in Python
From the solving side: Kakuro Unique Sums (Kakuro for Beginners) and Essential Magic Blocks (Advanced Kakuro Tactics).