You’re down to the last four squares of an expert Kakuro. You try one fill and every clue checks out. You erase it, try the other, and every clue checks out again. Your pencil hovers. One of them has to be wrong, and the puzzle has stopped giving you hints.
So you pick one. If you’re lucky, the grid finishes and you feel nothing. If you’re not, you find out eleven moves later, and now you can’t tell which of those eleven moves to undo.
That’s bifurcation: picking a branch and seeing if it breaks. It’s the move pure-logic solvers try hardest to avoid, and on a well-built Kakuro you never need it. When two fills both seem to work, one of them breaks a rule you haven’t checked yet. The job is finding which rule, before you write anything down.
The Problem with Bifurcation
Why Guessing Ruins the Game
A deduction carries its proof with it. When you place a 9 because a 16 crosses a 17, you know why the 9 is there, and that reason holds no matter what happens later. A guess carries no proof. Every digit you place after it depends on it, and if the guess is wrong, all of them are wrong too.
The real cost shows up late. A bad guess rarely fails right away. It fails ten or twenty moves later, in a part of the grid that looks unrelated, and by then you can’t untangle which digits came from logic and which came from the coin flip. Wikipedia’s Kakuro article puts it plainly: brute-force guessing is possible, but understanding the combinations an entry can take is the more efficient approach.
Guessing also misses the point of the puzzle. The setter hid a deduction in that corner. Finding it is the game.
Understanding Parity (Even/Odd Logic)
Basic Parity Rules
Three facts from grade school:
even + even = even
odd + odd = even
even + odd = odd
For a two-cell entry, that means the clue tells you the mix before you know a single digit:
Odd clue (3, 5, 7, 9, 11, 13, 15, 17) → one odd digit + one even digit
Even clue (4, 6, 8, 10, 12, 14, 16) → both odd, or both even
For longer entries the rule generalizes: the clue is odd exactly when the entry holds an odd number of odd digits. A 6 in three cells is {1, 2, 3}, two odds, so the total is even.
Two of the magic blocks have a parity bonus. A 4 in two cells is {1, 3}, and a 16 is {7, 9}, so every cell in a 4 or a 16 is odd. A 3 {1, 2} and a 17 {8, 9} always have one of each.
Applying Parity to the Grid
Parity becomes useful when it crosses another clue. Say a cell sits in a 16 across, so it’s odd. Its down entry is a 10 in two cells, which could be {1, 9}, {2, 8}, {3, 7}, or {4, 6}. The 10 is even, so its two digits are either both odd or both even. The shared cell is odd, so the partner is odd too. {2, 8} and {4, 6} are gone, and the partner is 1 or 3.
It also works with even cells. If a cell has to be even and sits in an 11 in two cells ({2, 9}, {3, 8}, {4, 7}, {5, 6}), it can only be 2, 4, 6, or 8. Its partner takes the odd half: 9, 7, 5, or 3. That’s half the candidates gone without listing a single combination.
Parity never finds anything a full list of combinations wouldn’t. What it gives you is speed. “Odd clue, two cells, so one of each” is a thought you can have at a glance, which makes it the right tool when you’re staring at a stuck corner and need to see its shape.
Advanced Conflict Resolution
The “Ghost” Duplicate
Sometimes a candidate looks fine against both of its own clues, but placing it forces a digit somewhere else that already exists in that entry. The duplicate isn’t on the board yet. It’s a ghost: it only appears if you follow the consequences.
Hunting ghosts means following a forcing chain: assume one candidate, follow only the moves that assumption forces, and watch for a collision. Three rules keep a chain from turning into a guess:
- Every link must be forced. “This cell is 4, so its partner in the 5 must be 1.” If a link has two possible outcomes, stop. You’ve started guessing.
- Keep it short. Three or four links is a deduction. Fifteen is a guess in disguise.
- Never write the assumption down. Trace it in your head or with your finger. If it collapses, you remove one candidate. If it doesn’t collapse, you’ve learned nothing and the grid is untouched.
A chain that ends in a contradiction proves the starting candidate is wrong, so you can cross it out. A chain that ends peacefully proves nothing, so you can’t place the starting candidate just because nothing broke.
Practical Example: Resolving a Deadly Pattern
Sudoku players know the deadly pattern, or unique rectangle: four cells in a rectangle where two digits could swap places without breaking any rule. Kakuro has the same trap. If a 2×2 block of white cells reads x, y over y, x, you can swap x and y and every across sum and every down sum stays the same. A puzzle with one solution can’t end in that shape.
Here’s a grid with something that looks like one. Clue cells read down\across, so 5\ is a down clue of 5 and \16 is an across clue of 16.
+-------+-------+-------+-------+-------+-------+
|#######| 17\ | 4\ | 13\ |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \17 | | | |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \11 | | | | 5\ | 9\ |
+-------+-------+-------+-------+-------+-------+
|#######|#######| \16 | | | |
+-------+-------+-------+-------+-------+-------+
|#######|#######|#######| \4 | | |
+-------+-------+-------+-------+-------+-------+
The top falls quickly. The 17 down is {8, 9} and the 4 down is {1, 3}. The 11 across starts with one of each. If it started with 9, its last two cells would need to total 2, which is impossible. So it’s 8, and the remaining 3 is 1 + 2, with the 1 under the 4 down. That fixes the 17 across at 9, 3, and 5. The 13 down then needs 13 − 5 − 2 = 6 at its bottom.
Then the corner stalls. The 4 across is {1, 3}, both odd. The 5 down and the 9 down are odd clues, so each has one odd digit and one even digit. Their bottom cells are odd, so their top cells are even: 5 − {1, 3} gives {4, 2}, and 9 − {1, 3} gives {8, 6}.
+-------+-------+-------+-------+-------+-------+
|#######| 17\ | 4\ | 13\ |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \17 | 9 | 3 | 5 |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \11 | 8 | 1 | 2 | 5\ | 9\ |
+-------+-------+-------+-------+-------+-------+
|#######|#######| \16 | 6 | 2 4 | 6 8 |
+-------+-------+-------+-------+-------+-------+
|#######|#######|#######| \4 | 1 3 | 1 3 |
+-------+-------+-------+-------+-------+-------+
Now look at the two ways to finish. Fill A puts 2, 8 on top and 3, 1 below. Fill B puts 4, 6 on top and 1, 3 below. Check the clues: 2 + 3 = 5 and 4 + 1 = 5, so both fill the 5 down. 8 + 1 = 9 and 6 + 3 = 9, so both fill the 9 down. 2 + 8 = 10 and 4 + 6 = 10, so both give the 16 across the 10 it needs. Everything balances. This is where you’d reach for a coin.
Notice, though, that this isn’t a true deadly pattern. In a real one, the same two digits trade places. Here the bottom row swaps 1 and 3, but the top row changes from {2, 8} to {4, 6}, which is a different set of digits. Different digits means some clue can still tell the two fills apart.
Step 1: Assume one candidate. Take the bottom-left cell and suppose it’s 1.
Step 2: Follow only forced moves.
Assume bottom-left = 1
→ 5 down: top-left = 5 − 1 = 4
→ 4 across: bottom-right = 3
→ 9 down: top-right = 9 − 3 = 6
→ 16 across: 6 + 4 + 6 = 16 … with two 6s
+-------+-------+-------+-------+-------+-------+
|#######|#######| \16 | 6 | 4 | 6! |
+-------+-------+-------+-------+-------+-------+
|#######|#######|#######| \4 | 1? | 3 |
+-------+-------+-------+-------+-------+-------+
Step 3: Spot the contradiction. The sum is right, but the 16 across now holds two 6s. The ghost was the 6 placed at the start of that row back when you solved the top. Nothing in the 2×2 corner pointed at it. It sat one cell outside the pattern.
So the bottom-left cell is not 1. It’s 3, and everything else is forced: 2 above it, 1 beside it, 8 above that. Check the 16 across: 6 + 2 + 8 = 16, no repeats.
+-------+-------+-------+-------+-------+-------+
|#######| 17\ | 4\ | 13\ |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \17 | 9 | 3 | 5 |#######|#######|
+-------+-------+-------+-------+-------+-------+
| \11 | 8 | 1 | 2 | 5\ | 9\ |
+-------+-------+-------+-------+-------+-------+
|#######|#######| \16 | 6 | 2 | 8 |
+-------+-------+-------+-------+-------+-------+
|#######|#######|#######| \4 | 3 | 1 |
+-------+-------+-------+-------+-------+-------+
You never wrote the 1, never wrote the 6, and never erased anything. That’s the difference between a forcing chain and a guess.
There’s also a faster, more controversial route. Some solvers argue straight from uniqueness: “if this corner could be a true swap, the puzzle would have two solutions, so it can’t be.” That reasoning only holds because the setter promised one solution, and purists consider it a shortcut rather than a proof. It’s your call. The chain above works whether or not you trust the setter.
Troubleshooting Common Issues
I lost track of the chain halfway through.
- Cause: Too many links, or a link that wasn’t actually forced.
- Fix: Start again from the same cell, saying each link out loud (“4, so 1, so 3, so 6”). If any link has two outcomes, abandon the chain and pick a different starting cell.
My chain ended without a contradiction, so I placed the starting digit.
- Cause: A peaceful chain proves nothing. It only means that branch didn’t break yet.
- Fix: Erase it. Then run the chain from the other candidate. If that one collapses, the first candidate is correct, for a real reason this time.
Parity told me a cell was odd, and it turned out even.
- Cause: You applied the two-cell rule to a longer entry. A 9 in three cells can be {2, 3, 4}, one odd digit, or {1, 3, 5}, three.
- Fix: For three or more cells, parity only fixes whether the count of odd digits is odd or even, not which cells hold them.
Conclusion
Key takeaways:
- A valid Kakuro never requires a guess. When two fills look right, one of them breaks a rule you haven’t checked.
- Parity reveals the shape of a stuck corner quickly: odd two-cell clues mean one odd and one even digit, and a 4 or 16 means both odd.
- Before committing, look for the downstream duplicate. Ghosts usually sit just outside the cells you’re staring at.
- A short forcing chain that ends in a collision is a proof. A long one that ends quietly is a guess.
Next steps: Pick up an expert-level puzzle and make one rule for the whole solve: no digit goes on paper unless you can say why it’s there.
Keep reading in this series:
- Previous: Intersection Logic: Cross-Referencing Kakuro Clues
- Start of the series: The Ultimate Guide to Advanced Kakuro Tactics