What Makes a Puzzle Hard? Search, Outcomes and Hidden Rules

A puzzle is hard when the set of possible answers is huge, when each test tells you very little, when the shortest route to the goal is long, or when the rule that rules options out is hidden from view. Mathematical papers on weighing coins, sliding tiles and packing squares give exact numbers for the first three of these sources ^{[1] [2] [3]}, chessboard papers do the same for the fourth later in this article, and this article uses those numbers to show that size on its own is rarely the whole story.

Rather than counting dead ends, this article measures difficulty in four other ways, one section each, and ends with a table that puts them side by side.

Is a puzzle hard just because it is big?

Size matters, and it grows faster than intuition expects: a paper on shapes that can be cut into smaller copies of themselves found that the number of possible tile designs in a square jumps from 784 to 29,611,120 when its grid grows from two by two to three by three ^[4].

Puzzles built on this kind of growth cannot be solved by trying everything in your head. Even a computer needs a plan, and the plan is what decides whether the search ever ends.

If size is not enough, what makes the search shorter?

Rules that cut away whole branches at once. The 70 square puzzle, which asks whether the squares with sides 1 to 24 can exactly fill a square of side 70, is a search far too big to finish by hand, and a paper published on 25 September 2018 approached it by first proving rules about what can never happen ^[3]. One of those rules says that none of the squares with sides 1 to 5 can touch the edge of the big square ^[3].

Those rules changed the scale of the work ^[3]. A first, looser list of ways to fill one edge of the big square had 9,285 rows; with all the restrictions of the theorem applied, only 391 rows were left ^[3].

This is the first lesson about difficulty. A puzzle feels hard when you can see no rule that removes many options at once. It feels easy when such a rule exists and you spot it. The number of possibilities is the same in both cases; what changes is how quickly you can throw most of them away.

How does the number of possible outcomes limit a solver?

Some puzzles let you ask questions, and difficulty then depends on how much each answer can tell you. Coin weighing is the cleanest example.

A balance has only three results: the pans stay level, the left pan goes down, or the right pan goes down ^[1]. A plan of several weighings therefore produces a string of these three symbols, and two different fake coins must never lead to the same string ^[1]. So the number of coins that a plan of w weighings can handle is at most 3 to the power w ^[1].

That ceiling explains why some versions feel much harder than others. If the fake coin is known to be lighter, every string can point to a different coin. If it may be lighter or heavier, each suspect uses up two strings, one for each case, and the limit drops to roughly half ^[1]. The classic version, with 12 or 13 coins and 3 weighings, sits at or just below that reduced limit ^[1].

A paper published on 26 June 2025 pushed the idea further with coins that change state each time they are placed on the scale ^[1]. One such coin alternates between light and normal, so after it shows up light, it behaves normally on its next trip to the scale ^[1]. Two unbalanced results can therefore never come one after another, and the number of usable strings shrinks to a much slower-growing sequence ^[1].

The lesson carries well beyond coins. When a test can only give a few different answers, you need more tests, and every wasted test hurts. A hard deduction puzzle is often one where each clue or each test narrows the field only a little.

Why do long solutions make a puzzle harder?

A puzzle can be small and still hard if the shortest solution is long, because every extra step is another place to drift off course.

Sliding tile puzzles show how fast the length grows. Every position of the small tray with three tiles can be solved within 6 moves, but the eight-tile tray needs up to 31 moves and the fifteen-tile tray up to 80 ^[2].

Even a simple method that does not look for the shortest route needs more and more moves. A paper published on 10 July 2015 showed that solving one row and one column at a time takes about eight thirds of n cubed moves on average for a tray with n tiles on each side ^[2]. The author checked this on 10,000 random positions for each tray size from 4 to 200 ^[2]. Doubling the width of the tray therefore multiplies the work of this method by about eight.

Can a few obstacles make a puzzle longer?

Yes, and one chessboard paper measures it exactly ^[5]. On an empty board, a rook can reach any square in at most 2 moves ^[5]. A paper published on 7 December 2021 asked how many blocking pawns are needed before some trip takes longer ^[5].

The answer starts small. A single pawn placed anywhere except a corner makes some trip need 3 moves, while a pawn in a corner leaves the longest trip at 2 moves ^[5]. More pawns stretch the longest trip further, as the computed values for a four by four board show ^[5].

Notice that the table does not climb evenly: the third pawn can stretch the longest trip to 5 or to 6 moves, depending on where it goes ^[5]. Placement matters as much as number, and as the table shows, a few well-placed obstacles can double the longest trip ^[5].

What are hidden constraints, and why are they so powerful?

A hidden constraint is a rule that follows from the stated rules but is never written down. The solver has to discover it, and until they do, the puzzle can seem hopeless or even impossible.

Knight's tours inside a cube are a striking case: in cubes with an odd side length there is never a closed tour, because every knight move switches between two colour classes of cells and the two classes have different sizes ^[6]. Nothing about the move itself hints at that rule; it only appears once you colour the cells.

Queen placement puzzles show the same effect. You can place n queens on an n by n board so that no two share a row, column or diagonal when n is 1 or any number from 4 upward, but not on boards of side 2 or 3 ^[7]. Nothing in the rules warns you about the small boards; you only learn it by trying or by proof.

Can a hidden constraint make a puzzle easier instead?

It can, because a strong constraint often forces your moves. A paper published on 8 December 2016 looked at boards where pawns may block the queens' lines and as many pawns as needed may be used ^[7]. On an odd-sided board, fitting the largest possible number of queens requires a queen in one particular corner square; the paper notes that this forces the placement of all the other queens, so for odd-sided boards there is only one way to place them ^[7].

For a solver this is the best kind of hidden rule. Once you notice it, the search collapses to a single line of reasoning. The edge rules for the 70 square work in a similar way: they do not just say what is impossible, they tell you where to look ^[3].

So hidden constraints cut both ways. Unseen, they make a puzzle feel impossible. Seen, they make it feel easy.

How do the four sources of difficulty compare?

Most hard puzzles combine two or more of these. A sliding tile tray has a large search space and long solutions. A weighing puzzle has a modest number of suspects but very few outcomes per test. A knight's tour on an unusual board adds a hidden constraint on top of a large search.

How can you use this when a puzzle feels hard?

Name the source of the difficulty first. The right response depends on which one you face.

• If the search is huge, stop trying cases and look for a rule that removes many of them at once, such as a piece that cannot go on an edge.
• If tests tell you little, plan each test so that its possible results are as evenly balanced as you can make them.
• If the route is long, break it into stages, fix one part, and treat the rest as a smaller puzzle.
• If it seems impossible, look for a hidden invariant: colour the board, count something that never changes, or find the one forced first move.

These habits do not make hard puzzles trivial, but they turn a blind search into a guided one.

Try them on this number riddle, where the clues together remove every candidate but one: