How an Escape Grid level is built and checked

People assume a puzzle game's levels are drawn by hand and tested by playing them. Half of that is true here. The first twenty boards in Escape Grid were placed by hand, one new idea per level, because the opening levels have to teach the movement rules and a generator does not know what a rule is. Everything after that is generated and then filtered hard, and the filtering is where the quality actually comes from.

Step one: generate something plausible

A candidate board starts with the target block and its exit, then adds blockers under constraints: pieces must fit, may not overlap, and the density has to land in a band appropriate for the level number. This produces boards quickly and most of them are worthless — either trivially open or completely stuck.

Step two: solve it exactly

Each candidate goes to a breadth-first search over board states. Because pieces slide along a single axis and nothing is random, the state space is finite and BFS returns the genuinely shortest solution rather than a good one. This is the step that matters: the solver either proves the board is solvable and tells us in how many moves, or proves it is not, in which case it is discarded immediately. No board reaches a player without that proof.

The shortest solution length becomes the level's par, which is what the star rating compares your move count against. It is also what the hint button reads from, which is why a hint gives you the next optimal move rather than a guess.

Step three: throw most of them away

A solvable board is not automatically a good board. Three kinds get rejected. Boards whose solution is mostly pushing the target forward, because they teach the wrong instinct and the game's whole difficulty comes from moving other pieces first. Boards whose par is far from the target length for their position in the progression, because difficulty has to rise smoothly. And boards with several very different solutions of the same length, because they feel arbitrary to solve — the satisfying ones have one route that can be reasoned out.

Why difficulty is not the same as length

Level 50 has a thirteen-move shortest solution. Level 100 has a ten-move one and is far harder. Length is a weak proxy; what actually makes a board difficult is how deep the dependency chain runs — how many pieces have to move before the one you care about can — and how few legal moves exist at each step. A tight board with five walls and ten moves gives you almost no choices, and finding the only plan is harder than executing a long one.

What the mechanics are for

Walls arrive first because they change what an empty cell is worth: space only counts if a piece can legally reach it. Locks and keys come next, and they add a required order rather than a preferred one. Ice tiles are the cruellest addition, because they take away your ability to stop a piece where you want — you have to plan a move to its endpoint. Chained blocks, in the last world, remove the assumption that you move one piece at a time.

When a level is wrong anyway

Occasionally a board passes every filter and still feels unfair to play. That is a design judgement a solver cannot make, and the only way I find out is when someone reports it. Several levels have been rebuilt off the back of a single message saying a board felt fiddly rather than hard. If you hit one, say which number it was — the solver can be re-run against it in seconds.