A word ladder looks like a word game and is really a graph problem wearing a costume. Each legal word is a node, each one-letter substitution that produces another legal word is an edge, and a puzzle is a pair of nodes with a path constraint. Once you write it that way, most of the interesting product decisions become questions about the graph.
The first decision is the word set. A ladder built on the full dictionary has nodes nobody can think of, which makes puzzles feel unfair rather than hard. A ladder built on a common-word subset has a different problem: some targets become unreachable. The practical solution is a curated core list plus a reachability check at generation time. If the target is not in the same connected component as the start within the budget, the puzzle is discarded.
The second decision is the budget. The shortest path length between start and target is a lower bound, but a fair budget is not the lower bound. Human solvers do not play optimally, so a par that equals the shortest route makes every board feel punishing. Adding one or two moves of slack changes the distribution of results dramatically, and the data is easy to collect: record move counts against par for a few hundred anonymous solves.
The third decision is the hint ladder. Hints should expose rungs of the shortest path in order, one at a time, with no forward preview. That structure is a product of the graph, not an afterthought. It gives the player a real choice about how much information to buy while keeping the remaining route a genuine search.
The graph model also tells you where puzzles get boring: dense regions are forgiving and short, sparse regions are punishing and long. The daily puzzle keeps to the dense core on purpose, and the unlimited mode lets players wander into the sparse corners when they want a harder board.
Top comments (0)