Overview
Doublets revives the word-ladder puzzle invented by Lewis Carroll and published in Vanity Fair in 1879. Players begin with one word and reach another by changing a single letter at each step. Every step must be a real word, and a shorter chain earns a better result.
The rules take only a moment to understand, but the possible routes reward vocabulary, logic, and experimentation. Four new puzzles arrive each day, ranging from easy to legendary, so a player can choose a quick success or a deeper challenge.
Doublets is a collaboration between Manley Studios and AstroBytes, the software engineering company founded by Porter McGary. Porter and I work together on the product, game experience, and implementation. The game is in active development.
The partnership is an example of how two independent engineering firms can bring different experience and judgment to a shared product. We were able to work as peers across both product decisions and implementation.
The challenge
A word ladder looks simple because the rules are small. Producing a steady supply of good puzzles is a data problem. A raw dictionary can show which words exist, but it cannot say which ones make a satisfying game. Some pairs have no path, some depend on poor intermediate words, and others produce routes that are valid but not enjoyable.
The work begins by finding possible doublets, then curating them for actual play. Holiday puzzles add another dimension: the endpoints should feel timely without sacrificing the quality of the route between them.
Difficulty also needed to be engineered rather than guessed. We developed our own metrics and evaluation process, then used them to tune puzzles within each level and across time. Because Doublets is a daily game, a puzzle has to work on its own and as part of a longer sequence. The four levels need to remain distinct, while the daily mix needs enough variety to stay interesting.
The approach
We built a data pipeline that turns a raw dictionary into a curated puzzle system. It identifies possible connections, finds viable doublets, evaluates routes using our internal difficulty model, and gives us tools to review and schedule the strongest puzzles. The details of the evaluation are proprietary, but its purpose is straightforward: make each difficulty level feel consistent without making every puzzle feel the same.
The pipeline also generates word graphs for fast lookup during play. Doing that work ahead of time gives the app quick access to possible routes and nearby words. Those graphs support player-facing features such as powerups without requiring expensive searches at the moment someone uses them.
The mobile app is built with Flutter and backed by Firebase. This gives the shared daily game a small, focused client while keeping puzzle delivery and product data consistent. A separate Vue website explains the history and rules, includes a playable tutorial, and directs players to the app.
The result is a game that feels simple at the surface because the data preparation, evaluation, and graph work happen behind it.
What we’re building
- A Flutter mobile game built by two collaborating engineers
- Four new daily puzzles across distinct difficulty levels
- A pipeline for turning raw dictionary data into playable doublets
- Curated daily and holiday puzzle sets
- Proprietary difficulty metrics and evaluation
- Tuning within difficulty levels and across the daily schedule
- Precomputed word graphs for fast lookup and player powerups
- Firebase-backed puzzle delivery and product data
- A web experience with an interactive tutorial and game history