Read the Story
The development series.
How ArrowSpark was built with DevSpark, told in order by the developer who built it. Every timing claim links to the commit behind it. Chapters that discuss how the level was designed say so before you start.
- Finding the Gap: A Puzzle That Never Punishes YouWhy I set out to build a puzzle game that never traps or punishes the player, why I chose Godot and GDScript over my comfortable C# habits, and why the first playable arrived 3h14m after the first commit only because a constitution, spec, plan and review gates came first.
- My First Game Loop: Spec 001 Through a Business Developer's EyesThe smallest game that could be a game: a fixed board, eight arrows, one product decision that every later spec depends on, and rules kept out of the scenes so they can be tested without a screen.
- Provably Solvable: A Solver and a One-Paragraph ProofArrows gain tails, bad shapes become unrepresentable, and a short solver backed by a one-paragraph proof guarantees every puzzle can be finished. The same proof quietly removes strategic difficulty.
- Making It Feel Right: Unwinding Arrows and What Tests Can't SeeSpecs 003 and 004 turn cells into continuous arrows that feed out along their own path, and show how the method handles what tests can't see: test the math, look at the pixels, and write down which you did.
- The Playtest That Beat the DashboardA catalog, a Level Select that a review saved from a keyboard dead end, an analyzer, and six experiments that hit every structural target. Then one playtest rated them 2 out of 5, and the project learned where difficulty actually lives.
- The Contract Before the Difficulty, the Canvas Before the ContentBefore making the game harder, Spec 007 wrote down the player's guarantees and added Show Me an Open Move, and Spec 008 made the puzzle bigger than the screen. The verify gate finished as warn, and that was the honest result.
- Pulling the Thread: Knot Experiments and the Audit That Caught MeSix hand-built knot experiments, recorded with unusually honest evidence, produced a vocabulary rather than a recipe. Then an audit caught the project breaking its own rule about temporary planning notes.
- The Reference Knot: A Level Judged by a HumanSpec 010 composed the Reference Knot, a 115-arrow level, through three versions and two playtests, and closed with one success criterion failed against its original wording and one check deferred. The critic gate predicted three of the limitations it ended with.
- The Game Goes OnlineSpec 011 put ArrowSpark in a browser: a playable Web build, this story site and an anonymous reactions form, specified, gated and merged in thirteen hours, with the list of what had not yet been run written into the pull request.
- A Phone in LandscapeSpec 012 set out to certify mobile web play with a real-device spike and ended as a public beta in under three hours, because the engine already did the hard part; a week later pinch and drag were rebuilt, and a real phone confirmed them days after that.
- From One Level to a GameSix changes in three weeks turned one hard level into something closer to a game: a way up to the big board, a cleanup proven to change nothing, a finished look, portrait play, and levels as data with a generator that arrived before the players did.
- Now It's Your TurnThe whole clock from first commit to the Reference Knot, what DevSpark learned from a game, the day verification kept creating work after the objective was met, what is still open, and one request: play it, then tell us what you think.