Cogworks

Route the thoughts. Build understanding.

Cogworks is a tile-based puzzle game about guiding mechanical thought tokens through a clockwork board. Players place and rotate routes, then run the machine to see whether the tokens reach an output in the required order.

It is also the current game project at AiD10t Games, my experiment in human-directed development with specialized AI agents.

The idea gives an abstract logic puzzle a small, readable fiction. Tokens such as “Dog,” “Animal,” and “Barks” become pieces of an ordered thought. Connecting the board is only part of the solution; the machine also has to deliver its ideas at the right time.

Depth from a small vocabulary

The playable campaign contains 20 levels across four chapters, built around three route tiles: Straight, Turn, and Arrow Router. An arrow accepts entry from behind and sends a token in its indicated direction. That restriction gives direction a meaning beyond the shape of the path.

The campaign develops those rules through missing connections, incorrect arrows, open routing areas, and arrival-order problems. One early puzzle launches three tokens simultaneously, so route length determines which reaches the receiver first. Later boards combine familiar rules into longer circuits instead of continually introducing new tile types.

That is the central design interest: how much variety can a compact set of rules support when geometry and timing interact?

Let players inspect their reasoning

The prototype includes placement previews, rotation, erase, and a one-step undo. Running a solution is deliberately reversible: Reset Simulation clears the moving tokens while preserving the player’s routes; Reset Board restores the authored layout.

Pause and 1×/2× speed controls make the machine easier to study. Tokens and spawners share the same simulation clock, so changing speed preserves their timing relationship and pausing freezes both movement and delayed arrivals.

Failure feedback points to the reason a route broke. An empty cell, an incompatible entry, a backward arrow, and an incorrect output order are different mistakes. Giving them distinct explanations helps the player revise a hypothesis rather than rebuild blindly.

Authoring and validation belong together

Each puzzle is a Godot Resource containing its board, spawners, expected output, and an authored reference solution. The board, simulation lifecycle, output validation, and UI have separate responsibilities connected through signals.

That structure makes the campaign practical to extend and check. The repository includes tests that apply reference solutions through the real simulation, alongside checks for tile behavior, level resets, pause, and speed. The reference routes also document the design intent behind each puzzle.

Where the prototype stands

Cogworks has a playable campaign, a paged level selector, and a clockwork visual treatment. Its current limits are intentionally modest: undo has one step, audio uses simple synthesized tones, and tokens do not collide or queue behind each other. Persistent progression, broader accessibility work, and richer presentation remain future work.

The project brings together puzzle design, readable feedback, data-driven level authoring, and simulation validation. Its strongest idea is that a machine can teach the player its rules simply by making the consequences of each route visible.