Darkwaar20 - From Noise

Darkwaar is an experiment in creating and publishing small playable ideas regularly.

Today's puzzle

Today's Darkwaar goes from noise to a world.

The board starts as noise. Five times, you choose Life, Order, or Void, and settle. The same seed with different choices becomes a different world. Then you walk what you made, from the core to the exit. A sequence that erases too much ends as a dead board.

How to play

• LIFE grows connected regions
• ORDER stabilizes patterns
• VOID erases unstable regions
• Choose one, then Settle. Five passes
• After the fifth, dark cells are void. Gray is ground. Orange is energy
• The cyan ring is the core. The pale square is the exit. You start on the core
• Tap the next open cell. Void blocks. Reaching the exit wins
• SAME SEED restores the noise, so you can try another sequence. NEW SEED moves on

Play on itch.io

Play in the browser: https://darkwoodcom.itch.io/darkwaar20

Screenshots

The core

Generative image models turn noise into a picture by denoising, step after step, under a guide. This puzzle keeps that shape and drops the network.

Diffusion goes from noise, through guided denoising, to an image. This puzzle goes from noise, through Life, Order, and Void, to a world, and then you have to cross it.

There is no neural network here. Each settle is a small deterministic rule on four integers per cell. The same seed and the same five choices always rebuild the same board. Life lifts cells that agree with their neighbors. Order lets a shared value take over. Void pulls isolated cells down. After five passes those numbers become void, ground, or energy. One core and one exit are taken from that field, not placed by a second roll.

A bad sequence is allowed to fail. Seed 20020 with Life, Order, Life, Void, Order leaves a path. The same seed with Void, Void, Order, Void, Life leaves nothing you can cross.

The useful kinship with constraint-based program synthesis is only that: a wide space, made specific by the choices you commit to. This puzzle does not search programs and does not run genetic programming. You spend five constraints, then you live in the result.

Development note

Seven by nine, four channels, five integer passes.