The Computology interaction suite · edition 01
Small rules. Unbounded curiosity.
A living path from a single cell to a learning machine. Change the inputs, follow the computation, and see what happens next.
13 experiments to take apart. No installation. No finish line.
Enter the workshop ↓Your computation workbench
Connect the dots.
Choose a guided circuit, or wire a node’s gold output to a blue input. Step the graph to pass real values through each connected experiment in order.
No wires yet. Each saved node remains a complete, independent snapshot.
One instance per module in this edition. No implicit resampling or thresholding. Whole workbenches export as JSON; account saves store individual nodes, including their resolved input state.
Opening the workshop…
Read the computed values as a table
Change the conditions
Applying keeps the current state. New dimensions or architecture require a fresh start. Reset returns to the last loaded or saved snapshot.
What is happening?
Try a different question
Inspect or edit the exact state
This is the real simulation state. Arrays, weights, kernels and graph connections are editable data. Export a copy before a large change.
Keyboard cell editor
Row and column start at zero. The left panel is editable when two panels are shown. Use value 0 to erase.
Wire another experiment into this node
Add a persistent connection and copy its current output now. Use Step connected nodes in the workbench to update the full circuit. Shapes must match; invalid connections explain the mismatch.
Keep the interesting bit.
Save privately to your Flaptown account, or export a file. Publish a saved revision when you want it on your public profile.
Checking account connection…
Your saved experiments
Measurements, limits and credits
Runs use JavaScript numbers and a saved xorshift32 random state. The same engine version, parameters and state can reproduce a run; floating-point math may differ slightly between browsers. History is capped at 512 samples. Training and simulations run in a cancellable worker. No result is a claim of scientific novelty.
The learned garden is an original small educational implementation inspired by Growing Neural Cellular Automata. Further reading: automatic differentiation and momentum.