Develops many random rules without displaying them and ranks them by the chosen goal (only rules that reach an attractor are ranked). Each rule runs from the seed graph until it reaches an attractor, has zero nodes, would exceed the size limits, or hits the step limit.
Evolutionary experiments run here in the browser, on a pool of workers. An experiment is a JSON config: the rule shape (states, hidden layers), the seed graph and evaluation limits, the fitness, the strategy (microbial GA, generational GA, MAP-Elites or random search), its operators and the evaluation budget. A list of configs runs one after another, each with its repeats.
Click a run to view it above. Per run: JSON (config, environment, histories, snapshots, top individuals and notable genomes), JSONL / CSV (one line per evaluated candidate), and for DGCA runs a PDF report (figures and tables of the run and its best rule). Runs are kept until the page is closed.