The dictionary is the genome. The code is the organism.

What you just watched

There is no separation here between the program and the thing being optimized. In unit, a Forth word — a named sequence of instructions in the dictionary — is a gene. To mutate the organism is to edit its code; to run it is to express its phenotype. Selection acts directly on the source.

That collapse is the whole idea, and everything else grows from it: a unit can inspect itself (SEE), mutate itself (SMART-MUTATE), share a discovered word with neighbours (SHARE-ALL), and spawn a child that inherits the entire dictionary (SPAWN). When a real unit can't solve a problem, it registers the failure as a challenge and broadcasts it; every unit on the mesh evolves in parallel, and the first verified solution installs as an antibody (SOL-*) that children inherit. The arena above is that loop, shrunk to one machine.

• ExecuteForth VM — stacks, a dictionary, an inner interpreter that defines new words at runtime.
• CommunicateS-expressions over UDP — units gossip, discover peers, and broadcast in a form any language can parse.
• ReplicateSPAWN — a unit reads its own binary, packages its state, and forks a child that boots with the parent's genome.
• MutateGenetic programming — tournament selection over a population, five token-level operators: swap · insert · delete · replace · double.
• PersistJSON snapshots — hibernate and resurrect exactly where it left off.