Your first evolution experiment
Step by step: seed a species with mutations on, let it run for tens of thousands of cycles and learn what to look at in Observe, Analyze and the inspector to tell selection from drift.
In A bot that looks for food you wrote a bot that makes a living, and in A bot that recognizes its own species, one that gets along with its own kind. This tutorial is different: it doesn't end in a new bot. It ends in a running experiment and in knowing what to look at while your species evolves.
Setting up the experiment
You need four things: a seed species that knows how to make a living, enough food, mutations switched on and time. Everything is set up in Experiment.
The seed species. Any of these will do:
- the final bot from A bot that looks for food, if you followed it;
- the Bestiary's Animal_Minimalis_mod_stress: it reproduces with
.reprowhen it is doing well and with.mreprowhen it is losing energy (Mutations); - any simple bot you know eats and reproduces: with no births there are no mutations to inherit.
A small one that you understand is best: when a descendant changes, you will be able to read what changed.
The steps:
- In Home, find the Primordial soup scenario and click Tweak: it opens in Experiment already loaded. It is the right template: a world with no costs, 15 algae, 5 animals and mutations switched on.
- Under Species to seed, remove the animal with the × on its row and add your seed with Add species: A bot from the library (by name) or Paste DNA. Leave the algae: they are the food. 5 seed bots are enough; every species you add from here starts with 3000 energy.
- Check that Mutations says Yes. And don't touch anything else: the app seeds each species with its mutation table on and the factory rates, so the switch is enough.
- Set the Seed and click New simulation. The app builds the world, sets it running and takes you to Observe.
Running it and watching
Set the speed to Max: evolution shows up over tens of thousands of cycles, not hundreds. Then, three places.
Observe (Observe). The Population by species chart and the Max. generation card tell you the basics: whether there are births (the generation goes up) and whether the population holds. The Events list notes the record generations. And the Color by selector changes the tint of the bots: Mutations paints by how many each one has accumulated, Generation by its genealogical age, and Genetic distance by how much it resembles the bot you have picked.
Analyze (Analyze), with the run live. Four tabs are useful for this:
- Species: one row per species, with the average of Mutations, the Max gen., the Mean DNA and the Offspring per bot.
- Phylogeny: the tree of individuals of your species, ordered by generation. Click a bot and the card gives its parent, its mutations, its DNA length and its offspring.
- Genetics: histograms of the population (DNA length, generation, mutations, offspring) and the most direct view of all, Dominant DNA vs founder: every 1000 cycles the app takes a snapshot of the DNA carried by the most bots and compares it gene by gene with the founder's — which genes stay the same, which changed, which are new.
- Events, and the Dashboard's Findings: a dominance, a replacement or a collapse, written out in words and with their cycle.
The inspector (The inspector). When you see a bot with a shifted color, click it: the header gives its generation and how many mutations it has accumulated, and the DNA tab shows its program as it is now, changes included. Copy takes it all with you; Family highlights its descendants in the world.
What to expect
With the factory rates, in a 20-instruction bot we measured that roughly 1 in every 60 children conceived with .repro is born changed, and almost 1 in 5 with .mrepro. It is a measurement, not a law: the proportion depends on the DNA length and on the rates. And evolution happens almost entirely through births: in life, almost nothing changes. The numbers are in Mutations and Mutations.
The rest is biology:
- Almost every mutation breaks something. Mutants show up one at a time, and most of them live shorter than the founder: a deleted
start, a shifted number, a gene that stops switching on (Mutations). It is normal for nothing visible to happen in the first tens of thousands of cycles. - A mutant that thrives is recognized like this: its descendants fill the population, the dominant DNA in Genetics changes from one snapshot to the next and, in Phylogeny, its branch is the thick one. The change can be small: a lower shooting threshold, a shorter turn.
- Color is your first indicator. Each mutation shifts one channel of the species color a little, so an evolving species turns into several shades: each shade is, more or less, a lineage.
- Drift or selection. A change sticking around doesn't mean it helped: in small populations, a neutral lineage can stay by pure luck. The test is to repeat: in Compare, the Replicates run the scenario again with other seeds. If a similar change wins in most of them, it is selection; if each run tells a different story, what you saw was drift.
If the experiment dies out
Three signs that something isn't working: the animal population falls and never picks up again (the Findings write it up as a collapse); the maximum generation doesn't move — there are no births, and without births there is no evolution —; the vegetables disappear and the world runs out of energy.
What to adjust, all in Experiment with Apply to current, without starting over:
- Food: raise the Solar energy (Solar share per cycle) or the Vegetable repopulation (Repopulate below (chloroplasts)), or enlarge the Vegetable cap (Vegetable cap).
- Costs: if you raised them, try again with No costs; a world that starves everyone lets nothing evolve.
- Starting energy: in Observe, Seed lets you choose the energy new bots come in with: seed a few with more energy to relaunch the population.
- Time: most experiments that “don't work” do work, just not yet.
Variants
- Exploring faster. Swap the
.reprofor.mreproin your seed (or use the Animal_Minimalis_mod_stress, which already does it when it is in trouble): each child is born with ten times the chance of mutating, just for that birth. You will see mutants right away — and also their price: most are born broken, and the mutations histogram in Genetics shifts to the right. - Sexual reproduction. With
.sexreprothe child is already born different because it mixes the DNA of two bots, even if nothing mutates; and when two branches drift far apart, they can no longer cross — genetic distance decides (Reproduction, Species and lineage). - A hostile world. The factory world is kind. Push: charge Costs (the basic control in Experiment), switch on Day and night — at night vegetables don't make energy and eyes see 20% less —, lower the vegetables or shrink the field. Each pressure pushes evolution somewhere else: with costs, the ones that spend little; with night, the ones that endure the dark.
And now
- Save the run (Save, in Observe) and build a report with its findings: Reports.
- If you like a descendant, copy its DNA from the inspector and save it as a bot of your own in Bots: library and profile; the DNA editor helps you read what changed.
- Before drawing conclusions, repeat with other seeds (Compare › Replicates).
- If something didn't go the way you expected, the list is in Frequently asked questions.