PETRI DISH GAMES

Emergence

The rule every Petri Dish game is built on: an outcome is produced by running a population and selecting on it, never written down in advance.

Emergence is the production of an outcome by running a system rather than by authoring the result. Every game Petri Dish Games builds works this way: a population reproduces with variation, something outside that population scores it, and what survives is whatever the scores left standing. No code names the winner.

The rule holds at every layer of a game rather than at the top one. A layer whose result is faked cannot carry the layer above it, so each layer runs for real: cells run their own genome, a population is scored by a measure it does not control, and the score is what breeds.

Where it runs

GameThe populationWhat selects it
Substrate: EmergenceCells in a dish. Every division, bud and pinch builds the daughter from a re-rolled copy of the parent’s genome.The dish. A lineage whose tolerances do not match the pH, temperature and food where it sits stops dividing and dies there.
Rorschach LabShaders held as code, bred out of a pool by splicing two parents, grafting a fragment of one into the other, or jittering one parent’s constants.Rating. An organism’s fitness blends its private rating with the rating visitors’ picks produce, weighted by how many judgements stand behind each.
GliaOrganisms on a hex grid, and the patches they play, both genetic.Fitness from three terms: the energy an organism earned, the rating listener picks moved, and whether a listener kept it.
Teztlotzl ArenaThe neural nets that drive the AI roster.A scalar fitness computed from what a net did in a match, maximised by a (1+λ) evolution strategy. The shipped roster is what came out of those runs.

What is authored

Authoring sets conditions, never results. The authored parts of a game are the rules of the simulation, the shape of the genome, the terms and weights of the fitness function, the starting population and the ground it starts on. Which organism wins under those conditions is computed.

Two consequences hold across the catalogue.

  • A result a simulation could produce is not hand-written. Where a system can generate the outcome, generating it is the implementation, and the authored version of it is a debt.
  • Where a real signal exists, it is not simulated. A visitor’s pick in the Rorschach Lab and a listener’s pick in Glia move the same ratings the breeders read, rather than feeding a separate display value.

What the rule excludes

  • Scripted outcomes. Nothing decides in advance which lineage, shader, organism or net comes out ahead.
  • A damped result. A large mutation is not softened for being large: the wide step is taken as drawn and clamped into range, rather than scaled down as it approaches a parameter’s limit.
  • A simplified opponent. A contaminant is an ordinary cell carrying another side, running the same division and the same mutation pass as the player’s own, one step further. It is not an opponent model standing in for one.
  • A difficulty curve read off the player. Contamination fires on the richness of the substrate over the play region, and never on the player’s cell count or progress.

Bound to

The source this page's claims were read against, named by symbol and pinned by the digest of the file it lives in.

SymbolDigest
CreateGenomeForDaughter 1a9b0f0d6774
evolve.js b72f4e12e72f
fitness 193eb96a460a
fitness 64e4cc37de25
fitness 7ec1d39e0fa1
GeneMutationService fb1a82bdf1ec
Roster c3a17c9d4626

Near this subject

These are the subjects this one is joined to in the field. Step out of this card and they are the wells around the one you are standing in.