I set out to prove emergence, I made a game

Limon is a world of proto-humans who live without me. On screen: a village growing by a river, a lineage passing knowledge down through generations, a neighboring civilization collapsing during a drought. They gather, they settle, they invent techniques, they wage war. I do not touch a thing.

My ambition was not to make a game. It was to prove so-called "open-ended" emergence: behaviors I had not programmed, appearing on their own, and above all measured against a null model, meaning against pure randomness. Not emergence you narrate in a blog post. Emergence you demonstrate. The map, the sprites, the interface, were almost just scaffolding for the measurement.

I proved generic Darwinian selection: an ordinary evolution engine that acts on any inherited trait, not a system rigged for a single case.

To check it, I broke heredity. In a control version, each child gets a genome drawn at random instead of its parents', and I compare the two worlds. The result: the adaptation passed down from generation to generation is about six times stronger when heredity works than when it is random. Transmission really does something. It is not noise.

But there is a trap, and it took me a long time to see it. Proving that selection acts on a trait I wrote myself is not proving that something emerges. It is proving that my code does what I told it to do. A tiny nuance on paper, a chasm in reality.

I deliberately put myself on trial. I ran a dozen adversarial reviews against my own plan, tasked with demolishing it rather than confirming it.

The verdict inverted my framing. I thought I had an execution problem: run this experiment, lift that limit, enlarge the map. The problem was in the architecture itself. Two flaws.

The first: my factor of six measures adaptation to traits I chose, not invention. A technology tree whose effects I write myself stays a catalog I fill by hand. The truly open systems, the artificial-life projects of the 90s like Tierra or Avida, leave the program free to build things nobody anticipated. I had confused filling a catalog with opening a space. Adding a hundred more tech cards makes a better management game, not a more open world.

The second: my only way to make a world die was to kill it whole, all at once. But to measure a strategy, the world has to survive long enough. So every time I had something to measure, it meant death had eliminated no one. Impossible to win that bet with pressure that blunt.

The question underneath it all: am I building a game, or am I chasing a proof, accepting in advance that it might fail?

I answered: a game. I removed "emergence research" from the project description, archived the design brief that carried that ambition, rewrote the store page copy.

The price is real. I spent months chasing a proof my architecture could not produce. Those months I do not get back, and no one refunds the detour. The only thing I take from it is that I stopped lying to myself about what I am actually building.

The world still runs, deterministic, indifferent. It discovers nothing. Neither did I, this time around. What remains is a game: a god-game, one of those games where you watch a world from above without steering it directly, contemplative, carried by real generic Darwinian selection.

I now judge it by a single measure: is it enjoyable to play? Never again: does it prove my thesis? It is a smaller horizon than the one I aimed for. It has the advantage of existing.

Emergence you narrate is not emergence you prove. I finally let mine go. The world did not notice.