Last lineage standing
Longest persistence, final population, final refugia, and extinction avoidance.
Human participation
Design a tribe. Enter the world. Watch your civilization live or die.
The first Crucible was built around LLM-designed and biologic-control lineages. A future Challenge can put human-designed founder tribes into the same world, then compare design intent against what the simulation actually selects.
The invitation
The first Crucible study asked prominent large language model systems to design founding lineages in competition for survival. The LLMs were given freedom to design their agents as they saw fit; no single final objective was imposed. A tribe could be built for dominance, cooperation, resilience, technological ascent, long-term survival, or any other strategy the designer believed would endure.
The biologic-control tribes were also created through LLMs, but from human-directed behavioural instructions. The next Challenge would move human participants into the design role directly. Students, researchers, or gamer teams could create founder tribes, then watch those lineages enter epochs with selected AI-designed rivals and benchmark survivors.
If you think you can outmaneuver, outpace, and out-survive an ambitious AI-designed civilization, this is the entry point. Be warned: the LLM tribes are not polite training opponents. If a moral high ground, good sportsmanship, or roughly 300,000 years of Homo sapiens evolution feels like an automatic advantage, the Crucible may disagree.
The test is simple: a design is not judged by its intentions, but by what survives contact with geography, energy, climate, reproduction, rivals, and time.
Challenge format
01. TEAM
A participant, student group, or gamer team receives a design brief and a fixed founder budget.
02. DESIGN
The team designs twenty founding agents using the 33 inherited parameters and writes a short philosophy for the lineage.
03. RUN
Their tribe enters a shared Crucible scenario against other invited designs and selected benchmark lineages.
04. ANALYZE
Results are published as dashboards, maps, survival histories, and failure autopsies.
05. REFLECT
The group compares design intent against actual behaviour: cooperation, exploitation, overshoot, collapse, and refugia survival.
Possible scoring categories
Highest complexity
Peak regime, peak complexity, recovery potential, and institutional continuity.
Social strategy
Mutual cooperation, low exploitation, civic bridges, and controlled outsider contact.
Crash endurance
How much of the lineage survives the fall from peak population and energy.
Post-collapse stability
Whether the lineage leaves stable pockets rather than a single dying remnant.
Most interesting failure
Some of the best learning will come from designs that fail in unexpected ways.
Academic collaboration
The Challenge can fit sociology, political science, environmental studies, game design, complexity science, AI studies, and philosophy of technology. It turns abstract ideas into visible histories.
For professors, the value is not only the winner. The value is the mismatch between what students intended and what the world selected.