Work / Living Society
Living Society
A seeded, repeatable society simulation used as an experiment bench, including the experiment that showed one of its own features made things worse.
What it does
Agents live in a seeded world, gather resources and interact, and any run can be replayed exactly. Multi-seed runs, scenario matrices and ablations test the design's own assumptions. One ablation found that switching the communication system off gave about 7.4% more population after 4,000 ticks. The project records that as a finding instead of tuning it away.
- Pick a seed and a scenario
- Agents gather, trade and interact through bounded decisions
- Watch the world and click any agent to inspect it
- Replay any run exactly from its seed
- Compare runs across seeds, scenarios and ablations
How it works
Living Society is more interesting as an experimental system than as a game. Agents make bounded decisions about needs, resources and relationships. The world is drawn on a canvas with click-to-inspect, and every run records its seed and telemetry. The test suite combines deterministic checks, scenario matrices, long-horizon runs, visual regression and performance proofs. An audit showed that several named policies were labels over the same behaviour, so the limitation is reported rather than presented as policy variety.
Skills shown
- AI, agents & ML
- Agents learn with tabular Q-learning, evaluated across multiple seeds, scenarios and ablations.
- Testing & QA
- Deterministic, scenario-matrix, long-horizon, visual-regression and performance tests, run with Node and Playwright.
- Frontend, 3D & graphics
- A Canvas world view with click-to-inspect agents.
Worth knowing
- Exact replay from a seed, so any interesting run can be reproduced.
- 276 tests across 14 suites at the documented checkpoint, plus visual-regression and performance proofs.
- A long-horizon run of 10 seeds of 4,000 ticks each succeeded on every seed.
- An audit found that several named policies were labels over the same behaviour. The project reports that limitation rather than claiming variety.