An "influence-not-control" evolution sandbox where players guide migratory micro-bird flocks using environmental signals. Agents learn across seasons via lightweight ML.
# Install development environment
make dev
# Run smoke test
make sim-smoke
# Run tests
make test
# Start client dev server
cd client && npm run dev/ sim/ # Python: env, physics, hazards, RL, evolution
/ client/ # React+Canvas/WebGL UI
/ levels/ # JSON contracts (campaign + test fixtures)
/ configs/ # YAML/JSON: training, sim, CI acceptance thresholds
/ tests/ # pytest, hypothesis, golden replays, Playwright
/ scripts/ # CLI entrypoints (bootstrap, run, train, bench)
/ docs/ # design docs, diagrams
/ experiments/ # notebooks, ablations (no prod code)
- Agents: 80-300 birds with energy, stress, genome, and social memory
- Environment: 2D world with wind, food, risk fields, and player beacons
- Machine Learning: PPO-lite within seasons, neuroevolution between seasons
- Determinism: Fixed RNG seeds for reproducible simulations
See CLAUDE.md for comprehensive development guidelines and quality standards.
make lint- Run Python/TypeScript lintersmake type- Run type checkersmake test- Run unit testsmake accept- Run acceptance suitemake bench- Performance benchmark
Copyright 2024. All rights reserved.