Skip to content

Ecofrontiers/mozilla-ags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 

Repository files navigation

Agent Governance Sandbox

A geoprospective simocracy platform: governance by simulation.

Communities managing shared natural resources configure AI agents to represent their stakeholders, choose institutional rules, and watch agents govern real territory under those rules. Configure values, delegate authority, run the simulation, review outcomes, adjust.

Built for the Mozilla Foundation Democracy x AI Incubator Cohort.

What is simocracy?

Governance decisions should be testable before they're binding. Instead of debating institutional designs in the abstract, communities run them and see what happens to their territory in economic terms.

Stakeholders express their values and constraints, delegate governance authority to AI agents, and review how those agents governed. It's democratic participation through delegation and evidence, not just voting.

Working prototype

The proof-of-concept is built on the Camargue, a 100,000-hectare UNESCO Biosphere Reserve in southern France. Five AI agents represent real Camargue stakeholders:

  • Rice farmer
  • Salt harvester
  • Flamingo conservationist
  • Tourism operator
  • Marine biologist

Users run the simulation under four governance configurations:

Configuration What happens
Tragedy of the Commons No institutional constraints. Ecosystem collapses within 12 rounds.
Ostrom's 8 Principles Graduated sanctions, collective-choice arrangements, monitoring. Territory stabilizes.
Cybernetic Quadratic voting and stake slashing. A futuristic governance form.
Camargue Real-World Mirrors the actual PNRC governance architecture operating since 1970.

All outcomes use the Camargue's real economics:

  • Commodity production: ~€360M/yr
  • Ecosystem services: ~€2.3B/yr
  • 19 territorial zones from OpenStreetMap
  • Per-hectare data from peer-reviewed sources

The system tracks eight governance failure modes with cascade relationships that signal when and why cooperation breaks down.

Scaling beyond the Camargue

The architecture generalizes to any region with sufficient open data (OpenStreetMap coverage, published economic valuations, identifiable stakeholder groups). The Camargue is the proof-of-concept. We want to deploy the Sandbox across biosphere reserves, national parks, marine protected areas, and municipal watersheds worldwide.

The UNESCO World Network includes 748 biosphere reserves in 134 countries. Most face governance questions with no comparative tools.

How it works

  1. Configure — express governance values and constraints
  2. Delegate — assign authority to an AI agent
  3. Simulate — agents govern the territory autonomously
  4. Review — examine outcomes in local economic terms
  5. Adjust — refine preferences based on observed results

Tech stack

  • React + TypeScript
  • Mapbox (territorial visualization)
  • OpenRouter (AI agent reasoning)
  • OpenStreetMap (territorial zones)
  • Supabase (state management)

Team

Pat Rawson — Independent researcher and technologist. Co-author, Green Crypto Handbook (Taylor & Francis, 2026). Built Regen Atlas, AI Design Atlas, Windfall. Nine consulting engagements including Celo Foundation and Glow governance.

Louise Borreill — Ecological economist. Doctoral research on ecosystem service governance in Mediterranean biosphere reserves. Institutional knowledge of Camargue governance bodies and stakeholder dynamics.

Ecofrontiers SARL — French research and technology consultancy. All outputs MIT-licensed.

Links

License

MIT

About

Agent Governance Sandbox — Simocracy platform for testing institutional governance frameworks on real territories

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors