AGORA — a sandbox for the economics of abundance

AGORA is an open, stock-flow-consistent model of the European economy built to answer one question rigorously: as AI raises productivity, who gets the gains — and which policies actually share them? It compares policies (cash UBI vs Universal Basic Capital) inside one accounting-complete world, calibrated to live data and validated against history.

Sandbox, not oracle. AGORA produces scenario comparisons under explicit, swappable assumptions — not forecasts. Every parameter is inspectable and sourced; every scenario must pass a consistency gate (no money created or lost in the accounts) before any result is reported.

📄 Read the study: Owning the Machine — A Governance of Abundance for Europe (12 pp) → AGORA_Manifesto.pdf · permanent copy: https://doi.org/10.5281/zenodo.20726928 · SSRN: (forthcoming) 📋 One-page policy brief: AGORA_Policy_Brief_UBC.pdf — practical paths to implement UBC and where the capital comes from. 🔗 Live dashboard: https://brano80.github.io/agora/

---

The headline result

Under an AI labour-share shock (Germany, 30-year scenario):

• Cash UBI lowers income inequality but leaves the top-10% wealth share ~56% — it redistributes the income capital throws off, never the capital.
• Universal Basic Capital (a citizens' fund paying a per-capita dividend) collapses it to ~20% — it distributes the asset itself.
• With fund reinvestment, the UBC economy ends larger than the cash economy — predistribution needn't cost growth.
• A pooled EU dividend roughly halves between-country inequality.
• Across a multi-objective frontier, UBC policies dominate both cash UBI and inaction.

At equal cost, Universal Basic Capital is non-dominated across all five objectives (growth, equality, poverty, fiscal sustainability, stability); cash UBI and “do nothing” are dominated everywhere. Per-country charts: DE.

How it's built

• Stock-flow-consistent core + pluggable modules (distribution · input-output · AI-shock · citizens' fund · multi-region) through a common schema.
• Live data: Eurostat national accounts, AMECO wage shares, BIS household debt, OECD wealth distribution, Eurostat FIGARO input-output — integrated via DBnomics; 26 EU members calibrated.
• Validated: back-tested to 2010–2019 history (~2.3% mean GDP error; government-debt path within the 10% bound); a consistency gate enforced on every run (residuals ~1e-10 of GDP); 241 automated tests.
• Pure standard-library Python engine.

Drive it in plain language (MCP + agent crew)

AGORA runs as an MCP server, so any AI assistant — Claude Desktop, Claude Code, VS Code — can drive the model directly, and a small agent crew turns a plain-language request into a gated, reported run. Ask in words:

"Compare cash UBI vs UBC under an AI shock in Slovakia" "What's the policy trade-off frontier for Germany?" "Run an AI shock with UBC at 40% in France"

The crew plans the request, resolves the assumptions for your approval, runs it through the consistency gate, and reports the result — a single scenario, a comparison, or the Pareto frontier — always with the numbers and never a "winner" (the values judgement stays with you). It runs offline with no LLM:

python agent_crew.py "What's the policy frontier for Germany under an AI shock?"

AGORA crew -- policy frontier on DE (30y). 4 non-dominated policies out of 13;
no single 'best' -- each trades one objective for another.
  - UBC t=20%: GDP 9,382,116, Gini 0.250, poverty 0.090, debt/GDP 143.
  - UBC t=30%: GDP 10,496,531, Gini 0.193, poverty 0.016, debt/GDP 136.
  - UBC t=40%: GDP 12,414,605, Gini 0.119, poverty 0.000, debt/GDP 125.
  - UBC t=50%: GDP 14,699,402, Gini 0.056, poverty 0.000, debt/GDP 118.
  Sandbox, not a forecast; every candidate passed the consistency gate.

Why it's trustworthy as a tool — the guardrails live in the tool layer, not the wrapper: every result passes the stock-flow consistency gate before it leaves (a failing scenario returns the failing checks and no numbers, with no switch to disable it); every payload carries its assumptions and data provenance; the comparison tools have no "winner" field; and AGORA bundles no model of its own — optional narration borrows the client's model via MCP sampling.

10 read-only MCP tools (agora_run_scenario, agora_compare, agora_policy_frontier, agora_preview_scenario, agora_narrate, agora_crew, …). Connect it:

pip install "mcp[cli]"
# then add mcp_server.py to claude_desktop_config.json (see docs/MCP.md)

Full tool list + setup: docs/MCP.md · crew design: docs/PHASE5.md.

Quickstart

python -m pytest tests/ -q              # run the test suite (241)
python -m streamlit run dashboard/app.py # explore scenarios
python scripts/run_triad.py --geo DE     # baseline / AI-shift / settlement

Repository map

• schema/ canonical SNA accounts · modules/ model adapters · consistency/ the gate
• data/ live connectors + cached snapshots · orchestrator.py · scenarios.py
• docs/ architecture, manifesto, findings · tests/

Citation

@misc{agora2026,
  title  = {Owning the Machine: A Governance of Abundance for Europe},
  author = {Branislav Ambroz},
  year   = {2026},
  note   = {AGORA — an open, consistency-gated model of AI's distributional impact},
  doi    = {10.5281/zenodo.20726928},
  url    = {https://doi.org/10.5281/zenodo.20726928}
}

License

Code & documents: CC BY 4.0 (attribution). See LICENSE.

---

AGORA is a research preview. It is a tool for comparing policies transparently — not investment, legal, or policy advice.