Compassionism aspires to enhance cultures by eradicating extreme poverty while incentivizing participation and contribution: see how it does.
The math and code of this simulation were engineered by Claude, an AI model made by Anthropic, from the concepts in Duke Johnson's book Better To Best and his related vision for eradicating extreme poverty while enriching cultures and supporting human flourishing. The model has not yet been reviewed by an independent economist; the full code is open for anyone to check, and expert collaborators are welcome.
A research-oriented, browser-based agent-based simulation exploring all five Compassionism architectures for comparative policy analysis — no installation required. BLEI-calibrated against US Consumer Expenditure Survey data.
The page opens with Compassionism's results. It follows the same 500 simulated adults and their children, living in households, who can lose their jobs and find new ones. It runs them for 20 years (or 40, with the Years switch), once with the program and once with no program, in three environments (Reference, Adverse, Stress). Basic living covered leads: the BLEI measure, the days of basic living a person's resources cover. One plain sentence each follows on poverty, savings, children, who gains, work, prices, cost and how it is paid, and the public costs of homelessness avoided. Prices are given as the typical run with the range of eight runs in ten (the 10th to 90th percentile). Above 1,000 times today's prices, the page describes the figure as a limit of the model's price rule, not a forecast. A table sets the Research Hub's own Year 7 targets (poverty under 2%, Gini coefficient 0.25 to 0.30) beside what the model reaches. Every figure is an average over 500 paired runs, with its 95% interval, and the page names the command that reproduces it. A button runs the model live in the browser. The model measures poverty, cost and work, not the cultural effects Compassionism aspires to. Poverty is measured against a living-wage basket, a higher bar than extreme poverty.
Since v5.2.1 the results are guided sections under one control strip (environment, 20 or 40 years, with or without Compassionism). The sections are: the answer in ten seconds; one run's 500 adults with their life paths; how poverty moves over time; who gains; families and children; jobs, community homes and shops, and how risky life is (v5.4); what each part does; what it costs and who pays; how sure we are; and run it yourself. Every result, chart by chart and table by table, with every earlier release, is on the findings explorer. Every number on the pages comes from one data file per release (data/releases/, listed in data/manifest.json); dev/ADDING-A-RELEASE.md says how to add the next one.
▶ Open the simulation · The findings, chart by chart · DOI: 10.17605/OSF.IO/QWTE2
The narrated walk-through is off the page while the model is still changing (it showed an earlier version); it comes back, re-recorded, once the simulation is settled.
This file covers what the project is and how to run it. For everything version-specific — the full changelog, current output metrics, methodology, and formulas — see the Replication Framework, which is kept current with each release. For open questions, known limitations, and how to contribute, see CONTRIBUTING.md.
The simulation models all five integrated Compassionism architectures and their interactions:
| System | Layer | Role in the model |
|---|---|---|
| CCO (Creative Currency Octaves) | Economic | Distributes flat Basic Units (BU) monthly to participants, redeemable at PTF businesses and convertible to primary currency at merit-scaled rates that improve with an agent's octave level and quality |
| PTF (Public Trust Foundations) | Asset | Community-owned essential goods/services that accept BU; modeled as a cost-reduction mechanism, not a wealth-addition channel |
| PTH (Public Trust Housing) | Asset | Community-owned housing; reduces housing-linked living costs and routes a share of that saving into Acre Equity, which appreciates over time and is accessible at exit subject to a liquidity haircut |
| SZH (Social Zone Harmonization) | Spatial | Zone-level coherence gives residents a coherence-linked benefit and gates a separate cooperative-synergy bonus once PTF merchant density crosses a threshold |
| CIP (Citizens Internet Portal) | Democratic | Digital platform administering CCO currency operations and democratic voting on PTF/PTH decisions; participation improves octave-advancement capability, quality accuracy, and reduces conversion-tax leakage |
Welfare outcomes are measured against the Basic Living Economic Index (BLEI) — a temporal-stability metric expressing how many days of basic living an agent's accessible resources cover, rather than a raw income or wealth snapshot. See the BLEI paper for the full framework, and the Replication Framework for the current formulas as implemented.
No Python. No installation. Just open index.html in any browser. The earlier engine (v4.22, with its own settings, presets and charts) is on its own page, earlier-engine.html, linked from the front door as "Explore the earlier engine".
- Download the repository (or at least
index.html, thesite/anddata/folders, andearlier-engine.htmlfor the earlier engine).index.htmlalone still shows every result and runs the model; the guided sections' charts needsite/, and the spread across runs and the 500 adults needdata/served by a web server (browsers do not let a page opened as a file read other files;python3 -m http.serverin the folder is enough) - Double-click — opens in Chrome, Firefox, Safari, or Edge
- Read the results, or press Run it yourself; the earlier engine's controls and Run Simulation are on
earlier-engine.html
- Upload
index.html,findings.html,earlier-engine.html,replication.htmland thesite/anddata/folders to your repository root - Go to Settings → Pages → Source → main branch / root
- Live at
https://yourusername.github.io/compassionism-simulation/
The simulation itself documents its own current controls, presets, calibration constants, and known limitations in-app — see the collapsible Assumptions, ODD Protocol & Known Limitations and References & Citations panels at the bottom of the earlier engine's page (earlier-engine.html), which are kept in sync with the shipped code.
This README intentionally stays stable across releases. For anything tied to a specific version:
- What changed, and when — the Replication Framework's Version History has the full line-by-line changelog, newest first, for every release; each release's detailed notes are in CHANGELOG.md (the current release's in CONTRIBUTING.md).
- Current results — the findings explorer, every table with CSV and JSON downloads, drawn from one data file per release (
data/releases/). (The Replication Framework's Performance Comparison section holds the earlier v4.22 engine's large-N figures, as history.) - The model as implemented — MODEL_SPEC.md (the release model only: what is simulated, the order of events in a year, every submodel, the random streams, what each headline measures), TESTBED_SPEC.md (every switch and labeled reading, with its status and the decision that set it), ODD.md (the ODD protocol, with each parameter's basis) and REPRODUCE.md (the command, file and fingerprint behind every published table). The Replication Framework's Mathematical Framework section describes the earlier v4.22 engine.
- How to cite — CITATION.cff (GitHub's "Cite this repository" reads it).
- Open questions, known limitations, and how to contribute — CONTRIBUTING.md, which tracks unresolved calibration items, model-architecture feedback, and good-first-issues.
The simulation needs nothing installed. The checks that keep it honest need Node.js (22 or later) and one development dependency:
npm ci # installs jsdom (pinned in package-lock.json), used only by domtest.js
node harness.js validate # the seed-42 reference run, asserted against the documented figures
node harness.js unit # pure-function tests
node domtest.js # drives index.html in a headless DOM (a few minutes)
Every harness.js study mode also accepts --agents=N, the population per run (default 500). For example, node harness.js largen 500 headline --agents=5000 reruns the headline figures at 5,000 agents per run, in about five minutes on one CPU core.
The same three run automatically on every push (.github/workflows/checks.yml). CONTRIBUTING.md explains what each covers and what none of them can see. Today unit runs 384 tests and domtest 159 checks; each run fails if these numbers are stale, and --write-counts (for either) refreshes them.
Split licensing, since v4.8:
- Source code (
index.html,harness.js,domtest.js) — Apache License 2.0. - Documentation and papers (this README, CONTRIBUTING.md, the Replication Framework, and the Research Hub papers) — Creative Commons Attribution 4.0 International (CC BY 4.0).
Cite as:
Johnson, D. & Claude (Anthropic). (2026). Compassionism Framework Simulation. Better To Best Research Hub. https://doi.org/10.17605/OSF.IO/QWTE2 (source code: Apache License 2.0; content and documentation: CC BY 4.0)
The DOI resolves to the archived record; cite the specific version and date you used alongside it if that matters for your purposes (the simulation's own footer and export files record the version and timestamp of any given run).
Research Hub: bettertobest.github.io/research-hub Wiki: bettertobest.github.io/research-hub/wiki BLEI Paper: basic-living-economic-index.html Replication Framework: replication.html Contributors' Guide: CONTRIBUTING.md DOI: 10.17605/OSF.IO/QWTE2 Contact: BetterToBestResearch@gmail.com