Compassionism beside other designs

How Compassionism compares with other whole systems. This page compares anti-poverty designs inside one simulation. For how Compassionism compares with other comprehensive systems (for example, systems of economics, governance and social organisation) on a wider set of values, see the Normative Evaluation.

Part 1. Six designs on the same simulated adults, current engine (v5.6)

Claude’s reading of the design This is the latest comparison: the six designs on the current engine (500 simulated adults, the same 500 paired runs in each environment). Every arm, including no program, has today's safety net (SNAP with its work rule, the earned income tax credit and the child tax credit).

• Equal cost. The UBI, the negative income tax and the asset endowment are each sized to cost the same as the Compassionism specification from the Research Hub. X-Cents and public grocery are shown at their own, much smaller, cost.
• Two ways to pay. Under the Source pays the program is paid for with newly created money, so prices rise (in this model the rise can be very large) and wealth poverty rises under every cash design because savings lose value. Under a flat contribution on wages pays there is no price effect, but earnings are taxed. The answer depends on which rule is used, so both are shown.
• Source pays, all three environments, 20 and 40 years. At equal cost, Compassionism raises prices less than the UBI and the negative income tax but more than the asset endowment, and it lowers poverty severity less than the UBI and the negative income tax. X-Cents and public grocery cost far less and move poverty far less.
• Wage contribution pays. The UBI lowers severity more than Compassionism in the Reference and Adverse environments at both lengths. The asset endowment raises it in every environment, sharply in Reference and Adverse. In the Stress Test the changes are small over 20 years (Compassionism −0.45, the UBI −0.99); over 40 years Compassionism is level with no program (+0.16) while the UBI and the negative income tax raise severity (+2.62 and +1.91).

Claude’s reading of the results Compassionism's wealth effect takes time to show. Between year 20 and year 40 the gap in wealth poverty (share with too little wealth) between Compassionism and no program narrows in all 11 Compassionism rows where the Source pays. In the Reference environment the Core row's gap falls from 31.0 to 21.8 points, and with every reward dollar matched by output (a = 1) or with the two untested mechanisms on it reverses, from +3.2 to −8.7 and from +18.3 to −17.9 points. Under the wage contribution the same narrowing appears in the Reference and Adverse environments but not in the Stress Test. On the cost-of-living measure the pattern is different: the share below the line falls over time with and without a program, so Compassionism's advantage there is a little smaller at 40 years than at 20. The Adverse and Stress environments repeat shocks for the whole run. These are model results, not forecasts, and Parts 2 and 3 have no 40-year version to confirm them.

The gaps between designs are results of this model, not forecasts, and the model does not measure cultural or creative effects. Every table is shown for 20 and for 40 years; use the switch below. The exact commands, and the full tables with price levels, are in the results repository (BetterToBest/compassionism-sim-results, folder s171-cfc3322).

How to read the rows. Compassionism Core is the like-for-like row: the Hub specification with the two untested mechanisms switched off, so it gets no extras the other designs do not. The other Compassionism rows show what changes if those mechanisms are on, if every reward dollar is matched by output, or if the price-damping rule is kept on. In the Stress Test the no-program row is the same as in the Adverse environment; the environments differ in how the program performs. A negative severity figure is good: it is the fall in poverty severity against no program (the range is 95%).

How to read the columns. Price rise is how much faster prices rise each year because of the program. Change in earnings is how much people's earnings change because they change how much they work in response to the program. Wealth poverty is the share of people with too little wealth at the end of the run (year 20 or year 40). Below the cost of living is the share of person-years below the cost of living; severity weights the deepest shortfalls most, and the last column is its average over the whole run (20 or 40 years).

Equal cost is set separately for each run length and rule. The UBI, the negative income tax and the asset endowment are sized to match Compassionism for that length and rule, so their amounts differ between the 20-year and 40-year tables. The dollar amount in a design's name is the payment as set; the cost column is what the 500 paired runs measured, and the two can differ (most under the wage contribution).

The Source pays (money created; the released reading)

A flat contribution on wages pays (the old comparison’s rule)

Part 2. Compassionism against today's US safety net (engine v5.5 and v5.6)

Here the "no program" run already contains SNAP (without its work rule), the earned income tax credit and the refundable Child Tax Credit, and so does the Compassionism run; Compassionism is added to them, as on the Hub. Source: the Health Study's rows for today's safety net (500 paired seeds, 20 years, today's health system). This part has no 40-year version; the Health Study was run for 20 years only. It is a separate, earlier study from Part 1, and its safety net lacks two things Part 1's has (SNAP's work rule, and the correction that removes SNAP's income-limit cliff), so its no-program figures differ from Part 1's (for example, 49.3% against 55.2% below the cost of living in the Reference environment). Compare designs within a part, not across parts.

Every gain should hold on all three measures. A design that lowers the first two while the third rises is not a confirmed gain. In the three rows above, poverty below the cost of living and below 30 days of basic living falls in every environment (except 30-days poverty in the Stress Test, where it rises), but too little wealth at year 20 rises in all three, so none of them is a confirmed gain on every measure. In the model that rise comes from prices rising while savings are not protected from inflation.

Part 3. The earlier engine (v4.22): six designs on the same simulated adults

Each design is paid for by the same flat contribution on wages, set each year to cover its cost, and adults adjust how much they work by the same rules under every design (earnings fall by 16 cents per unconditional dollar, the rate measured in the largest US study of unconditional cash). Poverty here is measured against a living-wage basket and the deepest shortfalls are weighted most (FGT2). Negative is less poverty. Brackets are 95% intervals across paired seeds. Today's safety net is not in these runs; Part 2 adds it. These runs cover 20 years only.

Compassionism Core and Compassionism with its untested mechanisms (earlier engine)

The like-for-like row, Compassionism Core, is the engine's Compassionism with the two untested mechanisms switched off: the wage raise for each octave held, and slower inflation in proportion to membership of the Public Trust Foundations (community-owned businesses) and Public Trust Housing. It can be set beside the other designs without giving Compassionism extras the others do not get. The row above it in the table, Compassionism, as coded, has both mechanisms on, which is closer to the Hub's picture of the design but rests on two assumptions that are still theoretical; they carry most of Compassionism's edge in these runs. The button specified on the Hub switches to the Hub's written conversion design (businesses convert the Basic Units, the community currency, that they accept, and expired Basic Units go to projects that participants convert), which costs about two and a half to three times as much per adult. The model measures none of the cultural effects the framework holds make that cost worthwhile. Compassionism as designed differs from the Core in exactly these ways. Read the Core row for the like-for-like comparison and the others as the design's aim.

Limits

• Parts 2 and 3 are 20-year results only. Part 1 is shown for 20 and 40 years. Part 3 (the earlier engine) predates the household, children, health and labour-market mechanisms and the safety net; it has not been re-run on the current engine. The like-for-like re-run of the six designs on the current engine, with the safety net in every arm, is Part 1.
• The comparison designs are simplified versions of real proposals; each is described, with sources, on the earlier-engine page.
• Public costs of poverty (crime, homelessness services, emergency health care) are credited in the current release but were not in Part 3 (the earlier engine), so no design there is credited with the savings from poverty it removes.
• This page is kept out of search results until Duke approves it.

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.
Simulation · Findings · Replication framework · GitHub