What social systems that actually build human capability look like
"What is the optimal welfare system?" is already the wrong question.
It accepts the premise that the goal is "welfare" — maintaining existence, providing a floor, catching people after they fall. The "safety net" metaphor biases toward crisis response over prevention and invites the "hammock" counter-metaphor that justifies making the net uncomfortable.
The right question: What does a system that builds human capability look like?
Not "how do we help the poor?" but "how do we convert resources into humans who can sustain and extend complexity over time?" The first question optimizes for sympathy. The second optimizes for syntropy.
The answer exists. Multiple systems have demonstrated it. They share common principles derivable from first principles. And they look almost nothing like what Western welfare states actually do.
Systems that require asset liquidation, impose 100% marginal tax rates, surveil every transaction, replace human judgment with algorithms, and reward incapacity performance destroy the capability they claim to build. For the full diagnostic of how Finnish toimeentulotuki exemplifies this pattern, see The Capability Trap. Here, we focus on what actually builds capability.
BRAC, a Bangladeshi development organization, developed what's now called the Graduation Approach. The insight: poverty is a multi-constraint trap. Address one constraint (give money, or training, or health care) and other constraints remain binding. The person stays trapped.
The solution: simultaneous "big push" across six pillars.
The Six Pillars:
Randomized controlled trials across six countries (Ethiopia, Ghana, Honduras, India, Pakistan, Peru) showed effects persisting 7+ years after the program ended. India ROI: 433%. The ultra-poor — people conventional programs couldn't help — graduated to sustainable livelihoods.
Why it works: it addresses the binding constraints simultaneously. Consumption support provides stability. The asset creates income-generating capacity. Training enables productive use. Savings builds buffer against shocks. Health removes physical barriers. Coaching provides the executive function that poverty depletes.
The key insight: you cannot give someone capability by giving them money. You can only give them capability by addressing all the constraints that prevent capability from developing. This requires coordination, which requires human judgment, which requires the thing that algorithmic welfare systems systematically eliminate.
Singapore's Central Provident Fund takes a different approach: mandatory asset-building for everyone.
Every worker — including low-wage workers — contributes to CPF. The government tops up accounts of poor workers. Critically: transfers go into assets (housing, retirement, healthcare accounts), not consumption.
The result: even working poor exit the workforce with home equity and retirement savings. Near-universal "graduation" to capability. No permanent welfare class.
Why it works:
Singapore is often dismissed as "authoritarian" or "unique." But the mechanism is replicable: mandatory savings plus government top-ups for low earners plus restrictions on consumption until retirement. The politics are hard. The engineering is straightforward.
An important caveat: Singapore succeeds at individual capability building but has the world's lowest fertility (TFR ~1.0). The Axiological Malthusian Trap analysis diagnoses it as a "High-Competence Hospice"—brilliant at building capable individuals who don't reproduce. Capability architecture solves one problem; it doesn't solve the selection question about who has children.
EMPath (Economic Mobility Pathways) in Boston developed Mobility Mentoring, which addresses the executive function problem directly.
Poverty depletes cognitive bandwidth. Mullainathan and Shafir's research shows scarcity consumes the equivalent of ~13 IQ points. The same person performs worse when poor than when flush. This isn't character — it's cognitive load.
Mobility Mentoring has participants define their own goals (restoring agency rather than imposing compliance), tracks progress visually across domains, provides mentor scaffolding that withdraws as capability builds, and uses small wins to build self-efficacy.
Results: gains in income, employment, and executive function. Breaking of intergenerational transmission patterns. Participants report restored sense of agency.
The key distinction: "entrepreneurial agency" versus "compliance agency." Standard welfare systems build compliance agency — the ability to navigate bureaucracy, fill out forms, satisfy caseworkers. Mobility Mentoring builds entrepreneurial agency — the ability to set goals, pursue them, and adapt when circumstances change.
Compliance agency is what the system needs from you. Entrepreneurial agency is what you need from yourself.
The Graduation Approach, Singapore CPF, and Mobility Mentoring all work. They also share a limitation: they're all adult interventions.
James Heckman's Nobel Prize-winning research shows that return on intervention decreases with age. Early childhood programs show ~7-10% annual returns. Adult job training shows ~0-2% returns. By age 25, you're trying to rebuild what should have been built by age 5.
The optimal intervention hierarchy:
The three models in this essay operate at level 4. They work—but they're working against established patterns, depleted cognitive bandwidth, accumulated disadvantage. Earlier intervention has higher returns.
Finland has excellent infrastructure at levels 2-3 (neuvola, daycare, schools). But it's universal, not targeted at capability-building for at-risk children. The 16% who end up on bad trajectories had the same access to universal services—the capability deficit was in the home. Level 1 (parenting capability) is unaddressed and connects to harder questions about what the system selects for.
Adult capability-building models remain necessary—people already in the trap need help now. But the real leverage is earlier. The question is whether that's politically tractable.
Despite operating at different levels and contexts, capability-building models share common principles:
1. Build capability, don't maintain existence.
The goal is not "keep people alive" but "create people who can sustain themselves." Existence is the minimum case. Capability is the target.
2. Allow asset accumulation.
Assets provide resilience, options, and compound growth. Systems that require asset liquidation before helping destroy the foundation capability requires. Finland's €50 threshold is capability destruction by design.
3. Smooth transitions.
Benefit cliffs (where earning €1 more means losing €100 in benefits) make formal work irrational. The math must never punish progress. Singapore's universal participation eliminates cliffs entirely. The Graduation Approach's time-limited intensity creates natural transition.
4. Treat recipients as agents.
Systems that require incapacity performance produce incapacity. Systems that treat recipients as agents — with goals, judgment, and potential — produce agents. The difference between "what do you need from us?" and "what are you trying to build?"
5. Integrate human judgment.
Algorithms optimize for measurable proxies. Humans can see the whole person. The Graduation Approach's coaching, Mobility Mentoring's scaffolding, and even Singapore's relatively simple system all preserve space for human judgment about individual circumstances.
6. Measure capability trajectories.
Not "how many applications processed" but "where are people 5 years later?" Not "exit rate" but "sustainable exit rate." Not "compliance" but "capability." What you measure is what you optimize for.
If we know what works, why don't we do it?
The Paradox of Redistribution. Korpi and Palme's research: the more you target benefits at the poor, the less poverty reduction you achieve. Mechanism: targeted programs exclude the middle class; the middle class (with political power) withdraws support; benefits erode; "programs for the poor become poor programs." Universal programs enmesh the middle class, maintain political support, and achieve more redistribution despite apparent "inefficiency."
Finnish toimeentulotuki affects ~4% of the population. They don't vote (26% turnout vs 39% general). They're stigmatized. They have no political constituency. The 70% who never touch the system have no stake in its quality. Stable equilibrium of poor outcomes.
Time horizon mismatch. Capability building takes years. Political cycles are 4 years. Sanctions show immediate "results" (people leave the rolls). Training shows immediate "failure" (people stay on rolls while training). The metrics that politicians can point to systematically favor what doesn't work.
Blame avoidance architecture. Finland's 2017 Kela transfer centralized benefits to achieve "legal equality." The actual effect: algorithmic adjudication means "the computer decided." No individual caseworker faces blame. The reform optimized for blame avoidance, not capability building.
The Copenhagen Trap. Approving benefits that later prove fraudulent creates career risk. Denying benefits that would have helped creates no consequence (the harm is invisible). Over time, systems select for people whose skill is avoiding decisions. The Unstained Incompetent rises precisely because they never made a consequential choice.
Revealed preference. The Finnish Basic Income Experiment provided evidence that removing conditionality improved mental health, trust, and cognitive function without reducing employment. The government's response was to let the experiment conclude without extension while implementing stricter activation. When evidence suggests punitive elements aren't necessary and the system doubles down on them anyway, the operational goals differ from the stated goals—regardless of policymakers' conscious intentions.
Finland already knows how to do this. Housing First (Asunto ensin) applies capability-building logic to homelessness: provide housing unconditionally, wrap services around it, build capability from a stable platform. Result: Finland is the only EU country where homelessness decreased. The logic is identical to the Graduation Approach — address constraints simultaneously, provide foundation first, then build capability. (See The Finnish Irony.)
Applying the same principles to general welfare:
The 2026 reform is bifurcated: Yleistuki improves the upper tier (no asset test, 50% taper) while toimeentulotuki becomes harsher for the lower tier (stricter conditionality, heavier sanctions). This helps the ~14% who would have exited anyway and punishes the ~4% who need help most. The Paradox of Redistribution in action.
There's a harder question this essay hasn't addressed.
Capability-relevant traits are substantially heritable. Intelligence, conscientiousness, time preference, executive function — all show 30-60% heritability in twin studies. This is not controversial in behavioral genetics. It is controversial everywhere else.
Policy affects who reproduces and how much. A system that subsidizes reproduction without capability correlation, maintained across generations, has selection effects. A system that makes capability-building the path to family formation success has different selection effects.
The question almost no one asks: What are the selection effects of welfare design over many generations?
This essay has discussed capability building at the individual level — helping existing people develop capability. It has not discussed what happens to population-level capability distributions when systems reward or punish capability across generations. The two questions are related but distinct.
The Aliveness framework forces this question because it operates on infinite time horizons. "Does this policy help people now?" is different from "Does this policy compound toward a population capable of sustaining complexity over deep time?" Standard welfare analysis asks the first question. The framework demands the second.
This is the question that makes the prescriptive analysis incomplete. You can design perfect capability-building institutions. If the population being served has been selected for incapability across generations, the institutions face a harder problem than the one they were designed to solve.
The selection question is addressed elsewhere. Here, it's enough to note that it exists and that ignoring it doesn't make it go away.
How should social systems be evaluated?
Not: "Does this relieve suffering?" (Sympathy optimization, often entropy.)
Not: "Does this reduce the welfare rolls?" (Exit rate optimization, ignores what happens after exit.)
Not: "Is this fair?" (Fairness is a proxy, not a terminal value.)
The question: Does this system convert resources into humans capable of sustaining and extending complexity over time?
This is the capability question at individual level. At population level, add: Does this system compound toward a population capable of sustaining Aliveness, or away from it?
Systems that pass the first test but fail the second are consuming the future to fund the present. Systems that pass both are genuinely pro-Aliveness.
The Graduation Approach, Singapore CPF, and Mobility Mentoring pass the first test. Whether they pass the second depends on details this essay hasn't examined.
Finnish toimeentulotuki fails both tests. It destroys individual capability while creating selection pressure toward incapability. The 2026 reform fails both tests for the bottom tier while partially passing the first for the upper tier.
The optimal isn't "better welfare." The optimal is replacing welfare with capability architecture — systems designed from first principles to build capable humans rather than maintain dependent ones. The models exist. The principles are clear. The political economy is hard. The physics is not.
This draws from Aliveness, a framework for understanding what sustains organized complexity over time. For the diagnostic analysis of Finnish welfare failure: The Capability Trap. For the selection question this essay gestures toward: The Selection Question.
Capability & Selection series: Diagnostic → Prescriptive → Selection → Institutional
Related: