Future Philanthropy

This brief is about philanthropy as an institution: what it is worth, what the public pays for it, what rules bind it, what it discloses, and whether anyone can tell what it achieves. It is not about philanthropic science funding, which is a question about contract terms and selection rules and is treated as a funding model at Scientific Funding Models. That brief asks whether an eight-year no-grant contract produces better science. This one asks whether a tax-subsidised perpetual institution answerable to nobody is a defensible thing for a society to run, and what the measurements say.

The framing under test is that private giving can be made strategic and accountable, and the two words fail differently. Strategic is achievable and there are demonstrated instances — organisations that set an explicit cost-effectiveness bar, publish their reasoning, move nine figures a year against it and revise it. That is a description, not a hypothesis. Accountable is where the framing breaks, and it breaks in a specific place: philanthropic accountability in the United States consists almost entirely of voluntary self-report. The statutory layer beneath it — an annual information return, a distribution requirement that counts administration toward the numerator, and an examination rate around four hundredths of one per cent — does not bind on anything the word is normally taken to mean. The organisations that behave best behave best because they chose to, and the mechanism that made them choose is reputational.

Established Start with scale, and with the line in the scale table that most summaries drop. Total US charitable giving in 2025 was $617.20 billion, up 5.7% in current dollars and 3.0% after inflation, passing $600 billion for the first time. By source: individuals $394.2 billion (63.8%), foundations $117.15 billion (19.0%), bequests $62.19 billion (10.1%), corporations $43.67 billion (7.1%). Established Two recipient lines matter more than the total. Education received $92.01 billion, up 11.7%. And giving to foundations was $79.05 billion, down 16.2% — a category that is a transfer within the philanthropic sector rather than a disbursement out of it. Frontier Roughly one dollar in eight of 2025 “charitable giving” was a transfer into an intermediary rather than a payment to an operating charity, and any argument about philanthropic scale that does not separate the two is counting the same dollar twice on its way through.

Established The public subsidy cannot be stated as a single number, and the reason is a disagreement between two federal estimators rather than a measurement error. For fiscal 2024 the Joint Committee on Taxation puts the charitable deduction tax expenditure at $55.4 billion; the Treasury puts it at $82.2 billion. That is a 48% spread on the same provision in the same year, it is the tenth-largest tax expenditure on either estimate, and both agencies report it as three separate items covering education, health and other charitable purposes. Established Report the range and name the cause; do not average. The spread is a definitional dispute about baselines and behavioural response, not noise, and averaging it would manufacture a precision neither estimator claims. Frontier Statutory change alters the base from 2026: a permanent deduction of up to $1,000 for single filers and $2,000 for joint filers who do not itemise, and a 0.5% floor on itemised charitable contributions. Speculative Both change who receives the subsidy and neither has an evaluated behavioural response yet.

Established Donor-advised funds are the sharpest measurement case in the sector, because the payout rate differs by a factor of nearly three depending on the unit of analysis. The aggregate, built on tax-return schedules for fiscal 2024: 3.59 million accounts, $327.87 billion in assets, up 27.9%, $90.57 billion contributed in, up 38.6%, $64.60 billion granted out, up 17.9%, an aggregate payout rate of 25.2%, an average account of $91,300, across 1,512 sponsors — 103 national, 803 community foundations, 606 single-issue. Established The account-level picture, from 57,539 accounts at 111 sponsors over 2014–2022 comprising 600,675 contributions totalling $40.5 billion and 2,270,447 grants totalling $31.6 billion, is a different world. On three-year average payout for 2020–2022: 22% of accounts were inactive, granting nothing; of the remainder, 15% paid out 0–4%, 16% paid 4–10%, 37% paid 10–50% and 10% exceeded 50%. Median payout across all accounts: 9%. Median across active accounts only: 15%. Mean: 18%.

Established And the mechanism generating the gap is itself measured, which is what makes this a finding rather than a complaint. Including closed accounts, which show 100% payout in their final year, raises the mean by five percentage points, from 17% to 22%. The aggregate rate is dominated by large, active and terminating accounts. Established Payout falls with size — very small accounts 31%, small 17%, medium 12%, large 9%, very large 11% — but 53% of inactive accounts were very small or small, so the size gradient and the dormancy problem point in opposite directions and neither summarises the other. By sponsor type: religiously affiliated 24%, community foundations 18%, national programmes 16%. Frontier What these numbers settle is narrower and more useful than the reform debate they are usually deployed in: the 25.2% headline is real and is not evidence about the behaviour of a typical account, and any argument quoting one figure without the unit of analysis attached is quoting half a dataset.

Established Foundation payout shows a floor behaving as a target, and the metric is more generous than its name. Private foundations must distribute approximately 5% of net investment assets annually. On tax-return data, median payout ran 5.2% to 5.6% over 2015–2021, and for billion-dollar-plus foundations 4.6% to 5.4% — hovering just above the statutory minimum, with the largest institutions frequently below the sector median. US foundations held roughly $1.1 trillion at the end of 2022. Established The definitional point is load-bearing and is conceded by the analysis making the reform case: qualifying distributions include administrative overhead, trustee and staff compensation, programme-related investments and travel. The payout rates above are not grant rates and would all be lower if only grants to charities counted. Established Mark the interest — the analysis comes from an organisation advocating a higher statutory payout, the underlying data are tax returns, and the concession that the metric is generous runs in the direction of its own argument. Established One methodological caveat travels with it: electronic filing covered about 65% of foundations through 2019 and roughly 95% in 2020–2021, so the earlier years are a partial sample. Frontier A minimum that a sector clusters immediately above is functioning as a target. Speculative Whether raising it would raise disbursement or accelerate spend-down and closure is not established by anything in this record, and the natural experiment — foundations that voluntarily set higher payouts or limited lifespans — has never been evaluated against matched perpetual foundations.

Established Now the statutory accountability layer, in the units it was measured in. In 2024 the tax authority received over 1.8 million tax-exempt returns and examined roughly 7,000 exempt entities, of which 666 were Form 990 examinations — down from 1,020 in 2023 and 1,343 in 2022 — plus 134 Form 990-T, 210 Form 4720 for excise taxes, which is the mechanism for payout violations, and 72 across all other exempt forms. That is an examination rate on Form 990 of roughly 0.04%, and it is falling. In the same year the authority processed 136,323 applications for exempt status, approving 113,312 and denying 74. Established A denial rate of 74 in 136,323 is 0.05%. Entry to the sector is effectively unrationed and exit is effectively unpoliced. Frontier This is the same order-of-magnitude enforcement shortfall that Scientific Governance Models documents for research integrity — a statutory office operating two to four orders of magnitude below the volume it names — appearing here in a sector whose accountability claim is that legal oversight exists.

Established The voluntary layer is better than the statutory one and is still thin, and the disclosure figure is the one to carry. A benchmarking survey of 127 senior evaluation or programme staff at foundations giving at least $10 million a year, with median assets around $530 million and median annual giving around $28 million, found that only 21% maintain a separate evaluation unit, that the median foundation has about one evaluation full-time equivalent per ten programme staff, and that about half spend $200,000 or more on evaluation while about a quarter spend $40,000 or less — with only a third of respondents confident in their own spending estimate. Established On sharing findings quite a bit or a lot: 66% with their own staff, 47% with their board, 28% with grantees, and 14% with the general public. Seventy-six per cent say evaluation findings inform mid-course strategy adjustments and 70% name incorporating results into future work as a challenge. Frontier A sector in which one institution in seven shares what it learns with the public, and one in four with the organisations it funds, cannot be described as accountable in any sense that would be accepted of a public agency. It can accurately be described as reflective.

Established And where a real evaluation exists at all, the two best cases count double, because the funder paid for the finding that embarrassed it. A teacher-effectiveness initiative cost $597 million across seven sites over 2009–2016, of which $237 million came from the foundation's grants and the rest from local general funds and federal grants, across three school districts and four charter management organisations. The commissioned evaluation's own words: “Overall, however, the initiative did not achieve its goals for student achievement or graduation, particularly for LIM students” — low-income minority students, the initiative's explicit target — and it also reports that the initiative did not generally increase those students' access to more-effective teaching, with outcomes not dramatically better than at comparable non-participating sites. Established The second case: $100 million matched to a $200 million total for a city school reform over 2011–12 to 2015–16. A value-added analysis against similar students statewide found English achievement growth falling from −0.014 SD at baseline to −0.082 SD by 2014 before recovering to a statistically significant +0.058 SD by 2016, and mathematics falling from +0.068 SD to −0.092 SD in 2013, recovering to +0.035 in 2015 and falling to −0.057 in 2016, not significant. 61% of the English improvement between baseline and 2015–16 was attributable to enrolment shifting toward higher-performing schools rather than to schools improving. The study was funded by the donor's own foundation. Frontier Both of the two most-cited evaluations of large-scale American philanthropy were paid for by the funder whose programme they found had not worked as intended. Interest running against the finding raises the weight of both substantially, and it is the strongest evidence in this brief that the accountable half of the framing is achievable — voluntarily, by donors who choose it, in a sector where nothing compels it.

Speculative The organising question is whether philanthropy has a justification that survives measurement, and the strongest candidate is a claim about horizon. The positive case — philanthropy as an instrument of discovery, funding long-horizon experiments that neither markets nor electorally accountable governments will attempt — is the sector's own self-description and also the strongest published academic defence of it. Speculative Its testable consequence is that philanthropically funded programmes should show fatter tails than publicly funded ones, and the closest thing to a test anywhere is the single investigator comparison recorded at Scientific Funding Models. One comparison, in one field, on a treatment cohort thirty years old, is a thin evidentiary basis for the legitimacy of a $1.1 trillion sector — and it is the best there is.

Frontier The opposing position is that philanthropy is a democratic harm, and its evidence is the institutional facts rather than a theory. Foundations exercise power that is largely unaccountable, often perpetual and heavily tax-advantaged, converting private assets into public influence with a plutocratic bias, subsidised by taxpayers who have no mechanism to hold the institution to account. Frontier As a description of the institutional facts this is well supported by everything in section 2 — the subsidy, the payout floor, the examination rate, the disclosure figures. Handwave As a normative conclusion it does not follow from them, and the same author who makes the critique makes the discovery defence, which is why this brief carries both.

Established On payout the useful disagreement has moved from level to distribution. The aggregate donor-advised-fund rate of 25.2% against a 9% account-level median and 22% inactivity is a distribution problem, not a level problem, and it would be settled by an account-level panel with donor identifiers — which sponsors hold and do not publish. Speculative The reform proposition that a higher statutory foundation payout would raise disbursement is genuinely open: foundations clustering just above the floor is equally consistent with the floor binding and with it being irrelevant to institutions optimising perpetuity, and no natural experiment has been evaluated.

Established The strategic case has a movement attached to it and the movement is small enough to state exactly. One evaluator raised $415 million in 2024, up from $355 million in 2023, and directed $397 million from more than 30,000 donors across 55 grants to 34 organisations in 22 countries, estimating 34 million people reached and 74,000 lives saved — and stating that the estimate covers only the 17 grants focused on reducing mortality and excludes the other 38. Established The principal institutional funder reported $87 million to that evaluator's recommended charities in 2024, about $25 million spent and $50 million committed on technical AI safety, $104 million raised for a lead-exposure fund, staff growing from just over 100 to nearly 150, and roughly 15% of funds directed coming from partners other than its principal donor pair. Frontier The self-criticism in that report is the against-interest content and is worth more than the impact figure: the organisation states that its rate of AI-safety spending was too slow, that it should have expanded earlier and is now playing catch-up, and that information security work has not had a clear internal owner despite early recognition of its importance. Established The structural fact is the concentration. A movement whose institutional funding is roughly 85% dependent on a single donor pair has a governance profile closer to a large private foundation than to a movement, whatever the participatory rhetoric.

Frontier The intellectual dispute about that movement should be stated at full strength on both sides, because it is the live one. The defence argues that the core beneficentric commitments are close to truisms, that the measurability objection misreads the evidential standards because any epistemic reason can count, that indirect helping is as legitimate as direct, and that the charge of laundering billionaire reputations is empirically unsubstantiated — while conceding that critiques identifying implementation problems in actual organisations deserve consideration. Speculative The strongest published attack frames the movement as a textbook case of moral corruption, and the charge is about the moral epistemology of the method rather than about any implementation. That argument is reported here as a position held rather than summarised, because the research pass behind this brief obtained the framing sentence and not the chapter. Frontier A third position sits between them and is the one with evidence: that the movement's real contribution is the norm of publishing your own bad news — the excluded grants in the impact estimate, the admission that priority spending was too slow. The norm is documented; whether it changes allocation is not.

Established The governance shock of the period is dated and its lessons are narrower than usually claimed. On 11 November 2022 an entire donor-funded grantmaking team resigned publicly, citing fundamental questions about the legitimacy and integrity of the business operations behind the fund, condemning deception in the strongest terms, and stating that many committed grants would not be honoured. The associated charity later paid $26,786,503 to the bankruptcy estate, representing 100% of the funds it had received in 2022, under a December 2023 settlement, after an independent law-firm investigation found no evidence that anyone there was aware of the fraud; it then decentralised, spinning three constituent organisations into independent legal entities and stating that the events made the costs of centralisation much more salient. Frontier What that establishes: grantees carry counterparty risk they do not price. A charity that has spent a grant can be required to repay it in full years later, and a movement's central legal entity was exposed to a single donor to the tune of $27 million. The response was structural rather than procedural. Speculative What it does not establish is that this governance was worse than comparable philanthropic structures. There is no base rate. Foundations routinely depend on one living donor, are routinely governed by boards that donor selected, and are not routinely subject to an independent legal investigation when the donor's fortune turns out to be fraudulent — because the failure mode is rare, not because the governance is better. Frontier The honest reading is that the episode revealed the general fragility of donor-concentrated philanthropy, in the one part of the sector that publishes its own post-mortems.

Frontier And the sector's flagship cost-effectiveness claim has been through replication, which is more than most claims here can say. On mass deworming, a systematic review found no demonstrated effect on nutrition, health or learning outcomes, having excluded long-term impact studies, and critics argue the exclusion criteria were too restrictive; a replication of the foundational study identified coding errors and reduced the estimated spillover radius from about 6 km to about 4 km while confirming that spillovers exist; and a third team reached different conclusions through defensible alternative analytic choices rather than by identifying errors. Frontier What is genuinely established is that a flagship estimate survived replication with its central claim intact and its magnitudes reduced, and that reasonable analysts still disagree about the policy conclusion. The summary this brief draws on concludes the policy case is largely unchanged and is written by an advocate in the dispute, which is marked.

Speculative Four further positions are live, are held by serious people and are not settled. Philanthropic effectiveness is unmeasurable in principle, because the counterfactual is a policy world — consistent with two large interventions in real systems producing small ambiguous effects that took years and expensive value-added modelling to detect. Philanthropy works best when it buys a specific, cheap, evidence-backed unit of good, which is the strongest evidenced position in the sector and comes with its own boundary attached: 17 of 55 grants were estimable. The subsidy should be replaced by direct public funding of the same goods, whose empirical premise — that the state would fund the same things — is untested and probably untestable. And the subsidy should be abolished without replacement, a minority convergence of libertarian and left positions, where the regressivity evidence is in the deduction's incidence and the evidence about consequences of abolition is nil.

Established The first bottleneck is that the sector's headline statistic double-counts. Of $617.20 billion given in 2025, $79.05 billion went to foundations rather than to operating charities, and that line fell 16.2% year on year. Frontier Nothing separates the dollar that reached a food bank from the dollar that reached a donor-advised fund on its way to somewhere later or nowhere, and every scale argument in the policy debate is made on the combined figure.

Established The second is that the most-quoted payout rate is a ratio of aggregates and behaves nothing like the typical account. 25.2% in aggregate; 9% as an account-level median; 22% of accounts granting nothing across a three-year window; and closed accounts alone moving the mean five points, from 17% to 22%. Frontier An aggregate dominated by large, active and terminating accounts is a perfectly correct number that answers a question nobody is asking.

Established The third is that the statutory payout metric counts things that are not grants. Qualifying distributions include administrative overhead, trustee and staff compensation, programme-related investments and travel, so the published rates are not grant rates and would all be lower if they were. Frontier That the concession comes from the analysis making the case for a higher requirement is what makes it usable.

Established The fourth is enforcement scale, and it is not a marginal shortfall. 666 Form 990 examinations against over 1.8 million returns, falling year on year from 1,343 in 2022, and 74 denials out of 136,323 exemption applications. Frontier At that rate the return is a disclosure document rather than a check, and the excise mechanism that would enforce the payout rule was applied 210 times in a sector of well over a million filers.

Speculative The fifth is that no base rate exists for philanthropic governance failure. The one well-documented collapse produced a public post-mortem, an independent investigation and a full repayment, because the organisation involved publishes such things. Frontier Comparable institutions do not, so there is nothing to compare against, and every claim that one part of the sector is better or worse governed than another is currently an argument from the availability of documentation.

Established The first dependency is an account-level donor-advised fund payout series with donor identifiers, published on a schedule. The 2014–2022 study exists because sponsors supplied data voluntarily to an academic collaborative; it covers 57,539 accounts at 111 sponsors out of 3.59 million accounts at 1,512 sponsors. Frontier Everything anyone wants to know about deferral behaviour — whether dormant accounts eventually grant, whether the deduction and the disbursement are separated by years or by decades — is in the sponsors' own systems and is not published.

Frontier Second, a payout measure that counts only grants to operating charities. This is a definitional change rather than a data problem: the components are already reported on the return and are simply summed together. Established Until it exists, the 5% rule is enforced against a numerator that includes the foundation's own running costs, and no published series says how much of the sector's distribution is grants.

Frontier Third, an examination rate that makes the annual return function as a checkpoint rather than as a filing. That is a resourcing decision, and its absence is the reason the entire accountability question in this brief resolves into voluntary disclosure. Speculative The structural finding imported from Scientific Governance Models is that interventions applying a rule to every item work while attempts to improve a fine-grained judgement fail. Philanthropy has no checkpoint at all. The framing under test proposes to improve the judgement.

Speculative Fourth, matched comparisons of limited-life and higher-payout foundations against perpetual ones. Both kinds exist, in numbers, with published returns, and nobody has assembled the comparison — which means the central reform proposal in the sector has no evidence for or against it beyond the observation that institutions cluster above the floor.

Frontier The most valuable study available is the one the sector's own data would support tomorrow: an account-level panel of donor-advised fund behaviour with dormancy tracked to resolution. The existing study covers 2014–2022 and shows 22% of accounts granting nothing in a three-year window. Speculative Whether those accounts eventually grant, and after how long, is the whole of the deferral debate, and it is answerable from records sponsors already hold. The obstacle is publication, not measurement.

Established Second, a natural experiment arrives in 2026 and nobody has announced a design for it. A permanent non-itemiser deduction of up to $1,000 single and $2,000 joint, and a 0.5% floor on itemised contributions, change the price of giving in opposite directions for two identifiable populations on a known date. Frontier That is as clean a policy discontinuity as this sector ever gets, and the outcome data are tax returns. Speculative A pre-registered difference-in-differences on giving behaviour around the threshold would produce the first credible elasticity of charitable giving with respect to the subsidy in a generation.

Frontier Third, evaluate limited-life and voluntarily-higher-payout foundations against matched perpetual ones. The reform argument is that a higher statutory payout raises disbursement; the counter-argument is that it accelerates spend-down. Speculative Both effects are observable in institutions that already made the choice voluntarily, and matching on assets, sector and vintage is a conventional design.

Established Fourth, and this one is a norm rather than a study: more funders paying for independent evaluations of their own flagship programmes and publishing the nulls. The two best evaluations of large American philanthropy exist for exactly that reason, and both found the programme had not achieved its stated goals. Frontier That is admirable and it is not accountability — it is a voluntary act by donors who chose it, evaluated by contractors the donors hired, in a sector where nothing compels it. Speculative A commitment by a group of large funders to commission and publish independent evaluations of a fixed share of grantmaking would convert a norm into something closer to a rule, which is the only mechanism in this brief with any evidence behind it.

Speculative Fifth, the incidence question that would settle the regressivity argument. The deduction's benefit distribution across income is computable from published tax data, and the subsidy's total is disputed by 48% between two federal estimators. Handwave Reconciling those two estimates is a methodological exercise that neither agency has an incentive to undertake, and until someone does, every argument about the size of the public subsidy is an argument about which agency to believe.

Established The machinery has four parts and each one leaks in a documented way. A deduction whose cost two federal estimators put at $55.4 billion and $82.2 billion for the same year. A distribution requirement of about 5% whose numerator includes administration, compensation, programme-related investments and travel. An annual information return examined at about 0.04%. And an excise-tax mechanism, the instrument for enforcing payout violations, applied 210 times in 2024. Frontier Each part exists, each is correctly described in the statute, and none of them functions as a constraint at the observed enforcement volume.

Established The donor-advised fund is the structural innovation of the period and its engineering is the separation of two events. The deduction occurs on contribution; the grant occurs whenever the advisor recommends it, or never. Assets stood at $327.87 billion across 3.59 million accounts with an average of $91,300. Frontier Nothing in the structure requires the two events to be close together, and the account-level data show 22% of accounts granting nothing across three years while the aggregate reports 25.2% payout. Speculative Whether the deferral is a feature — smoothing giving across a donor's life — or a defect depends on the resolution of dormant accounts, which nobody publishes.

Established Foundation evaluation capacity is the operational reason so little is known. Among foundations giving at least $10 million a year, 21% maintain a separate evaluation unit and the median has about one evaluation full-time equivalent per ten programme staff; about half spend $200,000 or more on evaluation and about a quarter $40,000 or less, with only a third confident in their own estimate. Frontier An institution that cannot state its own evaluation budget to within an order of magnitude is not equipped to evaluate anything, and the disclosure figures follow from the staffing figures rather than from any decision to withhold.

Established And the two real evaluations show what it costs to do this properly. A six-year multi-site evaluation of a $597 million initiative across seven sites; and a value-added analysis of a $200 million reform requiring student-level statewide comparison to detect effects in the range of a few hundredths of a standard deviation. Frontier Detecting the effect of a large philanthropic intervention on a real system requires exactly the machinery that 79% of large foundations do not have.

Within this map: Scientific Funding Models, which owns philanthropic science funding as a set of contract terms and takes the investigator comparison this brief cites only once; Scientific Governance Models, whose structural finding — that a rule applied at a checkpoint works where an attempt to improve a judgement does not — is the lens used here, and whose integrity-enforcement shortfall is the closest analogue to the examination rate recorded above; Long-Term Institutions, where the perpetual endowed foundation is one of the few institutional forms actually designed to outlive its founder; Institutional Design, where accountability without a principal is stated as a general problem; Future Public Administration, whose finding that evaluation is discretionary explains most of the absences here; and Existential Risk Governance, a field whose funding base is unusually concentrated in the philanthropic institutions described above.

Outside it: tax law and the economics of tax expenditures, which supplies the disputed subsidy estimate; non-profit law and the fiduciary duties of trustees; development economics, which supplies the cost-effectiveness apparatus and the replication dispute; and political philosophy, which supplies both the legitimacy critique and its best answer, usually from the same authors.

Established The institutional finding is that philanthropy has no checkpoint. Every mechanism this brief examined is either a filing nobody reads at scale, a threshold the sector clusters immediately above, or a voluntary disclosure. 666 examinations against 1.8 million returns; 74 denials against 136,323 applications; 210 excise-tax examinations; a payout metric counting the foundation's own overhead; and 14% of large foundations sharing evaluation findings publicly. Frontier The structural result imported from the science-governance brief names the problem exactly: what works is a rule applied to every item at a checkpoint; what fails is an attempt to improve a judgement. Philanthropy is judgement exercised by the donor and reviewed by the donor, and the framing under test proposes to improve the judgement.

Established The one accountability mechanism with a demonstrated effect is reputational, and its evidence is two funder-commissioned evaluations that found the funder's own flagship programme had not worked. Both were paid for by the donor; both were published; both are the most-cited evidence about large-scale American philanthropy that exists. Frontier That is a norm rather than a rule. It depends on the disposition of individual donors, it is not required of anyone, and it is exercised most by the funders already most confident of their own methods.

Frontier The concentration problem cuts across the strategic case in a way its advocates rarely address. The organisations that have gone furthest on strategy — an explicit bar, published reasoning, cause prioritisation — are also the ones whose accountability is most concentrated: a donor base of more than 30,000 in one case, a roughly 85% dependence on a single donor pair in the other. Speculative The more a philanthropy commits to a coherent strategy, the fewer people the strategy has to satisfy, and the more its accountability collapses into the judgement of whoever wrote it. That is a structural tension, not a criticism of anyone's conduct.

Established Interested parties in this brief run in identifiable directions and it matters which. The payout analysis comes from an organisation advocating a higher statutory requirement and states, against its own convenience, that the metric is generous. The benchmarking survey is by sector bodies surveying their own field. The impact figures come from organisations reporting on themselves, one of which publishes the exclusions in its own estimate and the other of which publishes that its priority spending was too slow. Frontier The deworming summary is written by an advocate in the dispute it summarises. Where a source's interest runs against its finding, this brief weights it up and says so.

Frontier The core ethical objection is precise and the facts underneath it are measured. Foundations exercise power that is largely unaccountable, often perpetual and heavily tax-advantaged, converting private assets into public influence — with a public subsidy that two federal estimators put between $55.4 billion and $82.2 billion, a $1.1 trillion asset base, a distribution floor the sector clusters just above, and an examination rate of 0.04%. Handwave Whether that is justified by a demonstrated capacity for long-horizon risk-taking is not a question any evidence in this brief can settle, and it is the step at which every argument in this area, on both sides, does its work by assertion.

Established The distributional facts of the subsidy are the strongest part of the critique. A deduction is worth more to a taxpayer in a higher bracket, and until 2026 it was available only to itemisers. Frontier The statutory change cuts both ways — a non-itemiser deduction widens access to the subsidy while a 0.5% floor removes it from small itemised gifts — and neither has an evaluated behavioural response, so the net incidence of the current design is unknown at exactly the moment it changes.

Established Grantees bear a risk they do not price and cannot insure. A charity that received and spent a grant was required to repay $26,786,503, being 100% of what it had received in the relevant year, after the donor's fortune turned out to be fraudulent. Frontier That is a general feature of receiving money from a concentrated donor, not a peculiarity of one movement, and it is entirely absent from how grant agreements are ordinarily written.

Frontier And the disclosure figures have an ethical content that is easy to miss. 28% of large foundations share evaluation findings with their own grantees. The organisations that carried out the work, absorbed the strategy and bore the consequences of it are told what was learned about a quarter of the time. Speculative That is a relationship with a power asymmetry doing exactly what power asymmetries do, and it is the part of the accountability problem that could be fixed by any single foundation deciding to fix it.

Established The scale is the civilizational fact: a $1.1 trillion asset base held by institutions designed to be perpetual, disbursing at a rate their sector clusters immediately above a statutory floor to reach, under a public subsidy nobody can price to better than a third. Frontier Whatever else it is, that is a durable parallel structure for allocating resources to public purposes, and it is the only one in a rich democracy that is neither elected nor priced.

Speculative The strongest defence of that structure is not charity but discovery: that a funder with no electorate and no shareholders can attempt long-horizon experiments neither markets nor governments will. It has a testable consequence — fatter tails in philanthropically funded work — and one thin empirical anchor, thirty years old, in one field. Handwave A justification for a trillion-dollar institutional form resting on a single well-identified comparison is the honest state of the argument, and it is not an argument either side should be comfortable with.

Frontier The deflationary reading is that philanthropic effectiveness may be unmeasurable in principle, because the counterfactual is a policy world. Two of the largest philanthropic interventions ever evaluated — $597 million and $200 million — produced effects of a few hundredths of a standard deviation, detectable only with expensive value-added modelling, and in one case 61% of the measured improvement turned out to be students moving between schools rather than schools improving. Speculative If that is the general case, the sector's strategic turn is an attempt to optimise against a signal too weak to steer by, and the correct response is more evaluation rather than more strategy — which is the opposite of what the staffing figures show is happening.

These horizons track statutory changes, filing series and disclosure norms rather than technology:

• 10 yr: Established The dated change is the 2026 tax base: a permanent non-itemiser deduction of up to $1,000 single and $2,000 joint, and a 0.5% floor on itemised contributions. Frontier Expect the giving series to move and expect the attribution to be contested, because two federal estimators already differ by 48% on the provision's cost. Speculative Expect the Form 990 examination rate to keep falling unless resourcing changes, and expect donor-advised fund assets to keep compounding faster than grants out, since contributions in exceeded grants out by about $26 billion in fiscal 2024.
• 25 yr: Speculative Either an account-level payout series becomes routine — in which case the deferral debate resolves on evidence — or it does not and the sector keeps arguing 25.2% against 9% indefinitely. Speculative A statutory payout change is plausible within this window given the sustained reform campaign, and if it happens it will be enacted without any evaluation of whether it raises disbursement, because the natural experiment that would have told anyone was never run. Handwave Which way it lands is a political outcome, not an extrapolation.
• 50 yr: Speculative If the perpetual foundation remains the dominant form, the sector's assets compound at investment returns while disbursing near the floor, and the gap between accumulated wealth and annual disbursement widens as a matter of arithmetic. Speculative The competing trajectory is the spend-down foundation, which several large donors have adopted voluntarily and which nobody has evaluated against the perpetual alternative on any outcome at all.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The one observation with that reach is that perpetual charitable trusts have repeatedly outlived the purposes they were written for and have repeatedly been redirected by courts rather than by donors — which is an observation about legal doctrine, not a base rate, and it is the strongest historical argument for limited-life structures that nobody has assembled into evidence.

• Depends on Nothing on this map. This brief waits on no result another brief produces. It imports one structural finding from Scientific Governance Models — that a rule applied at a checkpoint works where an attempt to improve a judgement does not — and uses it as a lens rather than as a dependency; no typed depends-on edge is claimed for that relationship.
• Requires (not on this map) Three institutional decisions, none of them a research result, and all three currently withheld. First, an examination rate that makes the annual information return function as a checkpoint. In 2024 the tax authority received over 1.8 million tax-exempt returns and conducted 666 Form 990 examinations — down from 1,020 in 2023 and 1,343 in 2022 — alongside 210 excise-tax examinations, the instrument for payout violations, and denied 74 of 136,323 exemption applications. That is roughly 0.04% and falling, and at that volume the return is a disclosure document rather than a check. Second, a payout rule counting only grants to operating charities: private foundations must distribute about 5% of net investment assets, median payout ran 5.2–5.6% over 2015–2021 and 4.6–5.4% for billion-dollar-plus foundations, and qualifying distributions include administrative overhead, trustee and staff compensation, programme-related investments and travel — so no published series states how much of a $1.1 trillion sector's distribution is actually grants, and the rates would all be lower if it did. Third, an account-level donor-advised fund payout series. The aggregate rate is 25.2% across 3.59 million accounts holding $327.87 billion; the one account-level study, covering 57,539 accounts at 111 sponsors over 2014–2022, finds a 9% median, a 15% median among active accounts, an 18% mean, 22% of accounts granting nothing across three years, and closed accounts alone moving the mean from 17% to 22%. The records that would resolve whether dormant accounts ever grant sit with sponsors and are not published. All three are things a legislature or a set of sponsors could choose to supply, and none is a question anyone needs new research to answer.
• Enables In principle any long-horizon programme on this map that assumes patient private capital inherits the payout, disclosure and enforcement facts recorded here. No typed enabling edge is claimed, because the relationship has never been measured: no study connects philanthropic funding structure to a downstream research or programme outcome at sector scale, and the two largest evaluations that exist were commissioned by the funders themselves.
• Adjacent Tax law and the economics of tax expenditures, which supplies the disputed subsidy estimate; non-profit law and trustee fiduciary duty; development economics, which supplies the cost-effectiveness apparatus and the replication dispute; political philosophy, which supplies both the legitimacy critique and its strongest answer; and within this map Scientific Funding Models, Long-Term Institutions and Institutional Design.

Established “The charitable tax subsidy costs about X billion.” There is no single number. The Joint Committee on Taxation says $55.4 billion for fiscal 2024; the Treasury says $82.2 billion — a 48% spread on the same provision in the same year, reported by both as three separate items. Frontier Report the range and name the cause. The disagreement is definitional, about baselines and behavioural response, and averaging it manufactures a precision neither estimator claims.

Established “Americans gave $617 billion to charity in 2025.” They gave $617.20 billion to the sector, of which $79.05 billion went to foundations — an intra-sector transfer, and one that fell 16.2% on the year. Frontier About one dollar in eight did not reach an operating charity, and the headline is routinely used in arguments where that distinction is the whole point.

Established “Donor-advised funds pay out about a quarter of their assets a year.” In aggregate, yes: 25.2%. At account level the median is 9%, the median among active accounts is 15%, the mean is 18%, and 22% of accounts granted nothing across a three-year window. Established Including closed accounts, which pay out 100% in their final year, alone moves the mean five points. The aggregate is real and is not evidence about the typical account.

Established “Foundations are required to grant 5% of assets a year.” They are required to make qualifying distributions of about 5%, and qualifying distributions include administrative overhead, trustee and staff compensation, programme-related investments and travel. Frontier The published payout rates are therefore not grant rates, and the concession comes from the analysis arguing for a higher requirement, which is the direction that makes it credible.

Established “Legal oversight of the exempt sector functions as a check.” 666 Form 990 examinations against over 1.8 million returns, falling from 1,343 in 2022; 210 excise-tax examinations, the mechanism for payout violations; and 74 denials out of 136,323 applications for exempt status. Frontier Entry is effectively unrationed and exit is effectively unpoliced, and this is the same order-of-magnitude enforcement shortfall the Institute's science-governance brief documents in a different sector.

Frontier “Foundations are learning organisations.” 14% share evaluation findings with the general public and 28% with their own grantees; 21% maintain a separate evaluation unit; and 70% name incorporating findings into future work as a challenge. Established Reflective is accurate. Accountable is not.

Established “74,000 lives saved is an estimate of that organisation's 2024 impact.” It covers 17 of 55 grants, by the organisation's own statement, and excludes the other 38. Frontier The exclusion is published by the organisation making the estimate, which is the behaviour the sector should be judged against and is also the reason the figure is quoted more confidently than its author states it.

Speculative “The FTX collapse showed that this movement's governance was worse than comparable philanthropy.” No base rate exists. Comparable institutions routinely depend on a single living donor, are routinely governed by boards that donor selected, and do not commission independent investigations of themselves. Frontier What the episode did establish is narrower and more useful: a charity that had spent a grant repaid $26,786,503, being 100% of what it received, and grantees carry counterparty risk they do not price.

Frontier “The Gates and Newark evaluations show philanthropy cannot work” and “the deworming evidence collapsed” are both overstatements. The first two show that two specific large interventions did not achieve their stated goals in the measured window, and both exist because the funders paid for them. Established On the second, the replication identified coding errors and reduced the estimated spillover radius from about 6 km to about 4 km while confirming that spillovers exist, and a third team differed through defensible analytic choices rather than by finding errors. Handwave And the largest claim of all — that philanthropy's legitimacy turns on any of these measured quantities — is where every argument in this area does its work by assertion. The discovery defence has one thin empirical anchor and the normative conclusion does not follow from it in either direction.