Why Scientific Funding Needs a Portfolio, Not a Single Winner

Getting your Trinity Audio player ready...

`Scientific funding should not attempt to identify one certain winner. It should construct a diversified portfolio of plausible discoveries, accept that some projects will fail, and expand support when evidence of value appears.

This is necessary because frontier research is conducted under fundamental uncertainty. Reviewers may assess whether a proposal is rigorous, relevant, and feasible, but they cannot reliably know which experiment will work, which mathematical idea will become foundational, or which obscure tool will later support an entire field.

A funding system that concentrates resources in the proposal ranked first therefore makes an unjustified assumption: that uncertain scientific futures can be ordered precisely in advance.

A portfolio approach makes a more realistic assumption:

No evaluator can reliably predict every breakthrough, but a well-designed collection of independent research bets can increase the probability that at least some produce exceptional value.

What Is Portfolio-Based Scientific Funding?

Portfolio-based scientific funding allocates resources across a deliberately varied set of projects rather than treating every competition as a search for one universally superior proposal.

A scientific portfolio may include:

• established research with a high probability of incremental progress;
• high-risk projects capable of producing major breakthroughs;
• replications and adversarial tests;
• foundational work without immediate commercial applications;
• scientific software, datasets, and infrastructure;
• early-career and independent researchers;
• projects from different disciplines, institutions, and countries;
• both prospective grants and retroactive rewards.

The objective is not simply to fund more projects. It is to select projects whose combined properties create a stronger research program.

A 2025 Research Policy article on the portfolio approach argues that funding decisions should consider the attributes of the combined set of proposals, not merely evaluate each proposal in isolation.

This distinction is crucial. Two individually strong projects may investigate nearly identical hypotheses using similar methods. Funding both could provide less scientific diversification than supporting one of them alongside a substantially different approach.

Science Is Not a Horse Race

Grant competitions are often structured as if proposals were contestants in a race:

1. Reviewers score each proposal.
2. Administrators rank the scores.
3. The highest-ranked applications receive funding.
4. Projects below a budget threshold receive nothing.

This process produces a clear administrative decision, but clarity should not be confused with predictive accuracy.

The difference between the highest-ranked proposal and the fifth-ranked proposal may reflect small variations in reviewer preferences, writing quality, disciplinary familiarity, or institutional reputation. It does not necessarily represent a meaningful difference in future scientific value.

The US National Science Foundation explicitly notes that a proposal does not need to receive uniformly excellent scores to be funded and that excellent scores do not guarantee funding. It also encourages high-risk, high-payoff research.

Scientific funding is therefore not merely the mechanical identification of an objectively best application. It is a decision under uncertainty, limited information, and constrained resources.

A portfolio makes that uncertainty explicit rather than hiding it behind a precise-looking ranking.

Breakthrough Returns Are Uneven

Scientific value is rarely distributed evenly across funded projects.

Many projects produce:

• useful but limited results;
• improved measurements;
• negative findings;
• technical refinements;
• datasets or tools;
• evidence that an approach does not work.

A much smaller number may create entirely new research programs, technologies, treatments, or mathematical frameworks.

This means that the value of research funding can be highly asymmetric. A project may lose its entire budget in the narrow sense that its hypothesis fails, while another may generate benefits many times larger than the cost of the full portfolio.

The rational response is not to eliminate failure. Eliminating failure would usually require funding only predictable research—and research whose conclusions are predictable in advance is unlikely to transform a field.

NIH’s High-Risk, High-Reward Research program recognizes this principle by supporting unusually innovative research with the potential for broad impact. Its Transformative Research Award explicitly covers projects that are inherently risky and untested but could create or overturn fundamental paradigms.

Failure Can Be a Portfolio Success

Consider a simplified portfolio of ten projects:

• six produce modest but valid results;
• three fail to confirm their hypotheses;
• one produces a major discovery.

Judging each unsuccessful project separately might suggest that the funder made three mistakes. Judging the portfolio as a whole may show that accepting those risks was necessary to create an environment in which the breakthrough could occur.

A failed project can also produce scientific value by:

• eliminating an attractive but incorrect hypothesis;
• revealing a methodological limitation;
• generating reusable data;
• identifying safety or reproducibility problems;
• redirecting later researchers toward better approaches.

The important distinction is between productive failure and avoidable failure.

Productive failure results from a serious test of an uncertain idea. Avoidable failure results from misconduct, weak methodology, undisclosed conflicts, or preventable operational errors. A credible portfolio system should tolerate the former while detecting and discouraging the latter.

Diversification Must Include Ideas, Not Just Institutions

A funding agency can distribute grants among many universities while still maintaining a highly concentrated intellectual portfolio.

True scientific diversification concerns more than recipient count. It includes variation in:

• hypotheses;
• methods;
• theoretical assumptions;
• disciplines;
• data sources;
• time horizons;
• risk levels;
• researcher backgrounds;
• expected forms of output.

Funding ten teams that use similar models, cite the same literature, and pursue the same fashionable hypothesis is not genuine diversification.

The portfolio must also account for correlated failure. If every project depends on the same dataset, experimental technique, theoretical premise, or AI model, one hidden defect could undermine the entire program.

This is similar to financial diversification, but scientific portfolios cannot be optimized using financial risk alone. Scientific returns include knowledge, public health, infrastructure, education, environmental benefits, and future options—not merely revenue.

Research on portfolios of risky scientific projects emphasizes that funding agencies face a distinct problem because research projects do not behave like ordinary tradable financial assets.

A Portfolio Should Combine Different Funding Mechanisms

A strong research portfolio should diversify not only projects but also the mechanisms used to fund them.

Prospective grants

Traditional grants remain useful when researchers need equipment, personnel, field access, or laboratory capacity before meaningful work can begin. Grants are particularly appropriate for early-stage exploratory research whose outputs cannot yet be specified contractually.

Milestone funding

Milestone payments release resources when defined intermediate results are achieved. They can reduce exposure to projects that cease progressing while preserving support for long, technically demanding programs.

Milestones should measure genuine scientific progress, however—not merely administrative activity or the production of reports.

Prize funding

Prizes are useful when a funder can clearly define a desired result but does not know which team or approach will achieve it.

They are less suitable when researchers cannot afford the work without advance financing.

Retroactive funding

Retroactive funding rewards contributions after evidence of usefulness, validation, reuse, or impact becomes available.

It reduces dependence on speculative proposal narratives and can support researchers whose work was initially overlooked. For a detailed comparison, see blockchain and alternative research-funding mechanisms.

Small exploratory allocations

Small, rapid allocations can let many ideas generate initial evidence. Successful projects can then receive larger follow-on funding.

This resembles a staged portfolio: broad exploration first, selective scaling later.

Baseline support

Some researchers, laboratories, software maintainers, and infrastructure projects require stable support that should not depend on winning repeated short-term competitions.

Portfolio funding must therefore balance exploration with continuity.

Fund Broadly, Then Scale Evidence

Portfolio funding does not require dividing money equally among every eligible project. A more effective structure is adaptive:

Stage 1: Broad exploration

Allocate limited funding to a diverse set of technically credible approaches.

The threshold should exclude proposals that are clearly unsound, fraudulent, or infeasible, but it should not pretend to distinguish precisely among all promising ideas.

Stage 2: Evidence collection

Track outputs such as:

• validated findings;
• open datasets;
• reproducible software;
• experimental progress;
• formal proofs;
• independent replication;
• documented scientific reuse;
• correction of previous errors.

Stage 3: Selective expansion

Increase support for work that demonstrates exceptional progress, produces valuable dependencies, or opens promising new research directions.

Stage 4: Retroactive recognition

Continue rewarding outputs whose importance becomes visible only later.

This approach replaces one enormous prediction with a sequence of smaller, revisable decisions.

The OECD’s work on mission-oriented innovation similarly emphasizes broad portfolios, systematic monitoring, and adaptive portfolio management rather than rigidly committing to a single intervention.

Why Selecting One Winner Encourages Conformity

When funding is concentrated into a small number of winners, applicants have strong incentives to resemble what reviewers already recognize as fundable.

Researchers may respond by:

• avoiding unfamiliar terminology;
• minimizing theoretical disagreement;
• overstating certainty;
• selecting fashionable topics;
• promising short-term applications;
• imitating previously successful proposals;
• excluding unconventional collaborators;
• concealing the true risk of the work.

This can create a paradox: every proposal is individually polished and defensible, while the total research portfolio becomes intellectually homogeneous.

A portfolio framework can deliberately preserve disagreement. Projects supported by conflicting theories may all deserve funding when available evidence cannot yet determine which theory is correct.

Reviewer disagreement can itself be informative. A proposal that some qualified reviewers consider extraordinary and others consider implausible may represent a different type of opportunity from one that everyone regards as moderately good.

The portfolio should contain both.

Portfolio Funding Is Not an Excuse for Careless Allocation

Diversification can be misused. Funding many weak projects does not automatically create a strong portfolio.

A legitimate portfolio still requires:

• minimum standards of rigor;
• transparent selection criteria;
• conflict-of-interest controls;
• independent auditing;
• fraud detection;
• evaluation of correlations among projects;
• termination or redesign of persistently unproductive programs;
• protection against political or institutional capture.

Funders must also avoid excessive fragmentation. Some scientific problems require large instruments, long time horizons, or coordinated teams. Dividing the budget into tiny grants could make those projects impossible.

Portfolio design is therefore an optimization problem involving diversity, scale, risk, expected value, scientific dependencies, and time.

How AI Could Help Manage Scientific Portfolios

Human committees cannot continuously analyze every publication, dataset, dependency, replication, code repository, and research update across global science.

AI systems could assist by:

• mapping overlap among proposals;
• detecting intellectual concentration;
• identifying projects dependent on the same assumptions;
• comparing risk exposure across disciplines;
• tracing reuse of scientific software and datasets;
• identifying neglected research areas;
• monitoring milestones and new evidence;
• estimating whether the portfolio is overly concentrated;
• proposing alternative allocation scenarios.

This is one potential role for AI meritocracy in research funding: not declaring one proposal the absolute winner, but helping donors and institutions understand how different contributions fit into a wider scientific ecosystem.

AI should not exercise unreviewable authority. Portfolio recommendations must remain explainable, auditable, contestable, and subject to independent governance. Systems that allocate money should also undergo adversarial testing and independent oversight.

AIIM as a Portfolio Allocation Layer

The proposed AI Internet-Meritocracy (AIIM) model could support portfolio funding by evaluating demonstrated scientific and open-source contributions rather than relying exclusively on advance proposal rankings.

An AIIM-style system could allocate resources across:

• foundational and applied science;
• conventional and independent researchers;
• papers, software, data, proofs, and replications;
• early work and mature contributions;
• global funds and purpose-specific funds;
• prospective support and retroactive rewards.

Its most valuable role would not be to produce a single universal ranking of scientists. That would recreate the winner-takes-all problem at a larger scale.

Instead, AIIM should help construct plural portfolios under transparent rules. Different donors or public institutions could select different objectives while relying on a shared, auditable evidence infrastructure.

One portfolio might prioritize neglected basic mathematics. Another might support reproducible biomedical research. A third might finance scientific software or climate measurement infrastructure.

The system could then measure how each portfolio performs without insisting that every legitimate funder adopt the same definition of value.

What Should a Scientific Funding Portfolio Measure?

A portfolio cannot be evaluated only by counting successful projects.

Useful portfolio-level indicators include:

Dimension	Relevant question
Scientific diversity	Are materially different ideas and methods represented?
Correlation risk	Could one flawed assumption undermine many projects?
Risk balance	Does the portfolio include both dependable and transformative work?
Time balance	Are short-, medium-, and long-term projects supported?
Reproducibility	Are validation, replication, and correction adequately funded?
Dependency value	Does the portfolio support tools and foundations used by other work?
Accessibility	Can independent and early-career researchers participate?
Adaptability	Can allocations change when new evidence appears?
Accountability	Are decisions, evidence, and financial flows auditable?
Societal value	Does the combined portfolio address important public needs?

No single metric should determine success. Metrics themselves create incentives and can be manipulated.

Quantitative evaluation must therefore be combined with transparent qualitative analysis and periodic revision.

The Correct Question Is Not “Which Project Will Win?”

Scientific funding often asks:

Which proposal is most likely to succeed?

A portfolio approach asks better questions:

Which combination of projects gives society the best range of scientific possibilities?

Which failures can we afford, and which risks would be irresponsible?

Which approaches are too correlated?

Which neglected ideas deserve a small opportunity to generate evidence?

When should successful work receive additional funding?

How should value be rewarded after its importance becomes visible?

These questions reflect the actual structure of discovery.

Conclusion: Fund the Search Space

A research funder is not merely purchasing predictable outputs. It is financing a search through a vast space of theories, experiments, methods, tools, and possible discoveries.

Selecting one apparent winner too early narrows that search before sufficient evidence exists.

A diversified scientific portfolio provides a better model:

• fund multiple credible approaches;
• preserve methodological and intellectual diversity;
• tolerate informative failure;
• combine grants, milestones, prizes, and retroactive rewards;
• monitor results continuously;
• scale projects when evidence strengthens;
• support the foundational work on which visible breakthroughs depend.

The purpose of diversification is not to avoid judgment. It is to make judgment more scientifically realistic.

No funder can know the future of science. But funders can construct portfolios robust enough to discover it.