Should Donors Fund Research, Evaluation, or Scientific Infrastructure?

Getting your Trinity Audio player ready...

Donors should rarely choose exclusively between research, evaluation, and scientific infrastructure. A healthy scientific funding portfolio needs all three: research produces new knowledge, evaluation identifies and improves valuable work, and infrastructure enables many researchers to work more effectively.

The best allocation depends on the bottleneck:

• Fund research when capable researchers have a specific, neglected opportunity but lack resources.
• Fund evaluation when good work exists but cannot be reliably identified, compared, reproduced, or rewarded.
• Fund scientific infrastructure when a shared tool, dataset, platform, facility, or institution could improve many projects simultaneously.

For donors seeking broad and lasting impact, infrastructure and evaluation may sometimes offer more leverage than selecting another individual research project. However, neither is useful without researchers producing work to support and evaluate.

Three Different Ways to Fund Science

The three categories solve different problems.

Funding target	Primary function	Typical examples
Research	Produces new scientific knowledge	Experiments, proofs, fieldwork, software development
Evaluation	Determines what work is credible or valuable	Replication, peer review, evidence synthesis, research assessment
Infrastructure	Makes future research possible or more efficient	Databases, laboratories, instruments, repositories, software and funding platforms

The distinction is not always absolute. A replication project is both research and evaluation. A scientific database may require continuous research to remain useful. A funding platform can simultaneously provide infrastructure and evaluate researchers.

Nevertheless, distinguishing the categories helps donors identify what kind of bottleneck their money is intended to remove.

When Donors Should Fund Research Directly

Direct research funding is the most intuitive option. A donor identifies a promising scientist, laboratory, or project and pays for the work.

This approach is particularly appropriate when:

• the research question is important and neglected;
• a credible researcher has a concrete plan;
• the required resources are relatively modest;
• existing institutions will not support the work;
• progress can be observed without forcing artificial short-term milestones.

Direct funding can be especially valuable for independent researchers, early-stage ideas, fundamental mathematics, research software, and unconventional projects that do not fit standard grant categories.

The main advantage is proximity to the actual work. Money can pay for researcher time, equipment, computation, data collection, publication, or collaboration.

Its main weakness is selection risk. A donor may not have enough expertise to determine whether a proposal is sound, whether the researcher can execute it, or whether another project would produce more value. Conventional peer review attempts to solve this problem, but it can also be slow, conservative, expensive, and influenced by institutional reputation.

Direct project funding is therefore strongest when the donor possesses relevant expertise or relies on a trustworthy evaluation mechanism.

When Evaluation Is the Scientific Bottleneck

Producing research is not enough. Scientific communities must also determine:

• whether a result is correct;
• whether it reproduces;
• how important it is;
• which earlier work it depends on;
• whether data and code support the claims;
• which researchers made the relevant contributions.

Evaluation includes peer review, replication, post-publication criticism, systematic reviews, evidence synthesis, research audits, and the empirical study of research funding itself.

The United Kingdom’s Metascience Unit, established jointly by UK Research and Innovation and the government, explicitly studies how to improve the efficiency, effectiveness, and inclusivity of research and development. Its current programmes investigate funding systems, institutional design, scientific measurement, and the effects of AI on research.

This illustrates an important principle:

Funding better decisions about science can improve the use of every research dollar allocated through those decisions.

Evaluation can therefore have multiplicative value. A better assessment system may redirect funding from weak projects to strong ones, identify neglected contributors, detect unreliable findings, and help donors avoid repeatedly financing fashionable but low-value work.

However, evaluation has limitations.

First, evaluation can become bureaucracy. Researchers may spend increasing amounts of time preparing applications, reports, impact statements, and compliance documents rather than doing research. Studies of research environments have noted substantial opportunity costs associated with grant preparation.

Second, evaluation can reward what is easiest to measure rather than what is most valuable. Citation counts, journal prestige, short-term deliverables, and institutional affiliation are imperfect proxies for scientific merit. The National Academies has warned that an excessive emphasis on short-term measurable outcomes can skew funding away from long-term research.

Third, evaluators themselves need incentives, scrutiny, and evaluation. Creating another committee does not automatically create a better funding system.

Donors should fund evaluation when it produces actionable information, not merely additional paperwork.

Why Scientific Infrastructure Can Have Multiplicative Impact

Scientific infrastructure consists of resources that support multiple researchers or projects. It includes:

• shared laboratory equipment;
• telescopes and research facilities;
• computing and storage systems;
• curated datasets;
• open-source scientific software;
• publication and repository systems;
• identity and attribution systems;
• replication networks;
• funding and research-assessment platforms.

Infrastructure can generate unusually high leverage because its benefits are reusable.

A microscope funded for one experiment helps one research team. A shared microscopy facility may support dozens of laboratories. A well-maintained dataset or software library may support thousands of analyses. A better scientific funding mechanism could influence an entire portfolio of research.

The US National Institutes of Health operates Shared Instrumentation Programs because expensive equipment can be more cost-effective when purchased for shared use. NIH states that instruments awarded through these programmes must be shared and benefit thousands of investigators.

Similarly, the US National Science Foundation funds research cyberinfrastructure—including computing, data systems, software, networks, and cybersecurity—because such systems enable scientific work at scale.

Infrastructure also has strategic value where markets provide insufficient support. A commercial company may finance software that generates direct revenue, but it may not maintain a mathematical library, open dataset, research repository, or funding mechanism whose benefits are widely dispersed.

This makes scientific infrastructure a form of public good: many people can benefit, while no single user has enough incentive to finance it alone.

Infrastructure Is Not Automatically the Best Investment

The word “infrastructure” can make a project sound foundational even when few researchers actually need it.

Infrastructure projects can fail because:

• they are built before demand is demonstrated;
• maintenance is underestimated;
• users are expected to change established workflows without sufficient benefit;
• governance becomes centralized or unaccountable;
• the project duplicates an existing service;
• the infrastructure survives administratively but becomes scientifically obsolete.

The relevant question is therefore not whether a project calls itself infrastructure, but whether it removes a genuine shared bottleneck.

A strong infrastructure proposal should answer:

1. Who will use it?
2. What work is currently impossible, expensive, or unnecessarily difficult?
3. Why can existing systems not solve the problem?
4. How will the infrastructure be maintained?
5. Can users export their data and leave the system?
6. How will scientific benefit be measured without relying on vanity metrics?
7. What happens if the original team stops operating it?

Infrastructure should be evaluated as infrastructure—not as a one-time research project with a launch date and no maintenance plan.

Research Funding Systems Are Themselves Infrastructure

Funding mechanisms are often treated as administrative overhead surrounding science. In reality, they determine:

• who can participate;
• which topics receive attention;
• whether unconventional work is considered;
• how quickly researchers receive support;
• whether past contributions are recognized;
• how much researcher time is consumed by applications;
• whether decisions can be audited.

A funding system is therefore part of scientific infrastructure.

AI Internet-Meritocracy proposes treating research funding as a continuously operating system rather than a sequence of isolated grant competitions. It aims to evaluate publicly attributable scientific and open-source contributions and distribute funding according to demonstrated work rather than only institutional status or proposal-writing ability.

This model combines all three funding categories:

• researchers receive money to continue producing work;
• evaluation determines the relative value of contributions;
• the software and governance mechanism constitute reusable scientific infrastructure.

The approach remains experimental and must be tested for manipulation, bias, inaccurate evaluation, governance failures, and concentration of rewards. Nevertheless, it demonstrates why the distinction between “funding science” and “funding the system that funds science” is important.

A donor supporting a better allocation mechanism may indirectly support many future researchers rather than choosing one project personally.

Which Option Has the Highest Expected Impact?

There is no universal ranking, but the following decision rule is useful.

Fund research when the opportunity is clearer than the system

Choose direct research funding when you have strong evidence that a particular researcher or project is valuable and underfunded. Do not delay obvious work merely because an ideal evaluation or infrastructure system does not yet exist.

Fund evaluation when uncertainty is the main problem

Choose evaluation when funds are available but decision-makers cannot reliably distinguish strong work from weak work, or when important published work remains unverified and poorly understood.

Evaluation is particularly valuable when its findings will change future funding, publication, clinical, or policy decisions.

Fund infrastructure when many projects share the same bottleneck

Choose infrastructure when one investment can reduce costs, expand access, or increase research quality across a broad community.

The strongest infrastructure projects have demonstrated demand, open access where practical, credible governance, interoperability, and a realistic maintenance model.

A Portfolio Approach for Science Donors

A donor who lacks a compelling reason to specialize can use a diversified model. For example:

• 50% for direct research and researchers
• 20% for evaluation, replication, and metascience
• 30% for shared scientific infrastructure

These percentages are a heuristic, not an empirically established optimum. The allocation should change with the field.

A mature experimental discipline with expensive shared equipment may require more infrastructure. A field suffering from unreliable results may need more replication and evaluation. An emerging theoretical field with capable but unsupported researchers may justify a larger direct-research allocation.

Donors with very small budgets should usually avoid dividing each donation into tiny pieces. Instead, they can select an organization whose overall portfolio already combines research, evaluation, and infrastructure.

Questions Donors Should Ask Before Contributing

Regardless of category, donors should examine:

• Counterfactual impact: What would probably happen without the donation?
• Neglectedness: Is the work already adequately funded?
• Scientific credibility: Is there substantive, inspectable work behind the claims?
• Openness: Will findings, data, software, or evaluation results be publicly accessible where possible?
• Governance: Who controls the money and resolves disputes?
• Maintenance: Who pays recurring costs after the initial donation?
• Evaluation: How will success and failure be detected?
• Adaptability: Can funding be redirected when evidence changes?

A useful donation does not merely support something associated with science. It changes what science can discover, verify, or build.

Conclusion: Fund the Bottleneck, Not the Label

Donors should not assume that direct research is always closest to scientific impact, that evaluation is merely overhead, or that infrastructure is automatically transformative.

Each category serves a necessary function:

Research creates knowledge. Evaluation makes scientific judgment more reliable. Infrastructure increases the capacity of the entire system.

The best donation targets the constraint currently preventing valuable science from happening.

When a specific researcher has a credible but neglected opportunity, fund the researcher. When poor selection or weak verification wastes existing resources, fund evaluation. When many researchers repeatedly face the same limitation, fund shared infrastructure.

In many cases, the highest-impact strategy will combine all three—and support institutions capable of learning how the balance should change over time.