Why Research Funding Should Follow Results, Not Promises

Getting your Trinity Audio player ready...

Most research funding is distributed before the funded work exists. Scientists submit proposals describing what they expect to discover, how they plan to proceed, and why their future work deserves support. Committees then allocate money on the basis of those promises.

This model is sometimes necessary. Laboratories cannot purchase equipment, hire researchers, or conduct expensive experiments without advance capital. But prospective grants have a fundamental limitation:

A research proposal is evidence of planning and persuasive writing—not evidence that the proposed scientific result will actually be produced.

Retroactive funding reverses the sequence. Instead of attempting to predict which proposal will succeed, it rewards research after useful outputs become visible. Milestone funding and prize funding occupy intermediate positions: they connect payments to predefined achievements, but differ in when recipients are selected and what evidence is evaluated.

The strongest research-funding system is therefore unlikely to use only one mechanism. It should combine:

• traditional grants where substantial advance resources are unavoidable;
• milestone funding where progress can be divided into verifiable stages;
• prize funding where a clear problem can support open competition;
• retroactive funding where outputs, use, reproducibility, and impact can be evaluated after publication.

AI Internet-Meritocracy (AIIM) is intended to make the last model scalable. Rather than treating a grant proposal as the main unit of merit, AIIM could continuously evaluate completed scientific contributions and distribute funding according to accumulated evidence.

What Is Traditional Grant Funding?

Traditional grant funding is prospective funding: money is allocated before most or all of the proposed research has been completed.

A typical process includes:

1. publication of a funding opportunity;
2. submission of research proposals;
3. peer or administrative review;
4. selection of recipients;
5. disbursement of funds;
6. progress reporting and post-award oversight.

The US National Institutes of Health describes its grant process as extending from planning and application through award administration and post-award reporting. In other words, the central funding decision is made before the expected research results are available.

This system finances work that might otherwise be impossible. It is especially important for:

• clinical trials;
• laboratory construction;
• telescope or accelerator access;
• field expeditions;
• long-term data collection;
• salaries for full-time research teams;
• expensive materials and instrumentation.

Traditional grants are not inherently defective. The problem arises when prospective selection becomes the dominant mechanism for nearly every kind of research, including work that can be evaluated more accurately after results appear.

Why Funding Promises Is Difficult

A proposal asks reviewers to estimate several uncertain quantities simultaneously:

• whether the hypothesis is sound;
• whether the methodology will work;
• whether the researchers can execute it;
• whether the results will be important;
• whether another approach would produce more value;
• whether the requested budget is proportionate;
• whether the applicant’s description is realistic rather than strategically optimistic.

Even highly competent reviewers cannot observe the future. They must use imperfect proxies such as institutional affiliation, previous publications, preliminary results, recommendation letters, writing quality, and conformity with an established research agenda.

Proposal quality is not identical to scientific quality

Successful proposal writing requires its own specialized skills. Applicants must understand the language, priorities, formatting rules, and strategic expectations of a funding institution.

These skills may correlate with research competence, but the correlation is incomplete. A researcher can be exceptionally good at mathematics, experimentation, software development, or conceptual synthesis while being relatively weak at:

• presenting speculative work as a predictable project;
• describing uncertain discoveries as scheduled deliverables;
• fitting unconventional research into established categories;
• maintaining institutional networks;
• producing polished grant narratives.

Conversely, a persuasive proposal may fail to generate important results.

The resulting system risks allocating money partly according to the ability to market future science, rather than according to the value of science actually produced.

Prospective funding can favor predictable work

Reviewers are accountable for their recommendations. When outcomes are uncertain, choosing a conventional project from an established institution can appear safer than supporting an unfamiliar researcher or a highly unorthodox idea.

This does not mean that peer reviewers deliberately suppress innovation. It means that the incentive structure can reward defensible selections. A proposal closely connected to accepted literature is easier to justify than one proposing a new conceptual framework that reviewers cannot yet evaluate confidently.

Traditional grants may therefore produce a paradox:

The more radically new a discovery would be, the less evidence may exist in advance that the proposed route will succeed.

What Is Milestone Funding?

Milestone funding releases money when a recipient completes specified stages of a project.

A milestone normally includes:

• a defined deliverable;
• completion criteria;
• a deadline;
• an associated payment;
• a verification procedure.

DARPA’s instructions for certain research agreements require milestones to specify a description, completion criteria, due date, and payment or funding schedule.

Examples of research milestones could include:

• producing a validated prototype;
• completing recruitment for a study;
• publishing a dataset under an open licence;
• demonstrating a specified measurement accuracy;
• passing an independent replication test;
• releasing documented research software;
• completing a preclinical study.

Milestone funding reduces some of the risk of a fully upfront grant. The funder does not need to release the entire budget before observing progress.

However, the initial project and recipient are still usually selected prospectively. The funder must still predict which team is likely to succeed.

Advantages of milestone funding

Milestone funding provides:

• stronger accountability than an unconditional upfront payment;
• earlier detection of failed approaches;
• clearer expectations;
• reduced financial exposure;
• opportunities to revise or terminate an underperforming project;
• incentives to produce concrete intermediate outputs.

Limitations of milestone funding

Milestones can also distort research when they are defined too rigidly.

Scientific investigation is not always linear. An unexpected negative result, failed experiment, or conceptual redirection may be more scientifically valuable than completion of the original deliverable. A team that follows the evidence could technically fail a milestone, while a team that optimizes for the contractual target could receive payment without producing comparable knowledge.

Milestones work best when:

• outputs can be defined objectively;
• the development path is reasonably understood;
• the project is engineering-oriented;
• deliverables can be independently tested;
• changes in direction can be approved without excessive bureaucracy.

They work less well for exploratory mathematics, foundational theory, open-ended discovery, and research where the important result cannot be predicted in advance.

What Is Prize Funding?

Prize funding offers a reward for solving a specified problem or satisfying defined performance criteria.

Unlike a traditional grant, a prize does not necessarily select one recipient before work begins. Multiple researchers or teams may attempt the challenge, while the reward is paid only to qualifying participants or winners.

Government and scientific institutions already use variations of this approach. For example, the US National Science Foundation’s VITAL Prize Challenge used several competitive stages, prototype development, evaluation, and final awards. Some finalists also received development funding before the final prizes were decided, making the program a hybrid of staged and prize-based funding.

Advantages of prize funding

Prize competitions can:

• attract participants outside conventional academic networks;
• define a concrete target;
• encourage parallel approaches;
• pay primarily for demonstrated performance;
• reveal unexpected solutions;
• reduce the need to predict the winning team in advance.

Prizes are especially useful when a funder knows what outcome is needed but does not know who can produce it or which method will work.

Limitations of prize funding

Prizes transfer substantial risk to participants. Researchers may need to finance unsuccessful attempts themselves. This favors:

• well-capitalized laboratories;
• companies with existing infrastructure;
• researchers who can work without immediate compensation;
• teams able to attract external investors.

A winner-takes-all design can also underreward valuable partial advances. Several teams may create useful methods, datasets, proofs, or tools even though only one receives the principal prize.

Prize funding is therefore not equivalent to general retroactive funding. A prize usually defines the desired result in advance. Retroactive funding can reward outputs whose importance was not anticipated when the work began.

What Is Retroactive Research Funding?

Retroactive funding allocates money after a scientific contribution has been produced.

The contribution might be:

• a research paper;
• a theorem or proof;
• a dataset;
• open-source scientific software;
• an experimental protocol;
• a replication study;
• a negative result;
• a literature synthesis;
• a correction of an influential error;
• a reusable scientific instrument;
• a new conceptual framework.

The funder evaluates evidence that already exists rather than relying primarily on forecasts.

This changes the central question from:

“Which applicant promises the most valuable future project?”

to:

“Which completed contributions have created the most verified scientific value?”

Retroactive funding is broader than prizes

Prize funding normally begins with a declared challenge. Retroactive funding does not require the funder to specify the winning discovery beforehand.

This distinction is crucial for basic science. Many important discoveries cannot be commissioned through precise challenge statements because nobody yet knows what form the discovery will take.

A retroactive system can recognize:

• an unexpected theorem;
• a method that becomes useful across several disciplines;
• software that gradually becomes critical research infrastructure;
• a neglected result whose significance becomes visible years later;
• an independent researcher whose work was initially overlooked;
• a negative finding that prevents others from wasting resources.

Retroactive funding is therefore particularly suitable for open-ended and decentralized discovery.

Traditional Grants vs Retroactive Funding

Criterion	Traditional grants	Retroactive funding
Main evaluation object	Proposal and applicant	Completed contribution
Funding time	Before results	After results
Primary evidence	Plans, credentials, preliminary work	Outputs, validation, use and impact
Prediction required	High	Lower
Capital available before work	Yes	Usually no
Access for independent researchers	Often limited	Potentially broader
Risk to funder	Pays for possible failure	Pays after evidence appears
Risk to researcher	Lower after award	Higher before recognition
Suitable for expensive experiments	Strong	Weak without complementary capital
Suitable for unexpected discoveries	Limited	Strong
Administrative emphasis	Application and compliance	Evaluation and attribution
Main gaming risk	Proposal gaming	Metric and attribution gaming

The comparison does not establish that retroactive funding should eliminate grants. It shows that the two mechanisms solve different problems.

Traditional grants answer:

Who should receive resources to attempt costly future work?

Retroactive funding answers:

Who has already created scientific value that deserves compensation and reinvestment?

A mature system needs both questions.

Milestone Funding vs Retroactive Funding

Milestone and retroactive funding both connect payment to observable work, but they differ in scope.

Milestone funding

• selects the project in advance;
• defines expected outputs in advance;
• pays according to contractual stages;
• works well for planned development;
• may penalize scientifically useful changes in direction.

Retroactive funding

• can evaluate contributions that were never preselected;
• does not require all valuable outputs to be predicted;
• can compare work across projects and institutions;
• can reward unexpected discoveries;
• requires robust post-publication assessment.

Milestone funding is best understood as conditional prospective funding. Retroactive funding is open-ended retrospective allocation.

Prize Funding vs Retroactive Funding

Prize and retroactive funding both reward completed performance, but a prize generally has:

• a predetermined problem;
• predetermined eligibility rules;
• explicit success criteria;
• a fixed reward pool;
• a competition deadline;
• a limited number of winners.

Retroactive funding can be continuous rather than episodic. It can distribute many smaller rewards across a network of contributors instead of selecting one winner.

For example, a prize might reward the first team to produce a specified diagnostic technology. A retroactive system might separately reward:

• the underlying statistical method;
• the open dataset;
• the replication work;
• the implementation library;
• the clinical validation;
• the documentation;
• later improvements;
• researchers who identify limitations.

Prize funding identifies a winner. Retroactive funding can model an ecosystem of contribution.

Complete Comparison of the Four Funding Models

Feature	Traditional grants	Milestone funding	Prize funding	Retroactive funding
Recipient selected before work	Yes	Yes	Usually no	No
Output defined before work	Broadly	Precisely	Precisely	Not necessarily
Payment before completion	Usually	Partly	Usually no	No
Supports initial costs	Strongly	Moderately	Weakly	Weakly
Rewards unexpected results	Sometimes	Poorly to moderately	Poorly	Strongly
Encourages multiple approaches	Limited	Limited	Strongly	Strongly
Can reward partial contributions	Yes	Yes, if defined	Often limited	Yes
Requires proposal review	Extensive	Extensive	Limited or staged	Minimal
Requires post-result evaluation	Moderate	Strong	Strong	Very strong
Best use	Resource-intensive uncertain work	Verifiable development stages	Clearly defined challenges	Open-ended completed contributions

Which Research Funding Model Works Best?

There is no universally superior mechanism. The best model depends on the type of uncertainty involved.

Use traditional grants when resources must come first

Prospective grants remain essential when researchers cannot reasonably begin without substantial funding.

Examples include:

• constructing research facilities;
• running clinical trials;
• maintaining long-term observatories;
• conducting expensive biological experiments;
• collecting data over many years;
• supporting researchers in fields without outside income.

In such cases, “fund only completed results” would exclude capable scientists who lack personal or institutional capital.

Use milestone funding when progress can be verified in stages

Milestone funding is appropriate when a project has a discernible path but the funder wants to limit exposure.

It is particularly suitable for:

• translational research;
• prototype development;
• scientific engineering;
• infrastructure deployment;
• benchmark-driven AI research;
• instrument construction.

Use prizes when the target is clear but the solution is unknown

Prize funding works best when:

• success can be tested objectively;
• many teams can attempt different methods;
• the problem is important enough to attract participation;
• the target does not prescribe the scientific route;
• unsuccessful attempts do not require unsustainable expenditure.

Use retroactive funding when valuable outputs cannot be predicted

Retroactive funding is strongest for:

• basic mathematics;
• theoretical research;
• open-source scientific software;
• replication and verification;
• independent scholarship;
• interdisciplinary synthesis;
• unexpected discoveries;
• contributions that acquire value through later use.

These areas often produce outputs that are difficult to describe convincingly before they exist.

Why Retroactive Funding Alone Is Not Enough

A purely retroactive system creates a financing gap. Researchers still need food, housing, equipment, computing, laboratory access, and time before their results can be evaluated.

It may also favor researchers who already possess:

• university salaries;
• personal savings;
• institutional infrastructure;
• commercial revenue;
• wealthy sponsors;
• substantial volunteer communities.

Therefore, retroactive funding should not be presented as a complete replacement for prospective capital.

A more defensible principle is:

Research funding should follow results whenever results can already be evaluated, while advance funding should be reserved for work that genuinely cannot proceed without it.

Retroactive rewards can also finance future research indirectly. A researcher who is rewarded for one completed contribution gains resources to pursue the next project without writing a conventional proposal. In this way, retrospective payment becomes prospective scientific capacity.

How AIIM Could Implement Retroactive Research Funding

AIIM proposes an automated and continuously updated funding mechanism for scientific and open-source contributions.

Rather than organizing all funding around occasional grant competitions, AIIM could evaluate published outputs using multiple forms of evidence.

Potential evaluation dimensions include:

• scientific validity;
• novelty;
• reproducibility;
• methodological rigor;
• usefulness to later research;
• quality of data and code;
• successful independent replication;
• correction of previous errors;
• downstream adoption;
• contribution to public scientific infrastructure;
• expert criticism and responses;
• uncertainty in the evaluation itself.

AIIM would not need to declare that an output has one final, permanent value. Evaluations could change as evidence accumulates.

For example:

1. a paper receives an initial evaluation;
2. independent researchers inspect its reasoning or experiments;
3. replications increase or decrease confidence;
4. software dependencies reveal practical utility;
5. later papers confirm, refine, or reject its claims;
6. funding is adjusted according to the updated evidence.

This creates a model of continuous post-publication funding rather than a single grant decision.

The comparison between AIIM and Horizon Europe describes a related distinction: conventional programs operate through calls, applications, work packages, and grant agreements, while AIIM could use continuous micro-funding and retroactive rewards as evidence develops.

AIIM as More Than a Scientific Prize

AIIM can resemble a scientific prize because it rewards achievement rather than rhetoric. However, it differs from a conventional prize in several ways.

A traditional prize normally:

• addresses one predefined challenge;
• has a closing date;
• selects one or several winners;
• distributes a predetermined sum;
• ends after the competition.

AIIM could instead:

• evaluate contributions across many scientific fields;
• operate continuously;
• distribute proportional rewards;
• revisit earlier assessments;
• recognize dependencies among contributions;
• reward both visible breakthroughs and supporting infrastructure.

The Science DAO article “AIIM Is an Automated Scientific Prize” describes AIIM as combining retrospective recognition with ongoing research incentives.

The critical difference is that AIIM would not merely award a trophy-sized payment to a small number of famous discoveries. It could make retroactive evaluation part of the ordinary economic infrastructure of science.

Can AI Evaluation Replace Human Judgment?

No funding mechanism becomes objective merely because it uses AI.

An AI system can inherit errors from:

• training data;
• publication bias;
• citation patterns;
• incomplete scientific records;
• prestige-correlated language;
• manipulated benchmarks;
• coordinated voting;
• fabricated papers;
• hidden conflicts of interest;
• poorly specified evaluation criteria.

AIIM should therefore not be described as an infallible scientific judge. Its advantage would need to come from auditable procedure, not presumed machine superiority.

A credible implementation would require:

• transparent evaluation criteria;
• uncertainty estimates;
• evidence-linked explanations;
• adversarial testing;
• appeal mechanisms;
• independent replication;
• plural models or agents;
• safeguards against prompt gaming;
• human review for high-impact decisions;
• public records of major scoring changes.

The goal is not to eliminate judgment. It is to make funding judgments more continuous, evidence-based, inspectable, and correctable.

A Hybrid Funding Architecture for Science

The most effective architecture would connect all four mechanisms.

Layer 1: Seed and infrastructure grants

Small prospective grants provide researchers with minimum starting resources. Larger grants remain available for capital-intensive work.

Layer 2: Milestone payments

Projects with identifiable development paths receive additional funding after verified intermediate outputs.

Layer 3: Open prizes

Clearly defined technical or scientific challenges invite competing solutions from any qualified participant.

Layer 4: AI-assisted retroactive rewards

Completed papers, datasets, proofs, software, replications, and other contributions receive continuous evaluation and proportional funding.

This structure avoids two extreme errors:

• funding everything on the basis of promises;
• expecting researchers to finance all socially valuable work themselves.

It also gives researchers several routes to support. A person excluded from a traditional grant may still qualify through a milestone program, win an open challenge, or receive retroactive rewards for completed work.

Funding Results Changes Research Incentives

A proposal-centered system tells researchers:

Predict what funders will consider promising.

A results-centered system tells researchers:

Produce work that survives examination and becomes useful.

That difference could affect scientific behavior.

Results-based funding can encourage researchers to:

• publish usable data;
• release code;
• document methods;
• conduct replications;
• correct mistakes;
• develop shared infrastructure;
• pursue neglected problems;
• make outputs easier to verify;
• create work that remains valuable beyond a grant period.

However, poorly designed metrics could replace proposal gaming with output gaming. Researchers might optimize for superficial indicators such as raw citation counts, repository stars, publication volume, or AI-generated assessments.

AIIM must therefore evaluate structured evidence, not a single popularity metric.

Conclusion: Fund Promises When Necessary, Results Whenever Possible

Traditional grants solve a real problem: many forms of research require resources before they can produce evidence.

But the grant model is often asked to solve a different problem for which it is poorly suited—determining the value of work that has already been completed. Once outputs exist, continuing to allocate rewards primarily through new promises wastes available evidence.

Milestone funding improves accountability by tying payments to predefined progress. Prize funding allows open competition around defined goals. Retroactive funding goes further by rewarding valuable contributions even when their form or significance could not have been predicted.

The appropriate principle is not “abolish all grants.” It is:

Use prospective funding to provide necessary research capacity, but use retroactive funding to reward demonstrated scientific value.

AIIM could make this division practical at scale. By evaluating completed contributions continuously, transparently, and across institutional boundaries, it could help research funding follow results—not merely the ability to promise them.