Why the Scientific Paper May Be the Wrong Unit for Funding

Getting your Trinity Audio player ready...

The scientific paper may be the wrong unit for funding because the boundaries of a publication rarely match the boundaries of a scientific contribution. One paper can contain several discoveries; one discovery can require many papers, datasets, tools, and years of maintenance. Funding that treats publications as interchangeable units risks rewarding how research is packaged rather than what it makes possible.

A better approach is to assess documented contributions, recognize the people responsible, and separately decide what future work needs support. Papers remain essential evidence in that assessment.

What does it mean to fund the “paper”?

Research grants commonly fund projects, people, teams, or institutions, rather than purchasing individual articles. The issue arises when publication counts, journal prestige, or paper-level citations become dominant proxies for deciding who deserves funding.

Three different questions then become blurred:

• What is the evidence? Papers, software, datasets, proofs, protocols, and other records.
• What deserves credit? The contributions those records establish.
• What should receive money? People, teams, infrastructure, or future activities.

A paper can help answer all three questions, but it does not answer any of them automatically.

This concern is consistent with the San Francisco Declaration on Research Assessment, which asks funders to assess research on its merits and consider outputs beyond publications, including datasets and software. DORA supports broader assessment; it does not establish a particular funding formula.

Why publications and scientific value do not align

One contribution can appear in many publications

Imagine two researchers producing equally useful advances. One explains the advance in a comprehensive paper; the other divides it into five articles.

If assessment mechanically rewards publication count, the second researcher receives an advantage without necessarily contributing more knowledge. The same problem arises when a preprint, conference paper, and journal article report substantially overlapping work.

The appropriate question is how much distinct, reliable knowledge was added. A funding system should recognize extensions while avoiding repeated credit for the same underlying result.

One paper can conceal many different contributions

A published discovery may depend on experimental design, instrument development, data collection, software, mathematical analysis, and validation. An author list alone does not explain who performed each task or how much it mattered.

The Contributor Role Taxonomy, CRediT, provides 14 roles for describing contributions to research outputs. This makes participation more explicit, but a role label does not determine a percentage of credit or a payment.

Knowing that two people contributed software, for example, does not tell us whether one built the central algorithm and the other corrected a minor defect. Both deserve accurate attribution; their contributions still require assessment.

Valuable work can continue after publication

A software paper describes a tool at a particular moment. Its usefulness may subsequently depend on years of debugging, compatibility updates, documentation, and user support.

If funding follows only the original paper, later maintainers can become difficult to recognize—even when their work keeps the tool usable.

The peer-reviewed Software Citation Principles argue that software should be recognized as a legitimate research product and that citations should support attribution and identification of specific versions. For funding, this suggests evaluating the software and its development history alongside its accompanying article.

The same reasoning applies to curated databases, reference collections, and experimental infrastructure.

Scientific importance can emerge slowly

An abstract mathematical result may become useful long after publication. A carefully conducted replication may strengthen confidence without producing a new headline. A correction may prevent other researchers from pursuing an invalid argument.

These contributions require different evidence of value. A short publication window or a single citation threshold cannot fairly represent all of them.

However, delayed recognition is not proof of importance. A broader assessment system must remain open to overlooked work while distinguishing supported contributions from unverified claims.

An example: who enabled the discovery?

Consider a hypothetical research project:

Contribution	Evidence beyond the final paper	What funding assessment should examine
Develops a reusable algorithm	Derivation, code, benchmarks	Correctness and usefulness
Produces a reliable dataset	Provenance, methods, quality checks	Reliability and reuse
Maintains essential software	Releases, issue resolution, tests	Continued functionality and substantive improvements
Makes the final discovery	Analysis and supporting results	Novelty, validity, and significance
Independently checks the result	Replication records or proof checks	What uncertainty the work resolves

The final paper brings these inputs together. It does not make every upstream contributor an author, nor does it measure their relative importance.

The record that announces a discovery is not a complete account of the work that enabled it.

This does not mean every dependency deserves an equal payment. It means publication boundaries should not decide in advance whose work counts.

A better unit: documented contributions within a body of work

Contribution-based funding would assess identifiable advances and services within the researcher’s broader record. That matters because dividing research into ever-smaller “contribution units” could reproduce the incentives of paper counting.

A practical assessment should ask:

1. What was contributed? Identify the result, resource, improvement, or validation.
2. Who contributed it? Examine attribution and supporting records.
3. How reliable is it? Consider appropriate checks, limitations, and unresolved disputes.
4. What does it enable? Examine demonstrated uses and explain uncertainty about potential uses.
5. Has it already been counted? Connect overlapping publications and versions.

Scientific dependency graphs—records linking results to the methods, data, and tools they rely on—could help organize this evidence.

But a dependency graph is not a ready-made payment formula. A citation does not prove necessity, an uncited dependency may still matter, and several contributors can each be indispensable. Allocating credit requires explicit judgments about shared value and uncertainty.

What this means for AI Internet-Meritocracy

AI Internet-Meritocracy, or AIIM, is Science DAO’s experimental approach to funding researchers and open-source developers using AI-assisted assessments of documented contributions.

Its relevance here is the possibility of evaluating a person’s research and software beyond publication counts. Whether that approach produces fairer or more effective allocations remains a question for testing.

For AIIM, the argument implies concrete design goals: connect duplicate records, recognize substantive software maintenance, examine individual contributions, and make assessments open to correction. These are proposed requirements, not a claim that every capability is already implemented.

AI also introduces its own risks. It can confuse fluent descriptions with strong evidence, overlook poorly documented work, or reproduce prestige bias. Replacing a publication count with an unexplained AI score would leave the underlying assessment problem unresolved.

Credit for past work and funding for future work differ

Even an excellent assessment of past contributions cannot determine every funding decision.

An early-career researcher may need support before producing a substantial record. A costly experiment may deserve funding despite its uncertain outcome. Essential infrastructure may require predictable operating support.

A sensible funding portfolio can therefore combine contribution-based rewards with prospective grants, support for emerging researchers, and infrastructure budgets.

The strongest case against paper-centered funding is not that papers lack value. It is that scientific funding should follow credible contributions and research needs, with publications serving as evidence rather than interchangeable units of merit.