Accelerating the American Scientific Enterprise

A response to the White House Office of Science and Technology Policy Request for Information

The White House recently asked for community input on the future of the American scientific enterprise. Below is our response.

---

The United States is the global leader in science, technology, and innovation. Investments in R&D in the 20th century put a man on the moon, created a medical renaissance, and launched the digital and the AI era. Yet despite these successes, the structure of the research enterprise today has become increasingly strained.

Today, scientists spend nearly a third of their time writing grant applications and are often unable to pursue their most ambitious ideas, as research increasingly follows the path of least funding resistance. Funders struggle to coordinate with one another, and the public has little visibility into or engagement with the process. The result is an opaque, low-throughput, high-friction system, with funding increasingly concentrated among a narrow set of institutions. The United States needs a fundamental restructuring of the scientific enterprise to unlock its full potential, maximize innovation, and drive economic growth.

America has long demonstrated the power of competitive markets and open experimentation. Capital markets show how to allocate resources toward the most productive ventures, tech platforms demonstrate how to build networks that connect billions of people, and startups illustrate the value of rapid iteration. These lessons, however, have not carried over into scientific research, where the infrastructure needed to connect ideas, funders, capabilities, and outcomes remains fragmented or nonexistent.

This gap is becoming untenable. New AI systems, such as those supported by the Genesis Mission, are poised to generate innumerable scientific hypotheses and discoveries. Every theoretical advance will require people, physical resources, and funding to be validated in the laboratory and translated into real-world applications. Without a corresponding upgrade to how research is prioritized, resourced, and executed, the scientific enterprise risks buckling under the weight of its own productivity.

What is required is a networked operating system for science: an integrated system that connects knowledge, funding, execution, and decision-making into a single learning infrastructure. Researchers should be able to surface bold ideas; funders of all types should be able to prioritize and support them; and resources should be mobilized to turn promising concepts into real experiments and outcomes.

Coordinated by AI-enabled decision-making, such a system would allow the scientific enterprise to operate at the scale and speed required for modern discovery and sustain U.S. scientific and technological leadership.

To achieve this vision, we propose that the United States:

1. Create a Widely Accessible Knowledge Layer (Question (v) &(viii))

AI cannot optimize science if data remains trapped in PDFs or as undocumented tacit knowledge among researchers and funders; we need a “Knowledge Layer” that captures the full lifecycle of research in a machine-readable graph.

Policy and Action Recommendations:

• Publish funding-decision metadata: Require agencies to release (with appropriate privacy and IP safeguards) structured metadata, including topic tags, scoring distributions, panel type, and reason codes, on all submissions.

• Enforce Machine-Readability via the GREAT Act: The Grant Reporting Efficiency and Agreements Transparency (GREAT) Act of 2019 mandates that grant reporting information be fully searchable and machine-readable. Agencies can interpret “federal award information” to include the application phase.

• Build Upon Proto-OKN: The NSF’s Proto-Open Knowledge Network (Proto-OKN) has demonstrated the viability of linking disparate, multi-agency federal data sources. The Administration should commission a permanent, enterprise-grade open knowledge network combining federal, state, academic, and commercial data sources .

• Unlock “Dark Data” (Negative Results): Align data policies with the “Restoring Gold Standard Science” Executive Order. This order lists accepting negative results as positive outcomes as a core tenet of scientific integrity.

• Securely Analyze Unfunded Proposals: Agencies can utilize the National Secure Data Service demo project to place unfunded proposal data within a secure enclave. This allows AI models to train on aggregate patterns of “near-miss” proposals without exposing specific IP or violating FOIA Exemption 4.

2. Create a Standardized Research Container (Question (ix))

An essential part of a functional Knowledge Layer is a set of common data standards. Researchers are currently trapped in endless cycles of bespoke grant-writing, and funders lack visibility into research data due to variable data structures across agencies and funding institutions. The solution is to make research data (proposals, evaluations, methods, and results) portable across different agencies and private funders.

Policy and Action Recommendations

• Adopt and Expand “RO-Crate” as the Technical Standard: The Research Object Crate (RO-Crate) is a community-led, open-source specification using JSON-LD to package research artifacts. Strong federal precedents exist, such as the NIH-funded Cell Maps for Artificial Intelligence project. We should designate RO-Crate as the foundation for the “Universal Grant Container,” enabling a single proposal package to be routed to NSF, DOE, NIH, and other funders simultaneously.

• Harmonize with FIBF: The Federal Integrated Business Framework defines the data dictionary for grants without a transport mechanism. The government should combine FIBF definitions with the RO-Crate technical container. This creates a standard that is legally compliant in its terminology and technically portable in its structure.

3. Create an Innovation Marketplace (Question (iv), (vii) and (x))

Current funding mechanisms favor large universities with administrative machinery. A marketplace model “unbundles” the funding process, allowing diverse entities to compete for funding based on the merit of their ideas rather than their institutional affiliation. This lowers the barrier to entry and allows the government to tap into innovation clusters outside traditional academic centers.

Policy and Action Recommendations

• Introduce a centralized digital marketplace for research: Leverage the constructed Knowledge Layer and Research Container to provide more visibility to more funders, allowing a marketplace to determine the most efficient allocation of resources.

• Scale the DOE “Small Business Vouchers” Model: The DOE Small Business Vouchers pilot awarded vouchers to 114 small businesses to buy capabilities from National Labs. The model should be re-authorized and expanded outside DOE. Awardees should include researchers outside traditional innovation hubs who could redeem federal credits for technical assistance at any FFRDC or university core facility, disseminating the expertise of major research institutions and economic benefits to previously-neglected regions.

• Leverage GSA Schedules for a “Shared Research Economy”: Agencies should utilize GSA Multiple Award Schedules to create a friction-free marketplace for laboratory services and equipment leasing. By pre-approving a diverse network of providers, including university core facilities and cloud labs, agencies enable PIs to contract out routine experiments or lease hardware rather than purchasing redundant equipment. This maximizes the utilization of existing national infrastructure and reduces capital waste.

• Deploy Consortium-Based OTAs for Rapid R&D: While GSA schedules streamline services, agencies should utilize Other Transaction Authorities (OTAs) to streamline innovation. Following the model of the Medical Technology Enterprise Consortium (MTEC), civilian agencies should establish consortium-based OTAs to create standing “marketplaces of solvers.” This allows agencies to issue challenge statements to a pre-vetted pool of non-traditional vendors (startups, academic labs, industry) who can form teams and launch new R&D prototypes in weeks rather than months.

4. Apply Index Logic for Research at all Stages (Question (vi))

Current funding mechanisms usually target individual projects that meet a specific program-specific quality threshold (that often punishes high risk endeavors). Federal agencies should encourage funding of research portfolios in which a set of promising approaches to a given problem (e.g. Alzheimer’s treatment or efficient battery design) can be grouped together and funded as a bloc rather than individually. Just as there are competing investment funds in public markets, so should there be competing portfolios to allow funders to easily choose among many various funding strategies.

Policy and Action Recommendations

• Establish Portfolio Selection Authority: Agencies should utilize Broad Agency Announcements (BAAs) and Other Transaction Authorities (OTAs) to solicit and review “thematic portfolios” rather than isolated projects. This empowers Program Managers to exercise “portfolio selection”, recommending a slate of projects that balances high-risk/high-reward bets with safer plays, rather than being bound strictly by individual peer review scores.

• Operationalize “Classes of Projects”: Agencies should leverage authorities similar to those granted to ARPA-H to fund “classes of projects.” This legal distinction allows an agency to review, approve, and fund a cohesive index of research (e.g., “all mRNA delivery mechanisms”) as a single strategic asset, rather than processing hundreds of disparate grants.

• Make research portfolios accessible to private funders. Agencies should make portfolios publicly fundable, lowering the barrier to entry for new funders, like index funds and ETFs did for the stock market. Individuals could access portfolios curated by trusted entities tied to cause areas they care about, massively increasing public engagement in R&D.

5. Promote Synergies through Syndication (Question (i), (ii) and (vi))

In venture capital and private equity markets syndication is the norm with multiple investors participating in any individual deal which allows for greater diversification and shared learning between firms. This rarely happens in research funding, constraining the number of entities that can fund large-scale experiments. Agencies should be able to participate in shared, open platforms where their validation serves as a signal to crowd in private investment for high-quality work.

Policy and Action Recommendations

• Create Syndicated Funding Pools for research: Agencies should leverage time spent in due diligence by producing a syndication-ready signal, including summary, assessment, and risks, that industry, philanthropies and family offices can co-fund.

• Empower “Agency-Related Foundations” as Syndicate Leads: Replicate and expand the Foundation for Energy Security and Innovation (FESI) model, which possesses statutory authority to transfer funds bi-directionally between the agency and the private sector. Agencies should designate their affiliated foundations (like the FNIH) to act as capital aggregators packaging federal initiatives to attract private capital.

• Crowd-in Funding via Prize Authority: Utilize 15 U.S.C. § 3719 (Prize Authority), which explicitly allows agencies to design competitions that accept private funds. Agencies can vet proposals with a “finalist” status, then open the portfolio to public matching dollars or philanthropic top-ups without navigating complex gift acceptance rules.

• Expand Partnership Intermediary Agreements (PIAs): For agencies without a chartered foundation, 15 U.S.C. § 3715 allows federal labs to partner with non-profits via Partnership Intermediary Agreements. Labs should use PIAs to matchmake federal technical assistance with regional economic development funding.

6. The AI Program Manager (Responding to Q. viii & Q. v)

Program managers are the cornerstone of ARPA agencies. They do the hard work of analyzing literature, identifying gaps in understanding, and structuring the financial incentives and performance management to move an initiative from zero to one. To expand these capabilities we should invest in “AI Program Manager” infrastructure that functions as an “air-traffic control” system for research. Over time, capabilities could include: summarizing literature, structuring portfolios and incentive structures, and simulating outcomes of comparative approaches.

Policy and Action Recommendations

• Integrate AI Program Managers into Genesis Mission Infrastructure. This represents a flagship use-case for the goal of creating an “integrated AI platform” for scientific discovery. Agencies should leverage the DOE’s American Science and Security Platform to host the “AI Program Manager” backend.

• Unlock “Evidence-Based” Funding: To fund this infrastructure immediately, OMB should issue guidance classifying AI management tools as “Evidence-Building” activities under the Foundations for Evidence-Based Policymaking Act. This allows agencies to utilize existing set-aside funds for program evaluation to build these AI capabilities.

• Scale Existing NIH Precedents: The NIH Assisted Referral Tool (ART) already uses AI to analyze grant text and route proposals to review panels. Similarly, the NIH Office of Research Information Systems (ORIS) uses text mining to forecast emerging scientific trends. These existing operational tools demonstrate that “AI Program Management” is a proven concept that must now be scaled government-wide.

Taken together, these recommendations outline a practical path to modernizing how the United States funds, performs, and translates research drawing on approaches that are already being developed by ourselves and others. Implementing this modernization would make the American research system more effective and reinforce our nation’s position as the world’s leading engine of scientific and technological progress.

---

This Substack series by Catalyze explores how science can work better to solve our most pressing problems, through new forms of coordination, capital, and intelligence.