NSF Put a Private Foundation Inside One of Its Own Solicitations. The Rules It Wrote to Make That Work Are the Real Story.

Federal agencies and private foundations co-fund things constantly. The usual arrangement is parallel: the agency runs its competition, the foundation runs its own, and if the same grantee wins both, everybody puts out a press release about alignment. The two funding streams stay in separate legal boxes.

NSF 26-501 — the Collaboratory to Advance Mathematics Education and Learning (CAMEL) for K-12 — does something structurally different. The Walton Family Foundation's money is inside the NSF solicitation. And the terms NSF wrote to make that legally and ethically workable are more interesting than the program's $9 million topline.

The program, briefly

CAMEL was announced on December 17, 2025, launched out of NSF's Directorate for Social, Behavioral, and Economic Sciences. The solicitation posted December 12, 2025.

The premise: K-12 mathematics research is starved of the kind of data that modern machine learning methods require. Not data in general — American education systems generate enormous quantities of it — but AI-ready data, meaning datasets with the quality annotations, structure, and documentation needed to actually train and evaluate models. CAMEL funds the construction of that resource base.

The numbers:

• $9 million total program funding for FY2026
• 6 to 7 anticipated awards
• Up to $1,500,000 per award in total costs
• Up to three years duration
• No cost sharing or matching required — and, notably, voluntary cost sharing is prohibited

Eligibility runs to U.S. accredited institutions of higher education, non-profit non-academic organizations with an educational or research focus, and federally recognized Tribal Nations.

The program's phase structure gates access. Phase I funds the creation of cross-disciplinary research networks that generate high-value K-12 math datasets, via two pathways: 1a, novel datasets addressing current gaps, and 1b, re-purposing existing datasets for new use cases. Phase II establishes the "collaboratory" itself — a shared national digital infrastructure with a National Coordinator, enabling cross-network collaboration.

Only Phase I awardees may submit Phase II proposals. Phase I required a mandatory Letter of Intent by January 9, 2026, with full proposals due March 11, 2026. Phase II proposals are accepted anytime, which for the teams who made it through is the live opportunity.

The four-role team requirement

CAMEL will not fund a math education research group. It will not fund a machine learning lab. It requires networks that integrate expertise across four distinct communities:

1. Basic science of learning researchers — neuroscience, cognitive science, developmental science, social sciences
2. Computer science and machine learning specialists
3. Education practitioners
4. Data scientists

NSF's public framing of the program compresses this to three components — learning science researchers, education practitioners, and data scientists — but the solicitation itself is explicit about the four-way integration, with computer science and ML called out separately from data science. The distinction is substantive: building a dataset that supports rigorous quantitative analysis is a different competency from building one that trains a model.

This requirement does what similar clauses do everywhere in NSF's portfolio — it functions as a filter more than as a description. A team that assembles all four roles authentically has usually been collaborating for years. A team that adds a machine learning co-PI two weeks before the LOI deadline reads exactly like what it is. The four-role structure is how NSF distinguishes between them without having to say so in the review criteria.

Combined with the rule that an individual may appear as PI or co-PI on no more than one Phase I proposal, the effect is that senior people in this space had to pick a coalition and commit to it. Sitting on three applications to hedge was not available.

The Walton clauses

Here is where CAMEL becomes unusual as an instrument rather than just as a program.

CAMEL is supported through an agreement with philanthropic partners including the Walton Family Foundation. That is stated plainly in NSF's own announcement. And the solicitation carries three provisions that exist because of it.

First: Walton Family Foundation partners and staff cannot be involved in any capacity. Not as PI, not as co-PI, not as senior personnel, not as consultants, not as subawardees. This is a categorical exclusion of the funder's own network from the competition its money helps support. It is a clean conflict-of-interest firewall, and it is more aggressive than what most foundations impose on themselves — plenty of philanthropic programs permit grantees to have prior relationships with the funder.

Second: voluntary cost sharing is prohibited. NSF's general policy already restricts cost sharing, but stating the prohibition explicitly in a solicitation with private co-funding closes a specific loophole. Without it, an applicant with a separate philanthropic relationship could effectively bid up its own proposal by attaching outside money, converting the competition from a merit review into a partial fundraising contest. The prohibition keeps every proposal on the same financial footing.

Third: reports will be shared with the Walton Family Foundation after NSF review, and publications must carry acknowledgment language naming both WFF and NSF.

That third provision is the one that carries real obligations for awardees, and it deserves a clear-eyed reading. Nothing about it is improper — the foundation is contributing to the program and receives post-review reporting, not pre-review influence, and the sequencing ("after NSF review") is doing deliberate work. But applicants should understand it going in: your project reporting has a second audience beyond your program officer. If your institution has policies about research reporting flowing to non-federal third parties, or if your data has sensitivity constraints, that is a conversation to have with your sponsored programs office rather than a detail to discover at the first annual report.

Taken together, the three clauses form a coherent design: the funder's money is in, the funder's people are out, and the funder's visibility is limited to acknowledgment and post-review reporting.

Why this structure is spreading

The reason to pay attention to CAMEL even if you do not work in mathematics education is that it is a template, and the conditions producing it are not going away.

Federal research budgets are under sustained pressure. NSF has spent 2026 restructuring its portfolio aggressively — consolidating twelve solicitations into a $1.5 billion foundational research package that abolished submission deadlines across whole directorates, dissolving the Biological Sciences division structure, and imposing per-person annual proposal caps in Chemistry. Those are the moves of an agency managing scarcity.

Philanthropy, meanwhile, has capital and wants federal-grade peer review, which it cannot replicate internally at reasonable cost. NSF has peer review and wants capital.

CAMEL is what the trade looks like when both sides take the governance problem seriously. And $9 million is a small enough stake to be a pilot — which is very likely what it is. If the structure holds without an ethics incident, expect larger versions.

For applicants across every NSF directorate, that means a specific new item on the pre-application checklist: read the funding-source language before you build the team. A solicitation with private co-funding may exclude entire categories of collaborator you would otherwise recruit without a second thought. Discovering the exclusion after you have a signed letter of collaboration is an expensive way to learn it.

What to do now

If you hold a Phase I award, Phase II is your live window. Phase II proposals are accepted anytime and are restricted to Phase I awardees — a genuinely small, closed field. The Phase II deliverable is the national collaboratory infrastructure and a National Coordinator role, which is a fundamentally different proposition from the Phase I dataset work: it is a service-and-stewardship role, not a research role. Teams that write Phase II as "more of Phase I, at scale" will be answering the wrong question. Make the case for durable infrastructure and cross-network coordination.

If you missed Phase I, your route in is partnership, not the front door. Phase II eligibility is closed. But networks funded under Phase I need practitioner partners, school district data relationships, and technical contributors, and those relationships form through subawards and collaborations rather than through NSF competitions. The 6-7 award cohort is identifiable through NSF's award database. Approaching a funded network with a specific dataset, a district relationship, or an annotation capability they lack is a real path.

Build for the AI-ready standard, not the publication standard. The solicitation requires datasets be AI-ready with quality annotations sufficient for model training. That is a materially higher bar than "documented well enough to support the paper we wrote from it." Annotation schemas, provenance, versioning, licensing, and access infrastructure are deliverables here, not overhead. Budget them as such.

Watch for the FY2027 signal. A $9 million pilot with 6-7 awards is a structure NSF can expand, replicate in another directorate, or quietly retire. The tell will be whether a second CAMEL-style solicitation appears with a named philanthropic partner and the same three clauses. If it does, the public-private-inside-the-solicitation model has cleared its first review, and the next one will be bigger.

Related: NSF 26-524's wave-based computing program and the Innovative Postsecondary Models Fund, where eight foundations pooled $7 million.