In July 1945, Vannevar Bush handed President Truman a report called Science, The Endless Frontier. It argued that the federal government should fund basic research through competitive grants administered by expert panels. The National Science Foundation was born. The modern grant system was born. The entire architecture that still determines how billions of dollars flow to scientists today was born.
That was eighty-one years ago.
Think about what the world looked like in 1945. There were roughly 5,000 active research scientists in the United States. The entire federal R&D budget was a fraction of what a single mid-sized NIH institute spends today. Scientific communication happened through printed journals mailed to university libraries. Computing meant a room full of people with slide rules.
The system Bush designed made sense for that world. A small community of researchers, known to each other by reputation, submitting proposals to panels of peers who could reasonably evaluate them. Low volume. Slow pace. Manageable scale.
Now look at today. There are over 1.5 million active researchers in the US alone. The NIH receives roughly 50,000 grant applications per year and funds about 20%. The NSF success rate hovers around 25%. Reviewers are overwhelmed. Panel fatigue is real. The average age of a first NIH R01 grant has climbed to 42 — meaning researchers spend their most creative decade begging for scraps.
The volume alone breaks the 1945 model. But the deeper problem is structural.
Bush assumed science was best evaluated prospectively — that expert committees could predict which ideas would succeed before any work was done. Eight decades of evidence say otherwise. Study after study shows that peer review of grant proposals is essentially random for proposals above a quality threshold. A 2016 analysis of NIH review scores found that changing the reviewer panel would change the funding decision for roughly half of all borderline applications.
We are running prediction software on hardware built for a world that no longer exists.
Other fields have moved past prospective-only evaluation. Software development discovered that shipping code and measuring usage beats writing elaborate design documents. Venture capital learned that small initial bets with follow-on funding for demonstrated traction outperforms committing large sums to untested business plans. Even government procurement has started experimenting with challenge-based awards that pay for results rather than proposals.
Science funding has not adapted.
The reason is partly institutional inertia — the people who succeeded under the 1945 system now run the committees that perpetuate it. And partly it is a lack of alternatives. Nobody has built a credible, scalable mechanism for evaluating research contribution after the fact and routing funding accordingly.
Until now.
Science funding innovation means building evaluation systems that match the scale and speed of modern science. Post-publication merit. Continuous contribution tracking. Funding that flows based on demonstrated impact rather than predicted potential. These are not radical ideas — they are obvious adaptations that the technology to implement simply did not exist in 1945.
It exists now.
The question is not whether the 1945 model is broken. Anyone who has spent time in academic research knows it is. The question is whether we have the courage to build something that actually fits the science of 2026 — or whether we will keep patching an operating system designed for slide rules and printed journals.
What would you change if you were handed a blank slate and told to design science funding from scratch today?
Comments
0 approved