Preprint models grant allocation as decision problem under Knightian uncertainty

The authors treat funding panels as agents facing ambiguity about project quality distributions and future citation impacts. They simulate panel decisions using robust optimization and ambiguity-averse utility functions drawn from decision theory. Sample size consists of 12,000 synthetic grant applications calibrated against real NSF and ERC award data from 2015-2022. The model shows that ambiguity aversion leads panels to favor incremental projects over high-variance ones even when expected value is identical.

This result aligns with observed patterns in which conservative funding correlates with slower paradigm shifts. It extends prior work on peer-review bias by quantifying how uncertainty aversion reduces portfolio diversity. The analysis connects to recent replication-crisis literature showing that low-risk grants produce more publishable but less robust findings.

The preprint does not contain empirical tests on actual funded versus rejected proposals. A natural next step would be a field experiment randomizing panels to standard versus ambiguity-adjusted scoring rubrics and tracking long-term citation and patent outcomes over five years.