planned
Forecasting drills that train grantmaking judgment.
Grantmaking quality is gated on calibrated judgment about counterfactual impact, and that judgment is currently tacit — it lives in maybe a few hundred humans globally. Phil's bet: it's trainable. Give grantmakers a steady stream of well-resolved questions about real allocations, watch how predictions track reality, and surface a number for grantmaking competence the way Tetlock did for forecasting.
Side effect once it works: a clean signal for training a grantmaking-LLM, and donors who can pick grantmakers (or models) by track record rather than reputation.