Who Gets the Call? Predicting Nobel Prizes from Prize Histories and Publication Records

Every October the Nobel Prizes are announced, and almost nobody predicts them. This paper asks whether they can be predicted at all, and what the answer says about how science rewards its own. Earlier attempts treat the problem as classification: take a list of scientists, learn from the records of past winners, and score everyone. Two things go wrong. The list is chosen by the researcher and largely determines the answer, and the method ignores that a prize is a choice: a committee picks two or three names from a few hundred plausible ones, so candidates should be compared with one another rather than with a fixed pass mark. We build the list from the awards themselves. Ninety-three prizes recur in the award histories of the 765 Nobel laureates in physics, chemistry, medicine and economics, and the 9,945 people who hold at least one of them form what we call the prize ladder. Following 4,958 of them year by year from 1940 to 2026, a model trained only on the past places the average laureate in the top fifth of the field, and its ten-name shortlist contains about one laureate in nine, six times more than chance. Earlier prizes matter far more than citation counts, and having a laureate as a doctoral advisor matters more than working at a famous university. Yet even the best model needs several hundred names to be ninety per cent certain of including the winner, which measures how much freedom the committees still have.