Why one failed experiment never actually kills a scientific theory

Lakatos argued scientific theories travel in 'research programmes' — a protected hard core surrounded by adjustable auxiliary hypotheses that absorb anomalies. A programme isn't falsified by one bad result; it dies only when it stops predicting new facts while a rival keeps predicting more. This is why Newtonian mechanics survived centuries of patches before giving way to relativity.

— Imre Lakatos, Falsification and the Methodology of Scientific Research Programmes — 1970 essay, in Criticism and the Growth of Knowledge (1970); expanded in The Methodology of Scientific Research Programmes, 1978
Go deeper: get the book →