AbstractEvaluates and replicates an improved specification of the transactions’ demand for money model, and shows that some empirical econometric equations which appear to ‘break down’ out of sample, had already ‘broken down’ but had not been rigorously tested for constancy. New observations can highlight pre‐existing failure, so regime changes help evaluation. Congruence defines the null of model validity (matching the evidence in all salient respects), but the nature of destructive testing reveals that there are no sufficient conditions for model validity in an empirical social science. However, a stringent list of necessary conditions exists. Personal computers with powerful, friendly software (PcGive) facilitated a leap in productivity: model evaluation could be undertaken at home in one evening. The M1‐demand equation did not exhibit predictive failure on data, which included the new 1979 policy on monetary control.