In 1994 a social experiment was conducted in Denmark, where unemployed applicants for classroom training were randomised into treatment and control groups. The data are contaminated by the presence of no-shows and crossovers, biasing the traditional experimental estimator. We interpret our experiment within an economic model of agents maximising outcomes facing different cost regimes and present results interpretable within this model. Surprisingly, we find that classroom training significantly increases individual unemployment rates and decreases employability. We discuss possible reasons for this finding and some related policy issues.