> That is definitely not compatible with actual mathematical “randomness”. Of course it is, the concept of randomness is much broader than you allow for. The

> That is definitely not compatible with actual mathematical “randomness”. Of course it is, the concept of randomness is much broader than you allow for. The specific issue you refer to here is if you are sampling with or without replacement - both are valid. There is another issue which is the distribution you use. Should songs be equally likely to occur, or should the ones you've played more occur more likely? You may also extend sampling strategies to genre, or to exclude repeated artists, or whatever - but all of these things could still be done via random sampling.