@jrich Unfortunately, afaik that string is eval'd in the global scope and will therefore not have access to s.

@zeppelin Yes, this last one is very nice. Thank you very much. :)

@ETHproductions Oh, good point. That is nicer. Thank you.

@zeppelin The problem is at the other end of the spectrum: (1+0.099999*10)*999 > 1997. As you have shown yourself, this will mean n===1 for the rest of the computation, but it will wait almost twice as long, as it should :)

@zeppelin I don't think I can remove |0. It's there to round down the random number. If I don't do then the timeout might be longer than the spec allows.

@zeppelin Fair point, according to the standard definition it doesn't count. (meta.codegolf.stackexchange.com/a/1325/56071). I shall change it accordingly.

I posted an attempt at a standard definition over on meta (meta.codegolf.stackexchange.com/a/10663/56071) to avoid ambiguities in the future. It is consistent with the answers I can read that have been posted so far, FWIW.

The issue with not defining what distribution the samples should approximately follow is that it we need to make assumptions on what is allowed and it therefore it becomes ambigous if a solution is valid. We could for example replace String.fromCharCode(...[...Array(n)].map(_=>Math.random()*95‌​+32)) with [...Array(n)].map(_=>'TX'[+(Math.random()>.5)]).join('') and lazily outgolf the current ES6 submission by 9 bytes. According to the spec it should count because it is a random character. However, it seems unfair, doesn't it? So should that count? See where this is going?