Yes, the experimental group that was induced to bribe for a license without testing were offered "free lessons" after the study. But this hardly "ensures that there was no social cost." There is no mention of lesson take up amongst the bonus group but statements elsewhere in the paper suggest this was not high. 40% of the experimental group that was offered free lessons at the outset did not take up those lessons, and we would expect an even higher proportion for the bonus group. (At one point the paper says that the "lesson group" were the better drivers, implying that the bonus group had even lower take up of the free lessons.) So we are left to conclude that the experiment increased the number of unsafe drivers, as I claim.