Various versions of the wild bootstrap are studied as applied to regression models with heteroskedastic errors. It is shown that some versions can be qualified as 'tamed', in the sense that the statistic bootstrapped is asymptotically independent of the distribution of the wild bootstrap DGP....