we want each tuner to freely decide the resampling experiment itself. but the function has become rather complex including errorhandling.

i think it should have something like a subset of hashes that it selects the best from