I am a newbie to randomization or stratified sampling, etc, so pardon my lack of accuracy in explaining my situation.

We have a set of covariates (binary and continuous) in our dataset. We do have not yet measured our outcome variable. We want to assign each observation to one of two treatment arms optimally enough so that for each covariate the distributions for both subsets (arms) look similar (t-test?). I was thinking that this entails fishing from a random assignment of 1 and 0 to observations so that the t-tests are clean enough (pvalues > say, 0.3) for each covariate -- and, if not, then running 1000s of iterations till we get such p-values across all covariates. Perhaps there is way to make this "search" more efficient using the current distribution of each covariate.

Is this all sounding like this could fall under an existing technique or a stata command? Or some references. Please note that we have not yet measured the outcome variable. psmatch2 sounds like it needs an outcome variable.