OK, suddenly I can’t recall what resampling exactly means in this context.
It’s a Monte Carlo simulation based on historical returns of asset classes so I am assuming, instead of having one single expectation for the asset class return as done in traditional MVO, we use a distribution of each asset class return, with probabilities assigned using historical norms; do the same for risk (std dev), and somehow get a fuzzy efficient frontier. Does that sound about right?
For the same “point” on the EF we could have two different portfolios with two different probabilities because in one of them, say, we assume stocks return 10% and bonds 8% with p = 0.5 and in the other, stocks return 5% and bonds 20% with p = 0.01, or something like that; but both have the same overall return and std dev.