Is it true that VaR "assesses potential minimum loss over a given time with a given degree of confidence" I thought it gave the maximum loss at a confidence level, or equivalently the minimum return, since it gives the lower (say) 5th percentile of the distribution?
no it's max loss. i.e. at 95% confidence level, the one day VaR is the maximum loss you expect to occur 95% of the time over a single day. Where did you see that min loss quoted?