How to complete analysis with small N?

ANCoVA and t-tests make assumptions about distributions but become resistant against violations of those assumptions with growing $n$. So with small $n$ it is reasonable not rely on these parametric tests but use non-parametric tests instead.

So why not use a simple ranksum test (aka Wilcoxon or Man-Whitney-U) or a permutation test?

Well the first question is whether the method is appropriate at all to the problem at hand ... Lets say that all the methods under consideration are arguably appropriate, you may then instead want to know which of the methods has most power (rejecting null when its false). You can get a sense of this using simulation. Steps:

1. Use your assumptions about the distribution of scores to generate random data that match the sample sizes you have, based on your null hypothesis being true.

2. Run all the models you're testing against the data. Log the p-values.

3. Repeat N>1000 times.

Power = (no of times null is rejected) / N

Pick the method with the most power.