Many alternatives exist for labeling arrays in scientific Python community. A really nice one looks like
One particular nicety is that broadcasting (arithmetic) is done on matched-label dimensions, not matched-shape dimensions. This would prevent a number of errors.
We should look at some of these alternatives to see what benefit they might provide in simulator and framework backend and how their metadata would fit with TVB's approach (of sticking it along side as related attributes on owner object or in trait metadata).