It would be great if, for each accepted sample we can store extra information, e.g., for GW stuff single detector SNRs. In theory, this must be possible since every sampler must know when a point is accepted, we'd probably have to dig further into the guts of the samplers to add it in though.

I would imagine this would be done by giving the likelihood a dictionary of extra_information (or some better name) and this is just added as rows to a data frame every time a sample is accepted. There may be some subtlety matching this up with the posteriors with nested samplers.