Thanks @robbie.morrison for the helpful response (as usual)!
In my case, I am trying to improve the standard LCA way of modelling electricity grid mixes by including the uncertainty in generation shares. Currently, this is simply skipped and annual averages are used, because the LCA software is incapable of including uncertainty and then normalizing back to 1 kWh of production for the entire mix. We could use calculated covariance factors, but this is lossy and unnecessary when we have the raw historical market mixes.
You are completely correct that to get the right answer, one should model the entire system. It would be very nice to get some quantification on the error induced by using average factors e.g. for the charging electricity for storage systems, for the relationship between load and CO2 emissions, etc. Barring that, I guess using the average mix when the storage system is not discharging (maybe weighted by price, where lower prices have higher weights) could be a reasonable proxy for the charging mix. It would be lovely to get actual data from specific storage systems, but I have no idea where to find this for now.
If anyone is interested in helping build such a system, please come on over to https://github.com/BONSAMURAIS/bentso 