We have ended up calculating averages for ourselves but its dependent on what gets included (some delete all the low recordings as noise) so having something that kicks out anything outlier wise and then averages the top 10 would be a pretty pure indicator of session robustness.

Agree, but to be a bit nerdy I think something like a median of top 10 instead of an average would be better since otherwise false very high jumps would still affect the average while it will not affect the median.