[TomGullen]

This is probably a preference thing (and scale is likely an important factor) and can never be definitively answered, but I find that denormalising data such as SSO ID's into a third table comes at the cost of slightly more complex code and more boilerplate if you've got a caching model, I used to denormalise to this level but have now gone back to putting it all into a fairly wide user table as a short string field is also very cheap.

[bfeynman]

Curious, do you mostly denormalize only immutable data? Otherwise how do you handle EC well?