I don’t think it should have been called “CAP” theorem. I think of it as “CAL” (consistency–availability-latency). It has always struck me that the P in CAP is different than the other two characteristics. C and A are continuous values, but P is binary (with a duration presumably).

The “P” in CAP means “tolerance of partition events”. But that’s really not what most engineering teams are thinking about when they talk about CAP theorem. Teams that really want tolerance to network partitions are really talking about building parallel instances, for example, cloud providers implementing multiple regions. If they need some level of consistency across those instances, then really, it becomes a discussion about latency.

If you think about it, what you’re really talking about with the network partition events is how stale or latent the data on each side will become, and whether or not, it will become consistent again, particularly without human intervention.

When you talk about it in terms of CAL, now all the properties are continuous, and you can discuss the trade-offs to each property with respect to network partitions.