Compared to the mathematical rigor that gets usually thrown at such problems, this is all pretty vague, but maybe this is just the informal version. I admittedly did not completely understand the idea, but I of course have thoughts about it.

Instead of “equality at all times” we should be asking “alignment whenever someone actually looks”, because this is all we can promise about someone else’s state.

It seems to me that we generally do not know when and where someone will look, so we will still have to be consistent at all times and all places because all times and all places is when and where someone could look. If the idea is to delay consensus from the write to reads, then I do not see how this makes things any easier. Instead of ensuring that everyone receives a write when it happens, you now have to ensure that you can gather all the writes that happened when a read occurs.