I have followed the link to read the specification itself.
SEE basically is an extension of RSS to carry additional change information about items of mutual interests between parties. When the publisher and the subscriber mutually publish and subscribe to feeds referring to the same item, changes are notified. End points processes can use the notification to handle changes as required.
The atomic size of change is based on item. Hence, if a post is an item (which is usually the case), then changes will be propagated even when it is a change in a single character in the item. In larger work, we can aggregate a set of items together. Changes are still item-based. There are provisions to enable changing the order of items in a set.
This model also requires that all participants of the SSE will have local copies of the items. Update is made by comparing the local copy with the published copy and a winner is determined. (see 3 and 4 of the specification)