I was getting the same kind of results using the search/streams endpoint. I did some testing and I believe what is happening is that when streams are ending, what would’ve been that 9th object is removed but the rest of the objects in the array in the server are still in place until the server reorders them. The same goes for the _total property in the response. One possible solution is to increase the limit, store the objects, and use your own pagination functionality instead of making a request for each new page. But this comes at the cost of accuracy of what streams are actually still live.