I've been following this thread for quite a while now, and the drop in page reads issue seems to be very odd, and perplexing. Some have seized on the Page-Flip being buggy as the possible reason but I am not sure that it isn't just a another bug in the overall system.
The key thing I have noticed is that it is not too easy to tell if your page-reads are decreasing normally or it is a systemic error caused by Amazon somehow intentional (algorithm changes) or unintentional (just incorrect lower counts due to errors). As the problem orginally seemed to be noticed by more people as occuring at some timeframe within Sept. and Amazon seeming to acknowledge via tech support that issues in that timeframe were legitimate; I decided to focus my analysis on this snapshot in time. (despite seeing some other zero day patterns in a smaller way appearing in early August too).
I publish one author's group of over 50+ titles of which 90% are in Select. Using the Sales Dashboard I looked at each of those titles KENP read for 90 days view to see if I could see any specific patterns. At first it was almost random or inconclusive, and I ended up discarding all the bottom 25% as slow sellers anyways. Of the rest I then began to see some chunks of time in September that had recurring zero days. Looking at the previous 90 days I also discarded a few more that had normally occuring dry spells of 3 or more zero days in the previous 60 days. That left about 12 titles that seemed to have a pattern of 3 or more days in a row (or more) - this data was transferred to a spreadsheet so I could see the overlaps. (image below - purple boxes are zero pages day for that title)
What I noticed was a set of 'chunks' where numerous titles all had zero pages read,
9/02-9/07,
9/08-9/12,
9/12-9/22 or 23. To me this really feels like an error of some sort due to a reporting code change (perhaps on the backend), that effected some titles groups as opposed to others. You can see the earliest chunk on my graph is only effecting one series. Then maybe even a attempted fix that then affected other titles than the earlier problems. Only a theory...but maybe others could look for similar time 'chunks' of zero page days reported too. And, see if like me, older titles had longer zero day drought periods which could point to a algo change favoring newer releases.
Would be great if we could get DataGuy to jump in with some more statistical analysis on this mystery for us...Hugh tell us if your Sept. page counts are looking wonky at all...inquiring minds want to know.