For the first week of November 2011, I went through the process of migrating 400,000,000 sales rank records into a new database system, in an effort to improve the performance of NovelRank, especially when you requested large amounts of data (e.g. charting). As of today, that migration was completed and NovelRank was updated (I didn’t eat today… it’s been a long day) to use this new system in 95% of requests. I didn’t think there would be as many problems as there were, thus why many of you reported strange changes hour-to-hour in your data. All of that should be resolved and the system mostly stable.
One big change is that as of Nov 1, sales estimates are being processed through a much more accurate (but not perfect) statistically based system for estimating the number of sales that occur per hour. The system still has a weakness for inaccuracy with a sales rank < 500 due to being based on sales rank changes, but it is a lot more accurate for Kindle Edition books, especially those not on Amazon.com. Due to some feedback today however, it seems that for some titles with sales rank < 500, it was still likely to be underestimating (and could be improved). I’ve made a tweak going forward that should increase sales estimates for those popular titles. I’ll continue to monitor the results throughout the month. This new system is constantly learning and reevaluating what constitutes a sale for a particular book/ebook, and it’s accuracy is increasing daily.
Please be aware that your sales estimates MAY HAVE CHANGED TODAY for the month of November as NovelRank was displaying the OLD sales estimates prior to today, while recording both the old and the NEW estimates (which are now being displayed).
There are still some issues with speed when calculating your BEST, WORST, and AVERAGE sales rank causing massive slowdown using the new system. For now, these calculations are still based on the old database structure, and I’ll be working this week to try and solve this problem. The end goal is to keep the speed and performance of NovelRank high and remove the downtime that occurs for a few hours every 2-3 months as well as for 30 minutes every evening at 1pm Pacific Time.
Finally, for those using the iPhone App, 40% of the data provided is from the old system (to keep the performance quick). Later this week you may see the stats change to ‘n/a’ for STD and AVG information as that data will be removed. I’ll be working with the App creator to see if an update is possible after the changes have been made.
Thanks again for using NovelRank, and I hope that it continues to be a useful tool in every authors arsenal.
Update: Nov 9, 2011
Some new fixes and adjustments have been put in place based on feedback. The system is now much tighter with the sales estimation algorithm, as well as it attempts to calculate sales for ‘holding rank’ when books have a sales rank that is better than 1000. As usual, estimates for popular titles will still be underestimated due to limitations of sales rank tracking.