I think I can see part of the cause for your observations. Your spreadsheet shows numbers with decimal points, which means it would have been data from Crowd Calendar 2.0. It was just about this time that the guys updated the system to Crowd Calendar 4.0, which does not have decimals.
This change automatically changed data almost every day. The numbers between the two calendars are not directly comparable. With the new calendar, there will be very few 1s, not many 2s, and a lot fewer 9s and 10s. Another aspect of the new calendar is that there are expected to be fewer significant updates as dates get closer.
Looking at the second set of numbers, the predictions were pretty darn close.
I've never heard these terms Crowd Calendar 2.0 and 4.0. Are those Touring Plans names for their prediction models? If the newer model will have fewer 1s,2s,9s, and 10s, does that mean they aren't percentile-based anymore. I always thought 10% of days would be 1s, 10% would be 2s and so on. The rolling nature of a calendar and changing predictions would cause that to not be perfectly true, but I thought theoretically each crowd level should be seen on about 10% of a year's days.



