In order to interpret the data you need to consider that at times uploading was not available. The recent upgrade for example will produce an artificial trough.
The data is not a count of uploads; it is a count of premiere date (as listed on a video page) processed into a percentage of videos for that year.
I guess it's possible that not having uploads affects the number of people who will create a video entry and fill out a premiere date, but I'm not sure how that explains 2006.
I'm going to be looking over my dataset for anomalies in data collection; again, if anyone else wants a copy, just ask. The dataset is a CouchDB database (v0.9.0 format), 138.5 MB in size. Views for performing counts, as well as the hacks I used to generate the charts, are included.
This is a block of text that can be added to posts you make. There is a 512 character limit.