The original data used over 1000 distinct classifications which I grouped into 36 wider classifications as best as my understanding of the classification names allowed.
The data provides a text description of each object's date as well as "start" and "end" dates which bookend the possible timeframe taking into account any inaccuracy of the date. So for an object described "early 7th-mid 7th century BC" the start and end dates are -675 and -650.
For an object with a range of possible creation dates this visualisation distributes the object across the full range. So an article made between 1840 and 1850 is distributed as 1/11 of an object for each year in the range 1840-1850 inclusive. This is why some years are shown as having only fractions of an item.