I don't think the author presented any evidence that invalidates the assumption of IID colors.
They show a histogram of the different counts of each color, which seems more or less consistent with a uniform distribution.
There are some fluctuations but they're not surprising; the Poisson error bars would be roughly +/- 0.16 in that figure.
With error bars that large, it would be surprising if the data were in more exact agreement with the flat line; it's actually related to the same question that the author is examining ("what are the odds to observe all 5 colors at exactly their expected rate of 1/5 within measurement error?").
They do speculate that the number of candies per pack is not IID, i.e., that there are (anti)correlations from one pack to the next.
But without knowing more about the packing process, and presumably also having some lot/serial number information for each pack, it would be pretty hard to establish this.
They do speculate that the number of candies per pack is not IID, i.e., that there are (anti)correlations from one pack to the next. But without knowing more about the packing process, and presumably also having some lot/serial number information for each pack, it would be pretty hard to establish this.