When heatmaps are wrong.

gistfile1.txt

Our heatmaps are generated via sets of events. The notable ones are "start" and
"stop" events. 
Let's say we have a 60-second video with some events attached to it.

A video watched all the way through exactly once looks like this:

a = start
z = stop
|------------------------------------------------------------|
a                                                            z

A start event exists at the beginning and a stop event exists at the end. This
represents a viewer watching the video without pause in its entirety.

Now let's take a scenario where the user watches the entire video except for
three seconds of it:

|------------------------------------------------------------|
a1                   z1 a2                                   z2

We see that at z1 the viewer stopped the video, and then skipped ahead to a2,
where he began watching it again. With this complete data, we can build an
accurate heatmap.

However, what happens if z1 and a2 are missing when we try to generate this
heatmap? In fact we will wind up with an event set equivalent to our first
example:

|------------------------------------------------------------|
a1                                                           z2

and thus the heatmap that shows at this time will be inaccurate, and will remain
so until events z1 and a2 are processed.