Re: Bucketing: Look for commonalities amongst timeouts

[Chris Warburton]

I initially plotted a bar chart, with each name getting a successes and
failures bar side-by-side. Since the names have no inherent ordering, I
sorted them by number of successes.

The result was a steadily-decreasing number of successes (as we would
expect from this sorting), each with a random-looking number of failures
(as we would expect if number of successes and failures is uncorrelated)
*except* for the very end of the graph, where there was a clear spike of
names which had many failures and no successes.

That graph didn't look like the best representation to me, so instead
I've turned the successes and failures into percentages and plotted
them on a stacked bar chart. This shows the gentle decrease we would
expect, followed by a sharp drop in successes and a few completely
unsuccessful names at the end.