In a particular field there are two algorithms being used:

• Algorithm Simple: you understand it in five minutes, you can implement it in fifteen.
• Algorithm State of the Art (SOA): requires very specific mathematical knowledge to understand and is hard to code. Beats Simple by a big margin, both in terms of speed and results.

However Simple is more popular because of its simplicity and not-so-bad results. Simple has 130 citations, SOA has 80. In 2016 Simple had 20 citations, SOA 17.

While trying to use the Simple paper, our results were so bad we spent more than a week looking for a bug in our implementation. Finally, we discovered a one line change that makes it work much better, improving a mathematical approximation in the Simple paper that is easily overlooked. After reviewing recent literature, it seems this is not known (we can reproduce recent papers).

We can prove that the one line change is statistically better and equally fast as Simple but statistically worse and slower than SOA. We also have a 5-line-change improvement which makes it both better and faster than Simple, worse results than SOA but slightly (10–20%) faster than SOA.

What should we do? Since we are not beating SOA's results and it is a one line change it may be hard to publish, but it is clearly something people in the field would profit from.