The difference between 97.1% uptime and 97.2% uptime is far less important than the difference between 99.98% uptime and 99.99% uptime, even though this graph shows the former as 10 times larger.

You are right to complain that a graph that exaggerates uptime differences on the scale of 10%, i.e. showing the full linear range from 0% to 100%, would be useless. But by the same token, a graph that exaggerates uptime differences on the scale of 1%, i.e. the OOP, is also useless. That we happen to live in a world where github’s uptime is varying at the scale of 1% doesn’t make the scale any more useful.

In this case, the graph of -log(1-uptime) would get you the “number of nines”, which is commonly used because it’s more insightful and more indicative of actual quality. Better still would be log(uptime(t)/(1-uptime(t)), which is functionally the same above two “nines”, but also allows the plotting of low uptime services, such as individual seeders for torrents or specific nodes of a mesh, on the negative part of the scale.