Michael Kuron

We would need to essentially replace float with a custom Kahan-summed datatype so we can keep the correction term for the entirety of the simulation. If we only did Kahan...

So your tests two weeks ago used `double` only in some places which was sufficient to make the test case pass but only cost a factor of two in run...

This is a known problem (https://github.com/mcuadros/ofelia/issues/157, #4272] that ofelia never fixed. It was suggested a while ago to replace ofelia with its fork chadburn (#4362), but that fork isn‘t receiving...