Artifacts in data analysis using floating-point numbers.

Using the Python script and data provided in Supporting Information S1 Data, we loaded the input data x and y in the top two rows (where one unavailable data point is interpreted as nan, floating-point "not a number"). Computing the ratio x/y, division by zero resulted in two further nonfinite values. Sorting the data (in ascending order, using Python's standard sorting algorithm) silently yielded an incorrect result. Computing the maximum based on successive comparisons failed as well, but this is not immediately obvious just from the value. Only the mean value is an obvious indication that something went wrong.