and also because a result always has the caveats of "if we did our experimental design and math right". A 1/1000 rate of experimental/code design errors will do

and also because a result always has the caveats of "if we did our experimental design and math right". A 1/1000 rate of experimental/code design errors will double the number of incorrect 3 sigma results.