Even for pure STEM subjects, the further you go into esoterica, the less checking there can be on it. I remember one instance, mainly because it added an hour or two to a problem set in grad school.

(Apologies for any errors, as it's been the better part of a decade since then.) Wikipedia gave simplified formulas for the asymptomatic forms of the modified Bessel functions, and each formulas they gave were correct. However, each simplification used a different branch cut, so subtracting two of the asymptomatic forms wouldn't give the correct result. Tracing back that error, and why I was consistently off by a factor of 2*pi took a long time.