one tailed t-statistic

Sorry, I know this is a really simple question, but why would I not use the t-value from the two-tailed column for df 40 (2.021), when the answer clearly suggests using a one-tailed stat (1.684)? Is it a genearl rule that you use the one-talied values for the t-distribution? If so, when would you use the two-tailed values? To me, the question seems to suggest that the deviation can above or below the average order size so two-tailed would be used…

[question removed by admin]