You end the do loop as soon as max greater than sum , which happens in the very first iteration.
After you switch this > to < , you get an arithmetic error, because ultimately you will bind sum to nil (when upgrading to (if (evenp curr) (+ sum curr)) , which is nil when (evenp curr) is false.You also need to specify an else-side: it must be sum .
More importantly, do binds its values in parallel, rather than in sequence, which means that for the second iteration, when you upgrade sum to (if (evenp curr) (+ sum curr) sum) you use curr and sum from the first iteration. If this is not what you intend, you should use do* .
Update
As requested in the comments , here is the full working version of the code. Note that it is almost identical to the code in the question; it only reduces the order of the arguments to > , so that completion occurs when the curr ent of Fibonacci is greater than the value of imum max , and adds the else case to the if , so that when (evenp curr) false, the sum kept.
(defun sum-even-fibs (max) (do ((curr 0 next) (next 1 (+ curr next)) (sum 0 (if (evenp curr) (+ sum curr) sum))) ((> curr max) sum)))