The cognitive load is not the issue here. The main issue is findability and learnability and amount of repetition. If the interaction is not familiar to user, it has to be learned and memorized. And if the target group is not mobile enthusiasts and designers / developers it will need to be learned.
If there are good enough hints to let users find the interaction, then you need good enough hints to show them what will happen when lift finger from screen, and let them also cancel action.
If the action is something to be repeated several times, like sorting stack of cards to two piles like voting (hot) (not) or cleaning up todo lists (done) (not done), I would go with swipe. If the actions are not pairs, like (next) and (like) I would make like separate button or selecting the item instead of swiping.
Swipe directions- for left handed person swiping from right to left is slightly more convenient. I'd make default action and positive action to be from right to left, and not default or destructive actions from left to right.
Also one option is to use up/down instead --> better mapping with good/bad.