1) When evaluating excuses (or other "low-probability" events), it's important not merely to evaluate the likelihood of this exact pattern, but of all patterns that are of the same level of impressiveness or more impressive. For example, if you're evaluating whether to accept "my cousin died", you really have to evaluate the whole class of excuses "my relative who is at least as close as a cousin had a significant occurence". This is one of the best observations in this essay.

2) Taking a game-theoretic approach to excuses, rather than merely accepting "true" or "believeable" excuses, has the advantage of allowing you to reject excuses without calling into question anyone's character. Nobody wants to be the jerk who said "I don't believe your grandmother really died" to the kid whose grandma actually DID just die; it's much easier to be the guy with the "late work is not accepted even if your grandma dies" policy.

3) The excuse evaluator can add a cost to making excuses. A teacher might require an essay explaining the excuse in detail, thereby wiping out the added utility the student derives from procrastinating, but not wiping out the utility the student derives from attending his grandmother's funeral. This creates a threshold that prevents people with no real excuse from gaming the system.