About Phase III Grading

As many of you realize, phase III was more like a term paper than a problem set, since the set of possible solutions was quite broad (and open ended). This was reflected in the grading scheme we used for this phase.

For each of phases I and II, we started with a maximum score and defined deductions for general classes of mistakes we anticipated.

For phase III, we did something different. For phase III, we started with a minimum score and then awarded points based on how good your trapdoors were. As the problem statement noted, we defined "how good" as a combination of hiddenness and efficacy, with a strong bias toward hiddenness. We also defined deductions for places a submission might fall short (such as for inadequate documentation).

The course staff made 2 passes through the submitted solutions. In the first, they summarized the trapdoors and proposed a score. In the second (which spanned two long meetings), each trapdoor was discussed and the staff reached an agreement on a grading for that trapdoor.

For those who like statistics, it turned out that 15.4 was the highest score that could have been awarded (but nobody got this), the actual mean grade was 9 (which we interpret as a respectable showing) the actual standard deviation was 1.8, and the grading distribution looked "normal".