We teach a capstone course on software engineering where students work in teams during one semester in a real world setting. By the end of each of the three iterations, students get a grade that takes into account several aspects including peer assessment. In the past, students with low commitment used to get only a slightly lower grade because peer assessment's weight was very low. Last semester we added a new grading rule: if the peer assessment is lower than a threshold, then such a grade would be final for the affected student(s). In this work we aim to show the effectiveness of this rule for promoting teamwork quality by recording the progression of peer assessments along the semester. We found that peer assessment decreases and is more disperse in the second iteration, but significantly improves in the third iteration. Our results suggest that the new rule is effective for improving teamwork by making students more responsible for their teammates' grades sooner in the project, when there is still time for improvement.