As a teacher, who has not faced this dilemma: you want your students to make an effort and learn, but you also want them rewarded with good grades so that they do well on the job market or with college applications. If your grading practices cannot be observed by outsiders, there is no reason for you to give bad grades, and students will not study hard.
Robertas Zubrickas rationalizes all this with a principal-agent model where the teacher cares more about the learning and the students more about the signaling. Teachers offer a contract to students, oberserving perfectly their performance and rewarding it with good grades. The problem is that issuing good grades is costless. Of course, what really matters are relative grades, but if the market cannot observe the grading rule or the grade distribution, all that matters are the "nominal" grades. The result: everyone gets the best grade, and there is complete pooling. And it is empirically observed that schools with laxer grading standards achieve lower SAT scores.
And how can one prevent this poor outcome. Either make the distribution of grades available, thus making grades "real" instead of "nominal." Or make it costly for teachers to give good grades (instead of making it costly to give bad grades by asking for extra administrative steps for those). For example by imposing a distribution or an average grade. I would certainly welcome this, as some of my colleague are clearly free-riding and giving good grades to everyone.