What would a good written policy for scaling exam grades look like?

In general scaling is a mistake as it introduces uncertainty into the student's calculations. But some adjustments are proper and you can publish statements about them.

First, you can agree that no one will "miss" the next grade by a small percentage. This is better done overall, than on a per exam/paper basis. But if it takes 90% (overall) to get an "A", then, at the end of the day give the A to someone with 89. This merely recognizes that your grading scheme isn't perfect and may disadvantage people in small ways occasionally.

Second, agree that your grading rubric is clear that you are allowed to adjust upward when you think it justified. Mine would say something like, "If you get 80% you will get at least a B". This is just a reinforcement of the first point.

Third, permit people to do work over if they have fallen short. Regrade it for "most of" the lost points. My policy was 90%. If you lost 30 points on an assignment you could get 27 points (max) back if you re did the assignment properly. My policy was generous, permitting several attempts. The repetition was good for the students and worked to assure learning.

Fourth, at the end of a course, look at how the students did overall and compare it mentally with what you think they really learned in the course. If you think the learning was actually better than the distribution shows, bump it a bit. This will push a few students up to the next partial grade, say from B to B+.

Finally, avoid marking using only a few high-risk exams, but spread the marks over a large number of tasks. All or nothing final exams eventually leads to trouble, even in less extreme cases than the one you mentioned. This leads to a practice of continuous study and learning rather than "cramming" for the big one. Cramming results in more memorization (short term learning) and less deep learning.

My students all knew everything about the above policies. I seldom got complaints about grading. I could be as demanding a professor as I felt necessary (students viewed me as very demanding). At the end of the course surprises were always happy ones and students felt good about themselves and encouraged to continue.

Finally, let me note that strict scaling, making the course distribution into something like a normal curve is, IMO, always a mistake. It turns the course into a zero-sum game for students who can only win if someone else loses. In theory it should be possible for everyone to do well, even full marks, based only on their own work. Such scaling is also unjustified as it assume that a given sample (your students) perfectly match a population (all students). Statistically that is a serious error to make unless your scale is huge.

Scaling or "curving" grades so that a fixed proportion of students get each grade is unethical. There is extensive evidence that student cooperation helps students learn more. If you fix the proportion of students that get each grade, then you incentivize students to stop cooperating. This will reduce how much your students learn.

@PatriciaShanahan is right that this grading approach is also incorrect because your student body does not remain the same across semesters.