As far as I remember, we did not learn an algorithm for making a lossless 3NF decomposition that is dependency preserving.
(In the presentation, it says it's "not so easy" and refer to the book, chapter 3.5; the book says it's beyond its scope).
However, in last year's exams there were questions that asked exactly that.
What should be the approach for such questions? Is it just trial and error?