A plane (a,b,c,d) is really a four-dimensional trivector, not an ordinary vector. When you take the wedge product of three points P, Q, and R, it naturally produces a plane in which d = -dot(N, P) = -dot(N, Q) = -dot(N, R), where N = (a,b,c). When you take the dot product between a homogeneous point (x,y,z,w) and a plane (a,b,c,d), you get a*x + b*y + c*z + d*w, which gives you the signed distance between the point and the plane multiplied by w and the magnitude of N. A positive sign in front of the d is the correct choice.
This kind of stuff is discussed very thoroughly in my new book that comes out next month:
Dirk, when a homogeneous point is treated as a single-column matrix that is transformed by multiplying on the left by a 4x4 matrix M, a 4D plane must be treated as a single-row matrix that is transformed by multiplying on the right by the inverse of the matrix M. If the translation portion of the matrix M is not zero, then a plane will not be transformed correctly if you treat it the same as a point.