= k * da * [a - b *(1-d_g) ]
Notice that nowhere in that calculation of the resulting pitching moment does the actual absolute AoA on either surface play a role - what matters is the gradient with alpha of the aerodynamic characteristics.
Nah, you're just hiding the dependence. The "gradient with alpha of the aerodynamic characteristics" (by which you mean k, I guess) is always positive in the unstalled regime. If that's
all that matters, we wouldn't care about all this aircraft stability stuff, because they'd always be stable.
But the problem is the rest of it, the [a - b *(1-d_g) ]. It's that bit that swaps sign and gets nasty when you move the CG. So how do we know what the sign is? The easiest way is to use the criterion that the total pitching moment is zero. i.e. that
a*CL - b*CLt = 0
And once you substitute that back in for a and b, you get:
k * da * a * [1 - CL/CLt *(1-d_g) ]
so the ratio of lift coefficients rears its ugly head, and suddenly it matters again whether CL is greater than or less than CLt, which, with the same k for wing and tail, is the proportionality argument again. Or more precisely it matters whether CL*(1-d_g) is greater than or less than CLt, which just means that we have to modify the proportionality argument to take account of the downwash gradient.