It is well known that :
$$\ \ \ \ \ \ \binom{(A,C;B,D)}{\text{harmonic division}}\ \iff \ \binom{\text{cross-ratio}}{ [A,C;B,D]=-1}.$$
(definition of cross-ratio here).
A non-classical formula in the case where $A,B,C,D$ are aligned in this order, expresses the general (negative) cross-ratio under the form :
$$[A,C;B,D]=-\frac{1}{\tan^2(\tfrac{\theta}{2})}\tag{1}$$
as a function of the angle $\theta$ between circles with resp. diameters $[AC] $ and $[BD]$.
Fig. 1 : Case $a=0,b=2,c=3,d=7$, one gets $[A,C;B,D]=-1/\tan^2(93.82°/2)$.
I provide two proofs of formula (1) below.
I hadn't seen previously this formula until I found it, under a different form (up to Cayley transform $Z=\frac{z-i}{z+i}$) in a recent question to which I haven't answered directly because there was no evidence of work. Unlike this previous question, I do not make reference to hyperbolic geometry.
My questions are : where can be found references to this formula in the literature on projective geometry ? To which other properties can it be connected ?
Proofs of formula (1) :
Proof 1 : (found afterwards, but placed first because it is simpler)
This issue is (classicaly) equivalent to this one (see figure 2), where the coordinates of $A,B,D$ are $0,1,\infty$ resp. More precisely, we keep the left circle as it was and we transform the right circle still making an angle $\theta$ with the first one into a vertical line in $C$, ("circle with an infinite radius").
Fig. 2.
In this way $\angle CEI = \theta$ (mutually orthogonal sides) ; therefore angle $\angle CAI = \theta/2$ (inscribed angle theorem).
As a consequence
$$AH=\tfrac12 \cos \tfrac{\theta}{2} \implies AI=\cos \tfrac{\theta}{2}$$
Then $$AC=\left(\cos \tfrac{\theta}{2}\right)^2$$
The cross ratio is
$$\frac{\overline{CA}}{\overline{CB}}\frac{\overline{DA}}{\overline{DB}} = \frac{-\left(\cos \tfrac{\theta}{2}\right)^2}{1-\left(\cos \tfrac{\theta}{2}\right)^2} \times \frac11$$ $$=-\frac{1}{\left(\tan \tfrac{\theta}{2}\right)^2} $$
Proof 2 : Let lower case letters $a,b,c,d$ be used for the abscissas of $A,B,C,D$ resp.
The angle between two intersecting circles is given by the formula :
$$\cos \theta = \frac{d^2-(r_1^2+r_2^2)}{2 r_1 r_2}\tag{2}$$
where $r_1,r_2$ are the circles' radii and $d$ the distance between the two centers by an immediate application of the law of cosines to triangle $C_1IC_2$ where $I$ is the intersection point of the two half-circles.
With the notations of the figure, (2) can be given the form :
$$\cos \theta = \frac{\left(\tfrac12(b+d)-\tfrac12(a+c)\right)^2-\left(\tfrac12(a-c)\right)^2-\left(\tfrac12(b-d)\right)^2}{2\tfrac12(a-c)\tfrac12(b-d)}$$
Different simplifications yield :
$$\cos \theta = \frac{(a-b)(c-d)+(b-c)(c-a)}{(a-c)(b-d)}$$
$$\cos \theta = \underbrace{[A,D;B,C]}_{\frac{r}{r-1}}-\underbrace{[B,A;C,D]}_{\frac{1}{1-r}}$$
(where $r$ denotes cross-ratio $[A,C;B,D]$). Therefore, we have :
$$\cos \theta = \frac{r}{r-1}-\frac{1}{1-r}= \frac{r+1}{r-1}= \frac{1+\tfrac{1}{r}}{1-\tfrac{1}{r}}.$$
From there, one can deduce formula (1) as a consequence of the following classical formula (see here) :
$$\cos \theta = \frac{1-t^2}{1+t^2} \ \text{where} \ t=\tan(\tfrac{\theta}{2}).$$



This answer is mainly to provide a reference, and also to provide a bit of background and context.
The reference is Coxeter, Inversive Distance.
Coxeter defines inversive distance between two circles and relates it to cross ratio. In section 4 he defines the quantity
$$ \gamma = \dfrac{a^2+b^2-c^2}{2 a b}, $$ where his $a,b,c$ correspond to your $r_1,r_2,d$.
He is concerned with non-intersecting circles, and defines their inversive distance $\delta$ as satisfying $\cosh(\delta)=\gamma.$ This is analogous to $\cos(\theta)=\gamma$ for the case of intersecting circles. (In general, $|\gamma|=1$ for tangential circles, $|\gamma|<1$ for intersecting circles, and $|\gamma|>1$ for disjoint circles.)
(According to one paper I read, $\delta$ is the imaginary part of the circles' imaginary angle of intersection. $\cosh()$ is of course the hyperbolic cosine, and Coxeter and Greizer provide a cheat sheet of hyperbolic functions in another write up of inversive distance and cross ratios in Geometry Revisited, pg 126.)
Back to the reference paper, the equalities in the middle of pg. 76 show that the cross ratio for disjoint circles is equal to $\tanh^2(\tfrac{\delta}{2}),$ which looks analogous to our $\tan^{-2}(\tfrac{\theta}{2})$ for intersecting circles.
I haven't done the work, but I conjecture that you can generalize the formula so that it works when the circles don't intersect. And I think that, had Coxeter not focussed solely on the case of disjoint circles, he would have derived the $\tan^{-2}(\tfrac{\theta}{2})$ formula for intersecting circles.