Stitching interferometry could be used to compare the centre of the convex reference with the entire surface of a concave testplate but that is both mechanically and computationally complex.
When testing an F/1 sphere by PDI the diffraction limited point test source should be close to its image formed by the test surface. This is difficult to achieve without introducing additional aberrations due to the use of beamsplitters etc.
However if the PDI operates at F/10 or slower together with a finite conjugate microscope objective with adequate working distance is used to transform the F/10 beam to F/1 or faster then on axis PDI is practical as the random ball test can be used to calibrate the errors of the microscope objective and interferometer including any additional optics such as beamsplitters etc.
Even with an aplanatic back distortion is incurred when imaging a spherical surface onto the flat surface of the image sensor unless the camera lens is designed to correct such distortion. Done correctly the linear displacement of the image on the image sensor of a point on the test surface from the image of the center of the test surface will be proportional to the sine of the field angle of the chief ray. Otherwise correction needs to be made for the effect of such mapping distortions.
A ball bearing can also be used as the ball in the RBT ( random ball test) but they are easily damaged unlike silicon nitride balls.
The RBT is mechanically less complex than the 10 axis stage required by Jensen's method.
A 3 axis stage will suffice to place the ball (sitting in a kinematic nest formed by 3 smaller balls) so that the center of the ball is coincident with the focus of the microscope objective.
It may be possible to use something like an eyepiece ( with low exit pupil aberrations) instead of an expensive long working distance microscope objective but raytracing would be needed to check the performance of a candidate eyepiece design in that application.