A few points:
The illumination source is a laser with substantial coherence length, so the interference is created (not projected) as a function of recombining the two beams, the direct (reference) path beam, and the test beam that is reflected from the test surface at a low angle. The diffraction gratings are what makes this method work as a fundamental principle. The lenses do some beam shaping, and prisms are simply items of convenience for beam handling to reduce the physical size of the instrument.
The calibration of the instrument is demonstrated by the measurement of the 18" optical flat (stated to be flat to 0.1 um, or about 1/6th wave), and then the subsequent subtraction of that calibration measurement as a reference map for comparison, thereby removing the anomalies that may be artifacts of the measurement system when processing the data acquisition from the object under test. This is a universal method for this sort of interferometric phase measurement map.
Stitching of a series of 1" x 15" maps can be done reasonably well by competent scientific programmers, based on matching pixels and phase information in overlapped map areas, and this is fairly well understood also. The shape of the data area in this particular instrument looks to me like a minor disadvantage in stitching, as the areas are small in one dimension, and may require a larger number of measurements to get a high-confidence stitched map in which all the sub-maps are leveled and fitted properly.
Interesting paper about trying to make a portable piece of grazing-incidence hardware; Tropel's work known to me was in fixed instruments that can measure 200mm diameter wafers and similar items.