Some time ago, I came to the realization that every real world part has some deviation from the nominal value. The question is how much.
Engineering drawings all have some implied or specified tolerance for dimensions. The tolerances specified take into account functionality of the part and cost of manufacturing.
Assuming a part is made to specification, which is often not a good assumption (think Asian manufacturing), there still can be a significant deviation from the nominal value,
When a part is reverse engineered, we measure a set of dimensions to construct a new engineering drawing. We have no idea as to what the nominal values were originally and we implicitly set new nominal values and some tolerances around those values. This new set of dimensions and tolerances may actually permit the manufacturing of a part which would be outside of the intended tolerance of the original design. I expect that many of us have run into exactly this problem. Where an aftermarket part advertised as a direct replacement for an OEM part doesn't fit. Automotive parts are a good example.
Another example is R8 tooling. Bridgeport designed and made the original buut never released the engineering drawings for the R8 taper. As a result, all aftermarket R8 tooling as well as the machines utilizing the the R8 taper were derived from reverse engineering. A good many of us have had to rework the keyway on R8 arbors. I have also had to rework the key on a Grizzly mill to keep it from binding.
So how do we utilize reverse engineering when we have no other option. One way is to make an assumption as Bruce did that dimensions were originally intended to be nice whole numbers. I do this myself and when measuring a dimension, check to see if it works in either Imperial or metric formats (I have seen mixed units in parts, mostly in those designed in metric units that throw in a few Imperial dimensions).
Another way is to assume certain symmetries in geometry. If three holes are located on a circle, we will assume the circle is concentric with other geometry and that the holes are equally spaced. When I fitted a backplate to an Asian chuck, I used that technique. I measured hole to hole dimensions of the three mounting holes. There was variation between the three measurements but when I plotted them out in a drawing and fitted a bolt circle to them, I was able to get a better picture as to what the original intent was. A check on the distance of each hole to the registration boss helped to verify the intent.
A third way is to measure a large number of parts, preferrably from the OEM and manufactured over a large period of time, and average the values. If you look at any distribution of values, dimensions include, the spread will typically follow a bell shaped curve. The true value will be somewhere close to the center of the bell. This is not always the case but even so the average value is still a better indicator of the true value