Let's start with the motivation - I'm trying to think of ways to talk about building coordinate systems operationally, ideally without directly using ideas like "space like geodesics" that one needs for fermi-normal coordinates. The ideas behind geodesics don't strike me as terribly complicated, but I get so many virtual blank looks when I mention them that I'd like a different approach. This is semi-inspired by the large number of people who want to build "frames of reference". Start by defining a bunch of tiny spheres of constant size - say one wavelength of krypton 86 , 605.78 nanometers. Or perhaps not so tiny, 16508 wavelengths of krypton 86, making them approximately 1cm in size. We define a sphere as the set of points n wavelengths away from the center. Close-pack the spheres in a close-centered cubic packing. For "large" spheres, the geometry of space-time would matter, for small enough spheres it shouldn't. Now imagine that the defining source of krypton 86 is pulsed in short bursts rather than continuous. The midpoint definition of simultaneity demands that the surface of the spheres all be at the same time, defining a particiular time-slice of space The cubic packing defines an array of three orthogonal spatial axes. The constant size of the spheres defines a distance scale. I would expect that this should be a realization of fermi-normal coordinates, but I' not sure how to prove it. Also, the scheme seems best suited for static space times, though I suppose you can imagine the construction working for non-static space-times, it's just that the close packed construction wouldn't be static either. [add]I'm not sure how rigorous and robust the idea of "close packed" really is. Especially if it realizes fermi-normal coordinates - we know that such coordinates are fundamentally limited in size. But I thought the idea was interesting, and I wonder if there's some refinement that would demonstrate the size limitation.