Our ability to image individual atoms and atom-columns only brings the practical problem of finding a statistically-useful number of nanoscale structures into sharper focus. In crystalline materials like metals and semiconductors, a key tool for locating lattice defects has been diffraction contrast from defect strain fields. For example, the lattice can be oriented just off the diffracting condition in darkfield, at which point defect strain fields (even at low magnification) light up like stars in the night sky.An important challenge in the gigascale silicon-device industry is the management of oxygen-related defects as allies (e.g. for "impurity gettering") in the device-making process. However, nucleation and growth of oxygen clusters from ~10-20 ppma of dissolved interstitial oxygen O i is a complex process, predicated on thermal history in the 600 o C range, and involving electrically active thermal donors, lattice vacancies early on [1], a variety of precipitate "shape changes" [2,3] e.g. from unstrained monolayer plates to (111) octahedra to platelets on (100), and expansion-related silicon self-interstitial dislocation loops and stacking faults as more and more oxygen comes out of solution. The population of oxygen clusters with fewer than 10 5 oxygen atoms, often associated with thermal history during crystal growth, is of special importance. However it has been resistant to quantitative characterization because of unstrained-configurations and surface O-intrusions [4] in that size range. Because silicon's diamond lattice naturally cleaves on (111), the preference for strained platelet formation on (100) is a mystery. This unexpected break in symmetry might be explained by anecdotal evidence [5] that unstrained monolayer oxygen-molecule or "ninja-plates" naturally form on (100) because of lattice potential considerations, but are very hard to find given their lack of strain. At some point these "decloak" [6] to form strained octahedra (cf. Fig. 4). As size increases and surface energies become less important, these octahedral clusters return to the original (100) plate for 2D growth above say 10-nm in size, as depicted in Fig. 1. What TEM images of the strained octahedral won't reveal are the broken bonds left by the O i as the puddle drains to form the octahedral defect. Figure 2 shows the three stage model for the ninja defect uncloaking in plot form.The problem is that systematic study of oxygen clusters below 10 nm in one or more dimensions has been hampered by number densities e.g. in the 10 9 /cc range, and by 10 nm (weakly strained) oxide intrusions which form (even during TEM observation!) on freshly ion-milled silicon surfaces exposed (even briefly) to air on their way into the TEM. As discussed previously [7], high-oxygen and highvacancy silicon results in more like 10 11 instead of 10 9 oxygen precipitates per cubic centimeter, with their panoply of extended defects (cf. Fig. 3) for impurity gettering. This means that the "sub-10 nm" cluster population is at least that abundant.We are thus e...