Two failures, and neither is the kind you find by being careful with arithmetic. The first is about dimension. A planar grasp analysis is cheap and everyone reaches for it, and the standard worry is that it will be optimistic: coplanar contacts obviously cannot resist a wrench out of their plane, so the spatial quality should collapse to zero. Measure it. The rank is six, not three, and the spatial wrench ball is positive. The reason is one line of algebra: at a contact whose normal lies in the plane, the tangent direction given by normal cross t-one points OUT of the plane, so the friction cone contains out-of-plane force. The grasp is not degenerate. So is the planar number optimistic or not? Here is where a second artifact hid. At the sixty-four directions this lab first shipped, the ratio read nine-tenths at three contacts and eleven-tenths at four, so it appeared to CROSS one and the direction appeared to depend on the arrangement. At four thousand directions it reads point six four, point six four and point six seven at three, four and five contacts, and across three to seven contacts it stays inside point six four to point six seven. It does not cross one. At twenty thousand directions it is lower still. So on this geometry, once the sampler is resolved enough to trust, the planar number reads consistently LOW. What makes it useless as a safety factor is not that it changes sign - that was the artifact - but that its value moves by a third as you refine the probe. There was a third claim here and it has been retracted, which is worth more than the claim was. The lab used to report that adding coplanar contacts left the spatial ball unmoved to nine digits while the planar number climbed twenty per cent - a designer optimising the cheap metric buying nothing. It was a SAMPLING ARTIFACT. Q1 is computed by probing directions on a sphere, which can only ever miss the binding one, so it is an UPPER bound that falls as you probe more finely. At sixty-four directions the coplanar values are a bit-exact tie; at four thousand they spread by three per cent, and the whole quality number falls from 0.164 to 0.064 - the shipped figure was three times optimistic. A quantity whose invariance depends on how coarsely you probe it is not invariant. Nothing about the geometry was wrong; the estimator was under-resolved and nobody had checked its convergence. The second failure is about state. Coulomb friction has two states, stuck and sliding, and real elastic contact has a third that sits between them for most of the useful range. As tangential load rises, a slipping annulus eats inward from the rim of the contact patch and the stuck core shrinks. Its radius goes as the cube root of one minus the load ratio, so the AREA goes as that squared - and a cube root near one is brutally steep. At a tenth of the friction limit the stuck area is already down to ninety-three per cent. At half, sixty-three. A monitor watching for gross slip sees nothing until the ratio hits one, by which time the object is leaving. A monitor watching the stuck area warns at under half of capacity, from the same sensor.