Do the cubes ever pass through each other on the way? Turn every hinge at once and in 33 of the 261 nets some do, near the end (up to a tenth of a cube deep): those cubes go red. Fold ring by ring instead (the cubes on the bottom cube first, carrying the rest flat, then the next ring out) and none of the 261 ever do. Checked here every frame: for each pair of cubes, a small linear programme finds the deepest point inside both.
Every way to unfold the tesseract is a net. Not so the 24-cell: some of its unfoldings put two octahedra in the same place (drawn red, the gold cross a point inside both). Draw random ones and keep count; about one in twenty-three collides. And a net that lies flat cleanly can still collide on the way up: with every hinge turning at once, 148 of 200 random nets have octahedra pass through each other mid-fold (red). Ring by ring, none of the 200 did.
A p-gon times a q-gon. Its cells come in two rings: p prisms on q-gons, one over each edge of the p-gon, and q prisms on p-gons. Each ring closes into a circle, and the two circles link through each other. The 4,4-duoprism is the tesseract. Small ones are all-net: every one of the 3,3's 1,296 unfoldings, the 3,4's 8,575, the 3,5's 53,824 and the 3,6's 326,592 is a net (checked one by one). Bigger ones aren't: about one in seventy unfoldings of the 6,6 overlaps, and so do some of the 4,12's, while 2,000 random draws found none for 4,6, 5,5, 5,6, 4,8 or 3,12. Folding them, the rule of the tesseract and the 24-cell holds: all at once, prisms cross; ring by ring, none did in any we sampled.
The one convex uniform polytope the kaleidoscope can't make (Conway and Guy, 1965). Built here from the 600-cell by deleting the 20 vertices of two great decagons at right angles to each other: what's left is 100 vertices, 300 tetrahedra and 20 pentagonal antiprisms, stacked pentagon to pentagon into two necklaces of ten that link through each other. It is not all-net: every one of 40 long winding unfoldings drawn overlaps (with points inside two cells by up to a third of an edge), while none of 40 bushy ones does. Fold a bushy net with every hinge at once and cells cross on the way (7 of 8 sampled); ring by ring, none did.
Cut a paper cube along seven edges and it lies flat: a net, six squares in the plane, hinged along their edges. Fold the squares up through the third dimension and it closes. A tesseract does the same one dimension up: cut it along its square faces and its eight cubes lie in our space, hinged along their squares, and they fold up through the fourth. Salvador Dalí painted one of these nets in 1954 (Corpus Hypercubus), a column of four cubes with four more round the second.
There are 261 of them. Choosing which 7 of the 24 squares to keep as hinges is a spanning tree of how the cubes touch: 82,944 ways. The page unfolds every one; none puts two cubes in the same place, and they come out as 261 different shapes, Peter Turney's count (1984). 194 of them come in left- and right-handed pairs.
Folding, each cube turns a quarter turn about the plane of its hinge square, out of our space into w, carrying the cubes hung off it. In the shadow you see them swing out (teal as they go ana) and close up into the tesseract's familiar cube-in-a-cube. In the slice (our space, where the net lies) a cube turning about a hinge square that lies in our space leaves it at once, all but the hinge itself: a 3D thing leaving a 3D world through a 2D door. Push the slice ana with the lever and you catch the rising cubes as their cross-sections; folded, our space holds only the bottom cube.
A left glove can't be turned into a right glove in our space. In four dimensions it can: turn it half a turn in the plane of x and w, and it comes back into our space with x reversed, its own mirror image, the way a paper L in a plane flips over through the third dimension. Halfway, it stands edge-on to our space: its slice is a flat section.
The snub dodecahedron comes in two hands, and so do exactly five of the 92 Johnson solids, J44–J48 (the gyroelongated bicupolae, cupolarotunda and birotunda); every other one is its own mirror image after some turn in 3D. The selftest tries every way of matching each solid to its mirror image: for these six, every match is a reflection.
The cubes' motions are exact: a quarter turn about each hinge square in the plane of the two cubes' normals, composed down the tree. Shadow: every vertex projected from an eye in w. Slice: each cube that lies in your slice drawn whole, each that crosses it as its section. Controls are CAD⁴'s: drag to look round, pinch or scroll to zoom, the pad twists your slice into w, the lever moves it, ⊞ squares it up. test/fold.selftest.mjs holds the counts (82,944 trees, none overlapping, 261 nets, 455 counting mirrors apart, Dalí's among them), the folding (rigid, hinged at every angle, closing exactly into the tesseract), and the mirror fold.