Typoena's screen is a sheet of 1 mm glass held without a drop of glue. Four printed posts were supposed to clamp it. Measured rather than eyeballed, they turned out to pass straight through the part they were holding, and nothing in the design said how hard the glass was being squeezed.
1The goal
Hold an e-paper panel so it never rattles, never gets over-squeezed, and can still come out with a screwdriver. No adhesive: glue on the glass makes the panel consumable, and a machine you can't open eventually gets thrown away.
2What holds the glass
Front to back: the deck's own bezel lip covers the panel's dead border, the glass drops into a pocket behind it, a foam gasket sits on its back, and a printed bracket screws to four bosses just outside the pocket. Tightening the bracket squashes the foam, which pushes the glass forward onto the lip.
Drawn to scale through one of those bosses, with the deck at the top and the cavity below:
before: the column runs through the bracket
Deck at the top, cavity below. The bracket (solid) is the part the boss is supposed to hold; the flagged band is where the two occupy the same space.
None of that is visible. The bracket is a separate printed part, so in CAD the two solids simply overlap and the software says nothing; on screen the dark bracket hides the column inside it from every angle. It shows up only if you intersect the two solids and measure what comes back. Tap a claim to see how it was tested:
3The idea
Stop the boss at the bracket, and its end face becomes the thing the bracket lands on. The screw pulls the frame down onto that shoulder and stops there. Foam squash, and therefore the load on the glass, stops being a matter of feel and becomes a number in the model.
That is the whole fix. A brittle part with no torque specification should never depend on how hard someone turned a screwdriver; put the limit in the geometry and the assembly can't be done wrong.
4The risk
Seating the bracket is only half of it. Where the seat sits also decides how much boss is left to cut a thread into, because the screw goes forward from the seat into the deck and has to stay blind: nothing may break through onto the face the writer looks at.
With the original 1 mm gasket the seat landed barely half a millimetre below the deck's inner face, leaving about one screw-diameter of thread to carry the clamp, pulling straight along the printed layer lines. A 3 mm gasket pushes the seat back and buys 4 mm of bite. Thicker foam is the better gasket anyway, since it absorbs the layer-height slop that 1 mm has no room for.
5The change
The column drops from 12 mm to 3.6 mm and the gasket goes from 1 mm to 3 mm compressed to 2.6. Everything else in the stack follows from one number: the depth of the seat.
after: the column ends on the bracket face
The boss now stops at the bracket's near face, the seat. The dashed outline is the blind M2 thread, stopping 1 mm short of the deck the writer looks at.
The same corner pulled apart along the screw, so the order of the sandwich is visible:
Exploded along the screw axis. Every thickness through the stack is the model's; the plan is cropped to one corner. The screw enters from below, through the bracket, and stops in the boss.
# hardware/case/typoena-case.scad-foam_t = 1.0;+foam_t = 3.0; # free thickness+foam_c = 2.6; # squashed, once the bracket bottoms on the seats-br_boss_len = lip_t + G_t + foam_t + bracket_t + 6; # seat depth PLUS a screw+br_seat = lip_t + G_t + foam_c; # the seat: this is the only number that matters+br_boss_len = br_seat - lip_t; # the column: pocket floor -> seat+pilot_skin = 1.0; # deck left over the blind pilot
The blind pilots also stop a full millimetre short of the outer face instead of 0.4 mm, so there is real material over them rather than two bridged layers.
6The result
before277mm³boss inside bracket
after0mm³touching, not overlapping
target 0 mm³: parts may touch, never share spacethe intersection of the boss solid with the bracket solid, measured on the built geometry
Zero, but not empty: the two parts still return a surface with no thickness, because the boss's end face and the bracket's front face are now the same plane. Coincident faces measure as zero volume, and coincident faces are what a seat is.
The rest of the re-run: the bosses clear the glass and the baseplate hardware, the pilots come back open with their 1 mm of cover intact, the bracket touches the case and nothing else, and the body is still a single connected solid, checked on the standard case and on both integrated-keyboard variants.
How it gets printed
Two of the three parts are decided for you. The baseplate and the bracket are flat plates, so they go face down and print without an overhang between them. The body is the argument.
Left: on its base, the deck spans the cavity as a shallow roof. Right: turned over, that plane is the first layer and the cavity opens upward, so nothing is left overhead.
The case is a shell that opens downward, so it has no roof except the deck. Print it sitting on its base and that deck becomes a ceiling over the entire cavity, held at 21.1° off horizontal, which is 68.9° away from vertical. Printers hold about 45°. Turned over, the same plane is the first layer on the glass, the walls lean 21° off vertical instead, and there is nothing overhead to hold up.
The cost is that the deck is the face you look at all day, and the first layer is the face that carries the texture of the bed and whatever squash the first layer gets. The engraved nameplate ends up printing against the glass too.