A machine that carries steel balls up a metre and drops them.

What it is

A cleated belt runs between two pulleys. Balls sit in the cleats, ride up, and fall out at the top. The frame is 1/4 inch bamboo skewers held by printed joints.

The parts

Each diagram is drawn from the real printed path, seen from above — one layer, except the weighted foot, which shows its first eight so the cavity is visible.

Drive hub

What you turn. The bore is printed round — push it onto the motor shaft while the part is still hot and the plastic moulds itself to the shaft's flat. The shaft is the mould, so the fit is exact and can't be misaligned.

60 mm · round 6.25 bore · 6.9 g

Belt pulley ×2

Slightly barrel-shaped — 0.6 mm fatter in the middle. A flat belt always climbs toward the widest part, so the bulge keeps it centred instead of walking off the end.

40 mm · +0.6 crown · 8.6 g

Bracket

Holds the pulley axle out from a bamboo upright, with 4.7 mm of clearance so a 40 mm pulley spins past the stick. Filled with rings rather than criss-cross infill — every pass follows the load around the holes instead of cutting across it.

axle 6 mm · stick 6.35 mm · 32 mm apart

Coupler

Joins two skewers end to end so the frame can be any height. A plain tube — two walls, one unbroken line.

14.35 mm outside · 6.6 bore · 3.9 wall

Spacer

Holds two uprights parallel. Use them in pairs — one plate sets the spacing but acts as a hinge, so a single spacer lets the sticks splay. Two, spread apart, fix the angle.

2 / 3 / 4 bores · 32 mm pitch

Found by building it, not by design.

Adapter

Motor shaft at one end, bamboo at the other. The shaft end goes at the bottom so that going up, the hole only ever gets bigger — nothing has to print over thin air.

6 mm shaft → 6.35 mm stick · 20 mm long

Weighted foot

A 1 m ladder with a belt running up it wants to walk. So the foot is printed as a shell with a floor and an open mouth — you fill it with sand and gypsum after printing. Sand alone damps well, because the grains rub and eat energy; the gypsum sets it so it stops migrating and becomes structural.

It is lighter to print and heavier to own: the shell is about a third of the plastic, and filled it beats solid PLA outright — sand+gypsum is roughly 1.8 g/cm³ against PLA's 1.24. Mass is the point, because it drags the whole structure's natural frequency down away from the belt's.

36.35 mm · 25 mm tall · 2.4 mm shell · ~15 cm³ of fill

Open at the top on purpose — a roof would need to bridge over the cavity, and there is no support and no bridging anywhere in this toolchain.

One plate, every part

Each part used to be its own file — which meant six heat-ups, six homings, and six chances to start a print onto a plate that had not been cleared. The whole frame now goes down in one run: 15 parts across 315 mm, packed into shelves tallest-first.

The catch is the brim. Every part's brim reaches brim_gap + brim × bead = 4.98 mm past its outline, so two neighbours need 11.16 mm between them or the brims merge and the parts come off the plate welded together — with the join hidden underneath a brim that is supposed to snap away. The first plate generated here used an 8 mm gap and would have done exactly that. It is now a check that refuses to emit the file, which is the only reason it was caught before the bed and not after.

Before you assemble: the gauge decides everything

Every bore in this build is modelled at stick + 0.25 mm, because a printed hole comes out about a quarter of a millimetre smaller than modelled. That figure is confirmed at 6 mm and was extrapolated to a 6.35 mm (¼″) stick — so one part on the plate is a gauge that tests it, and it is the first thing to pick up.

Push a real stick into each. The one that grips without splitting is the answer. Ten seconds with one stick either validates the other fourteen parts on that plate or condemns them before you have glued anything.

Why the fits differ on purpose: brackets are bored at zero clearance so they grip the upright and hold their height by friction — that is the tensioner. Spacers get 0.25 mm more so they still slide while you square the frame. Same nominal bore, opposite jobs.

How it goes together

1. Gauge first, then fill the feetSand and gypsum into the open mouths, about 15 cm³ each, and let it set before standing anything on them — loose fill damps worse than an empty foot.
2. One upright, one bracket, one pulley — turn it by handBefore building the ladder. A pulley that binds is a ten-second discovery on a single stick and an hour of disassembly on a finished frame.
3. Stand two uprightsSkewers into feet, joined with couplers for height. Spacers in pairs — one alone is a hinge.
4. Bracket at each endTop and bottom, carrying the pulley axles clear of the uprights.
5. Pulleys on the axlesCrowns facing the belt. They self-centre once it's running.
6. Belt over both, cleats outwardThe cleat side carries the balls.
7. Hub onto the motor, while hotPress it straight off the plate. It forms its own flat on the shaft.
8. Tension it by sliding a bracketThere is no separate tensioner. The bracket is a collar on a round stick, so you slide it up until the belt is taut and friction holds it. That only works because its bore is printed at zero clearance — spacers get 0.25 mm more so they still slide while you square the frame. Same part, opposite fit.
9. Tilt itVertical needs a chute wall beside the belt. At 60° the belt face holds the balls on its own, and you still get 0.99 m of lift.

Two things that were wrong first

• The cleats pulled flat. They were built by pushing the belt line outward — so the cleat's own strands were the belt, and pulling the belt taut pulled them straight. Fix: leave the outer 3 mm at each edge cleat-free. Those edges run 180 mm shorter around the loop, so they go tight first and carry everything; the cleats between them never take load.
• A ball rolled off sideways. The belt is printed on its side, so the belt's width is the print's height — which means cleat height can change across the belt. Making them 2 mm shallower in the middle turns every cleat into a valley the ball settles into.

Off the plate

The belt finished at 00:14 on 2026-07-26 — 76 minutes, 24.6 m of filament, 2.41 m of belt from a 350 mm plate.

Still open

• Nothing has been lifted yet. Every part exists; the machine has not run.
• Drive — hand crank first, motor after.
• The chute for running it fully vertical.
• Where the balls go at the top — the drop is the point; the sound is the product.

Make it yourself

These are the exact commands. The generator refuses to write a file it cannot verify — it will tell you why rather than hand you something that fails on the plate.

Every command above was executed and its output measured before being published here — an earlier version listed --part foot with no cavity arguments, which emits a solid 39 mm puck rather than the shell this page describes, and quoted the TPU belt at 230 °C, which is the temperature that jammed. Note also that the belt generator thins the requested 60 cleats to the 52 that actually fit the straight runs, and says so on every run — the page now quotes what it places, not what it was asked for.

--flat 0 on the hub is deliberate: the bore stays round and you press it onto the motor shaft while the part is still hot, so the shaft moulds its own flat. --clearance 0 grips a stick, 0.25 slides on one.