How I build a mechanical Garmin watch face: a real gear train for a watch with no gears

I build watch faces for Garmin. Some of them show a mechanical movement: Skeleton, Regulator, Skeleton Bridge and Smoked Crystal. The question I get about them is whether the wheels are only a picture. The wheels are a picture. What sits behind them is a calibre I had to solve the way a real one is solved, and this is why I bother. 1. First, a folder of photographs Every one of these faces starts as a folder of photographs of real openworked watches: a Royal Oak Openworked, a Moser Streamliner Tourbillon Skeleton, a Blancpain Villeret Squelette, and a dozen more. I spend the first days doing nothing except looking. I also drop one cheap homage in with them, on purpose, and it turned out to be the most useful image in the folder. At a glance it has everything a skeleton should have: bridges, wheels, screws, an open middle. It is also obviously wrong, and working out why is the whole of this first phase. Its spokes are all the same width. Its openings are symmetrical. Its bridges carry no bearings, so the wheels hang in space with nothing to turn on. Every surface has the same flat brushed texture. It looks like a skeleton drawn by somebody who has only ever seen one. The rest of the folder is the opposite. A skeleton is a load-bearing machine with the metal taken away down to the last safe gram, so what is left is whatever carries a force, and no spoke is wider than its job. They also all keep an unbroken ring at the outside. Chaos in the middle, order at the edge. Which is what a watch person clocks in half a second, long before they could say what they clocked. Is the movement complete, and is every part held by something? That became rule one: everything on the dial is carried by something, and that something traces back, piece by piece, to the wall of the case. 2. Solving the calibre Then I design the movement: which wheels exist, how many teeth each one has, how big it is, where it sits. Barrel, centre wheel, third, fourth, escape wheel, pallet fork, balance, plus whatever the layout needs to get drive back out to the hands. This phase decides what the finished dial looks like, and almost none of it is a drawing decision. Tooth counts come first and the radii follow. Two wheels can only work together if their teeth are the same physical size, and if you go the other way round, picking radii that look nice and deriving counts from them, you get teeth of different sizes on the same dial. That is what happened on the first pass of Skeleton: every pinion count had been taken from its radius, the mismatches ran to sixty percent, and nothing on that dial could have driven anything else. Positions get solved too. An arbor a fixed distance from two others has one place it can be, where two circles cross. On Skeleton Bridge the barrel had to sit further out than its teeth could reach, so a plain idler goes in between to pass the drive along. It changes no ratio. Sixteen teeth is the smallest wheel that spans the gap and still meshes at both ends. Other positions choose themselves, and that is my favourite thing I learned doing any of this. The fourth wheel turns once a minute, so the fourth wheel is the one that can carry a seconds hand, so the small seconds subdial goes wherever its pivot comes up through the plate. That is the whole reason small seconds sit where they do on a real watch. Nobody chose the position. The train chose it. Underneath all of it is a tension with no free answer. Big wheels spread across the dial are what make an openworked face worth looking at, meshing pulls every pair of them tight together instead, and something still has to reach the middle to drive the hands. Run the train off-centre, as the look wants, and you need a motion works chain to get back, which only fits if the barrel is small enough to leave the centre clear. Regulator began as eleven candidate layouts, drawn flat and cut down to one. The solved train of Skeleton Bridge. Each circle is a wheel, and each centre is the only point at the right distance from both of its neighbours. 3. Does the train actually run A drawn movement is a claim. Before anything gets a surface, I test it as a machine. Every mesh first, for teeth of the same size. Then the rates, from the barrel out to the balance. Skeleton beats at 28,800 an hour, eight beats a second, and the escape wheel advances one tooth per full oscillation, so one tooth every second beat. That two matters. Sources will happily pair the classic 80/10, 75/10, 70/7 train with 28,800, and that is an 18,000 train. Take one beat instead of two and the watch runs at double speed. Skeleton has a tourbillon, so the chain from the barrel to the cage has to come out at 480 to 1. Then the cage turns once a minute, which is why a tourbillon runs at sixty seconds in the first place: at that speed it doubles as the seconds indicator. Then the physical checks. Two parts at the same height may not overlap, every arbor needs a bearing above it and below it, and the click and its spring have to be there, because a mainspring that can unwind backwards is not a watch. These find things. On one pass the tourbillon cage had no pinion on it, so the wheel meant to drive it had nothing to push against, and the cage was the only object on the dial held at neither end. On another the ratchet wheel was animated at three times barrel speed, when the click holds the ratchet still the whole time the watch runs. Nobody would have named that in a review. Plenty would have felt it. Skeleton before the review of the movement (left), and after its fixes (right). 4. Could a watchmaker assemble it Running and buildable are different questions, and the second is harder. In a drawing a wheel turns perfectly well with nothing underneath it. So the render goes through a pass where it is treated as an engineering drawing, part by part, and asked whether somebody could put it together at a bench. The mainplate starts as a full disc reaching the caseband, and the openings are cut out of it. Build it the other way round, as a ring with a few arms, and it stops short of the case, and every bridge foot outside that ring is screwed to nothing. A bridge is a beam, so it wants an anchor at each end; a cock gets one, out near the edge. A jewel goes where an arbor turns and nowhere else, and a screw holding nothing gets deleted. The review of Skeleton at this stage came back with nine fatal faults, and it was the most useful hour in the project. Nine arbors had a jewel above them and open air underneath, so a lower plate went in behind the train, on skeletonised arms, reaching every one. The click spring had been screwed to the barrel, which turns; it moved onto the fixed bridge above. And the barrel had drifted far enough over the middle that its tooth roots passed a tenth of a millimetre from the hands' arbor, so the cannon pinion could never have reached the hands at all. What comes out the far side is a dial where the eye can follow any part to a bridge, the bridge to its feet, and the feet to the plate. Left, bridges cut open at every wheel. Right, continuous beams running over each arbor, with a jewel at every bearing. 5. Materials and finishing Only now does anything get a surface. The useful way to think about a finish is what it does in light rather than what it is called, and there are four behaviours. Black polish is binary: jet black at almost every angle, then blinding at one. A brushed or striped surface carries a soft bright band across its grooves, and the band sweeps along the part as the angle changes. Perlage scintillates, hard little points winking on and off inside a softer patch. Sandblasted matte simply gets brighter as it turns towards the light. Only the last of those four falls out of a roughness setting; the other three have to be built. Then each face picks from a catalogue. Rhodium plate for the mainplate, a gilt train, blued screws, and that blue is an oxide film rather than a pigment, which is why it is more saturated than paint and shifts as you tilt it. The bridges get anglage, the edge filed back to forty-five degrees and polished to a mirror, about a fifth of a millimetre wide on the real thing. There is one more trap, and it cost me a round. Polish on a part a few across comes out as a dark wire with a bright rim, because on a narrow curve the dark half of a binary finish is most of what you see. The tourbillon cage taught me that. With polished arms it read as a peace symbol. Thin parts get satin instead. The Regulator mainplate at six times magnification: a brushed face, and the filed edge reading as a single bright line. 6. Lighting the metal One light, from above, the way a ceiling lights a watch on a wrist. Then the only question that matters: what on this dial is allowed to be bright? The rule I keep coming back to is that the biggest object is never the brightest. Bridges are the largest mass of metal on a skeleton and the least interesting thing on it, and getting that backwards is most of what makes a rendered movement look cheap. I got it backwards on the first Regulator render. The two small gauge pointers were the brightest metal on the whole dial, and a made-up bounce of light off the wrist lit the movement from below, so the plate glowed at the bottom, where nothing interesting was. The answer is not to floodlight the tourbillon. A focal point wins by being different rather than brighter. On the best photograph in my folder it measures 119.5 against a movement mean of 119.9, which is to say exactly as bright as everything around it. It pulls the eye with hue and local contrast: a gold rim against black cage arms, a coloured jewel at the centre, the highest local range on the dial. That survives being shrunk to a watch screen. A brightness difference does not. Where the bright pixels belong is the filed edges, and those are lines rather than areas, each one changing along its length as the edge turns away from the