Picture the bench from the operator’s stool: a dozen cranks waving up and down, each carrying one constituent — and a single pen that must draw their sum. Gears multiply; they do not add. The solution, used in every predictor from Kelvin’s prototype to Machine No. 2, was a cascade of floating pulleys and one long wire. It is the most beautiful mechanism in the archive, and the subtlest.
One pulley adds two motions
Picture a single pulley hanging on a wire: the left end of the wire is pulled up by crank A, the right end by crank B, and the pulley itself floats between them. The pulley’s height is the average of the two ends — half the sum. Chain the output onward (or double the crank throws to compensate), and you have addition. The sketch above shows three cranks feeding one line: each floating pulley folds its crank’s motion into the common wire, and the free end — connected to the pen carriage — carries the total.
The cascade
With N constituents you need N−1 floating pulleys in series (Kelvin’s ten cranks used nine; No. 2’s thirty-seven used thirty-six). The wire snakes from pin to pin across the whole frame — on big machines the total run measured tens of metres. German engineers shortened and symmetrised these runs deliberately; Bidston’s team obsessed over pulley bearings. Everyone understood the same arithmetic of loss: each pulley steals a little motion to friction, and the small constituents — the very waves Doodson’s rigour had won — drown first.
Why wires, not levers?
Lever adders existed on paper and in a few analysers, but wires won for predictors: they route around corners via guide pulleys, they weigh almost nothing, and they tolerate the long throws of big spring tides without binding. Their vices were stretch (temperature and humidity moved the pen) and wear (a fraying wire corrupts a whole run silently — operators inspected them like rigging, which several operators had been).
The operator’s view
On setup day, the adder was checked before anything else: with all pins wound home, the pen had to sit dead still through a full crank revolution. Any drift meant a sticking pulley — found by feel, freed with oil, logged in the book. Veterans claimed they could diagnose a machine blindfold by the sound of its pulleys.
Doubts, answered
Is the sum exact? In principle, yes — analogue addition has no rounding error. In practice, friction, stretch, and backlash set the floor, which is exactly why 37 cranks was the economic limit.
Did any machine add differently? A few late designs experimented with electrical summation (resolvers adding voltages), a halfway house to the digital era. They worked, but by then the stored-program computer had already won.
What should I trace on a museum machine? Pick one crank wire and follow it with your eyes all the way to the pen — the full journey I recommend in the museums guide. It is the best five minutes in the gallery.
Listen to the pulleys
Veterans claimed they could diagnose a machine blindfold by the sound of its cascade — a sticking pulley clicks, a fraying wire whispers, a healthy machine hums evenly through the whole run. You cannot hear brass on this page, but you can trace one wire with your eyes from crank to pen in the sketch above, and then do it for real on any machine in the museums guide. For the gears feeding the cascade, read the gear primer; for the giant that stretched it furthest, Inside Machine No. 2.