Beat Error sits you at a watchmaker’s bench with a movement in the holder and a timegrapher on the desk. First you build the going train — fit each wheel in the order power actually travels — and then comes the part that fills a real watchmaker’s day: regulation. Two lines of dots crawl across the timegrapher, their slope is your rate, and the numbers a spec sheet throws around — −4/+6 seconds a day, chronometer certified, tested in five positions — turn out to describe exactly this bench work. It lives on the Watches library because that library keeps using those numbers, and this is the machine behind them.
How to play
Everything works with a mouse or a fingertip; on a keyboard, TAB cycles parts, ENTER fits them, P changes position, SPACE runs the movement, and R restarts. Assembly is strict about order because power is: mainspring barrel first, then centre, third and fourth wheels, the escape wheel, the pallet fork, and finally the balance and its hairspring. Try to seat a part early and the bench tells you why it does not go in yet.
Once the train runs, the timegrapher wakes up and the real game starts. You have three sliders and three position buttons. The regulator changes rate; the stud changes beat — they are different screws for different problems, and confusing them is the classic beginner mistake the game exists to un-teach. Read the trace dial up, crown down and crown left, find a compromise all three can live with, and case it up. There are three movements on the bench: a forgiving modern workhorse, a chronometer-grade calibre with much tighter tolerances, and a sixty-year-old vintage piece whose hairspring sits slightly off-centre and whose mainspring is tired. Nothing here runs on a clock except the one you are regulating.
The physics on the bench
The trace is not an animation of success; it is a readout of a simulated oscillator. The regulator index moves the effective length of the hairspring — longer runs slower, about 0.6 seconds a day per unit, which is the whole mechanism behind the fast/slow marks on a balance cock. The stud lever shifts the hairspring collet and moves the beat error by fractions of a millisecond; a healthy movement wants it under 0.5 ms. Gravity pulls on the balance differently in every position, so each position carries its own rate shift, and a good regulation is a compromise across all three rather than a perfect number in one.
Amplitude behaves too: it falls as the mainspring unwinds and again when the movement leaves dial-up, and the simulation honours isochronism — a low-amplitude balance runs slow on most escapements, and it scatters its dots where a healthy one draws two clean lines. Badly out of beat loses a little time as well. The verdicts — chronometer work, good regulation, serviceable — are read from those numbers, not from how tidy your clicking looked.
How it was built
Beat Error is hand-written WebGL and JavaScript, no engine, and the commit history is a diary of taking the movement seriously as an object. The wheels “stopped being stickers on a plate” when they got real depth and shadow; the first pallet fork “was aimed at nothing in particular” until its jewelled stones were angled to actually lock and release the escape wheel’s club teeth; the balance was at one point “quietly emptying the matrix stack” every frame. The decoration is geometry rather than texture — the Côtes de Genève on the bridges are modelled waves, so the highlight really walks across them as you orbit the bench — and the hairspring is drawn at its real proportions: eleven microns of Nivarox, a fine dark line whose outer turn rises like a Breguet overcoil.
The timegrapher had its own lesson: the trace originally ran on the wall clock and stuttered, so it settled on frame count instead, with a fixed jitter pattern that keeps the dots looking alive without dancing. If your trace slopes, reach for the regulator. If the two lines split apart, that is the stud. They are different screws.