The cable was yellow so you'd think twice before cutting it. Half an inch thick, stiff as a garden hose, with black bands printed every 2.5 metres, and those bands told you where you were allowed to make a connection. You drilled. A jig clamped over the jacket, a bit down through to the braid, and then the vampire tap went on, one spike into the centre conductor and a couple of teeth biting the shield. Slip with the drill and you shorted the segment and the whole floor lost the network at once. The fault was usually found by whoever had been in the building longest.

Digital, Intel and Xerox put the specification out in 1980 — 10 Mbit/s, thick coax, 500 metres a segment, 50-ohm terminators on both ends or nothing works at all. Three companies agreeing on a document in public, which in that trade was not the usual arrangement.

Here's the clever bit. The protocol doesn't try to stop two machines talking at the same moment. It assumes they will. You listen, you transmit, and if you hear your own signal getting mangled you stop, send a short burst of noise so everyone knows it was a collision, then wait a random number of slot times before trying again — and the range you pick that random number from doubles each time you fail. Polite rather than precise. I never got a proper feel for why the doubling settles as gracefully as it does under load; the queueing maths was above my pay grade. It worked, though, on cable I'd drilled myself.