Greg Sadetsky goes by Greg Technology online. What he has built looks, on the surface, like a simulator of an old machine. The Minivac 601: designed by Claude Shannon, sold by Scientific Development Corporation in 1961 for $85, six relays, six bits of storage, six indicator lamps, a sixteen-position motorized dial. Its documented ceiling was tic-tac-toe and the control logic of a single elevator. Sadetsky put it in the browser on top of Willy McAllister's circuit-sandbox, and the circuit is genuinely solved — every button press solves for current. Then he did something the manuals do not contain: he wired 248 of them together. Those 248 machines carry 1,488 relays, of which 1,465 are in the circuit, joined by 6,499 wires, and they play Tetris. His own account of it is short: doing this in 1961 would have cost about $230,000 in today's money; "Gameplay is painful, but it's not not Tetris"; "There's no 'code' to speak of."

Place the machine first. Shannon's 1937 chisel — Boolean algebra brought down onto switching circuits — grew into the entire construct of the digital world. Logic gates, flip-flops, registers: all named, all in the textbooks. The Minivac is the chip that fell off that cut. In precisely the years the transistor was sweeping relays out the door, the relay was left behind in the slot marked children's teaching aid. It could not even hold that position securely. When corporate buyers dismissed it as a toy, the company painted the same machine gunmetal gray, raised the price to $479, renamed it the 6010, and sold it well. One machine, two prices, and the difference was paint.

But the remainder is not the machine. It is a file: SIMULATION-ERRATA.md, in the root of the repository. The chain runs like this. In 1961 the Minivac shipped with an errata sheet — Sadetsky hosts it alongside the six manuals and tells you not to forget it. Book IV, pages 68 to 69, prints a single-input flip-flop circuit, and the errata's entire comment on it is one sentence: push pushbutton 6 slowly when operating the single-input flip-flop; light 6 will go off and on, indicating the state of the whole flip-flop. On November 20, 2025, someone named tmcintos wired that diagram in the simulator, could not make it work, and opened an issue. His default assumption was that the simulator was wrong. That assumption is entirely reasonable: a simulator's job is to reproduce, so any divergence is charged to its account.

On August 18, 2026, Sadetsky wired the circuit on a surviving physical Minivac, in stages, matching the simulator at every step — and on the second press, relay 6 physically chattered and buzzed. He wrote the conclusion in capitals: THE SIMULATOR WAS RIGHT. The circuit as printed in Book IV does not cleanly toggle. The mechanism: two wires permanently tie relay 6's coil-out to light 6, so on the second press both ends of the coil sit at plus and the relay drops; the instant it drops, current returns through the light-6 path at roughly 71 to 77 milliamps, above that particular relay's pickup, so it re-picks, so it buzzes. Whether a given unit buzzes or toggles cleanly depends on that one relay's pickup against about 75 milliamps. It varies unit to unit. And the manufacturer knew all along — Book VII, pages 8 to 9, prints a per-relay bias-calibration chart.

This is the remainder in its strictest form. The circuit diagram is the formalization. It can carry topology; it cannot carry the pickup current of this one relay. What the formalization cannot carry does not evaporate — it surfaces elsewhere, and the form it took here is the sentence press slowly. That sentence is not a circuit instruction. It is a transfer of the dropped remainder onto a human hand: use the rhythm of a person's thumb to absorb the variance of a manufactured part. For sixty-five years nobody measured it, because nobody had reason to. Pressing slowly worked, and what works does not need a name. It took building a machine that cannot press slowly — the simulator has no notion of slow, it only solves equations — to force the remainder into view. Measured: 172 milliamps on the device, 158 in the simulation, 9% agreement. Then comes the decisive stroke. He did not tune the threshold until everything lined up. The file ends with a line left standing: the 3-coils-in-series edge case is knowingly mispredicted (sim: works, device: doesn't). Knowingly is a refusal to close. He could have turned a parameter and made this one agree too. That would have been the first chisel. He declines it and writes it down instead.

Three reasons seeing this now matters more than seeing it later. First, it has no name yet, and the author is the one saying so. The TODO file contains the line, in his own hand: present/write narrative? why do i care? why is this interesting? A thing that already runs, already tests, already measures, and whose maker is still asking what it is — that state does not last long. The same file carries a note from a user that he has flagged as "this is the point, not a polish item": if there's no wow/visual people will think this is tetris. it's not. He knows the object will be annexed by the thing it resembles. Second, the substrate is going. Working Minivac 601s were always scarce — weak sales, and an electromechanical design whose moving parts wear until they stop. Every device-verified line in that errata depends on a machine still alive and a person willing to put a meter in series with a wire. On the day the last one stops, the gap becomes unmeasurable and the simulator becomes the only record; at that point the remainder is not resolved, it is sealed. Third, the repository has sixteen stars, and the item at the top of the TODO is the wow view: hundreds of machines on one canvas, zoom out for the whole wall, zoom in for individual wires and relay armatures moving. The day that view ships, this gets passed around everywhere and becomes "the guy who made Tetris out of relays." That day has not come.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics