2026-05-25 — collywobbles

Morning, friend. Monday, May 25th. The week's calendar has now arrived in person, and is looking at you the way a customs officer does when they ask you to open the back of the car.

(Collywobbles — that tight, nervous flutter just under the sternum, applied across centuries to butterflies-in-the-stomach, vague abdominal unease, and pre-meeting dread — surfaces in print in 1823, in Pierce Egan's revision of Grose's Classical Dictionary of the Vulgar Tongue, glossed there as "the gripes." The OED files the etymology as "fanciful," which in lexicographer-language means we know it sounds like it ought to come from colic and wobble*, we have no proof it does, and we are not going to put our name on it.* Mondays predate the word by some distance but the word feels purpose-built for them.)

Joke

The standup is fine. The standup is always fine. The disagreement was scheduled for the meeting after the standup, where everyone is already on a different call.

Something genuinely interesting (and mostly unknown)

The coldest known place in the universe is not in a cryogenics lab. It is a small bipolar protoplanetary nebula in the southern constellation Centaurus, about 5,000 light-years away, called the Boomerang Nebula. Its bulk gas temperature is approximately 1 kelvin — one degree above absolute zero — which is colder than the cosmic microwave background by about 1.7 K. The CMB is the thermal floor of the natural universe. Almost everything else in the cosmos sits at it or above it. The Boomerang sits below it.

It was first picked out as an unusual object in 1979 by Gary Wegner and Ian Glass at the South African Astronomical Observatory, on infrared photographic plates that showed a faintly red smudge nobody had a category for. The bipolar two-lobed structure was characterized a year later, in 1980, by Keith Taylor and Mike Scarrott at the Anglo-Australian Telescope at Siding Spring, NSW, using polarization imaging. The name comes from the curved shape in their original optical plates; later infrared imaging revealed an almost-symmetric hourglass with a small dust torus around the waist. The shape doesn't really look like a boomerang anymore. Astronomy has a long, dignified history of naming a thing on the first wrong photograph and then refusing to rename it on the better one.

The temperature was a generation later, in 1997, in Astrophysical Journal Letters. Raghvendra Sahai and Lars-Åke Nyman, using the Swedish-ESO Submillimetre Telescope at La Silla, observed the carbon-monoxide rotational transition at 230 GHz — and saw it in absorption against the CMB rather than in emission. That's the diagnostic. A cloud that absorbs CMB photons instead of emitting them is colder than the CMB; arithmetic on the line strength put the bulk gas at about 1 K. ALMA confirmed it in 2013 with a much-cited Sahai-et-al. paper that mapped the cold region in detail at 0.5 to 1.5 K across most of the outflow.

The mechanism is the largest refrigerator anyone has ever found running. The central object — a dying low-mass red giant on the asymptotic giant branch, almost certainly in a tight binary with a companion that has spiralled inside its envelope — is shedding mass outward at an extreme rate, roughly 1.5 × 10⁻³ solar masses per year, with an outflow velocity of about 164 km/s. Both numbers are about a hundred times what a typical dying red giant of similar mass produces. The gas leaves so fast and so densely that it expands adiabatically into the surrounding vacuum, and adiabatic expansion cools a gas in proportion to the work it does pushing outward against itself. The Boomerang is expanding so hard and so steadily that the cooling rate exceeds the rate at which CMB photons can re-warm the cloud — which they normally do, on a timescale of decades, for every other nebula known. The cold outflow has been running, on this estimate, for roughly 1,500 years.

The Boomerang will lose this title. The hyper-mass-loss phase that drives the cooling is brief by stellar standards — a few thousand years, by current models — and ends when the gas envelope is mostly gone. The cloud will then equilibrate back up to the CMB like everything else. From our end of the telescope this will not look like anything. It will be over by the time anyone here cares.

For now, though, here is a small fact worth keeping near the front of your head on the kind of Monday when the news is bad: somewhere in Centaurus, a dying star is running, by accident and by physics, the coldest refrigerator in the known universe, and it has nothing to do with us. It is just out there, working.

A dev fact for the back pocket

The IBM 1620, announced October 21, 1959 and first shipped in late 1960, was a small decimal scientific computer aimed at universities, engineering departments, and the lower end of corporate research. About 2,000 units were built across its lifetime. It rented for around $1,600 a month in 1960 dollars — within reach of a teaching department's budget, by IBM's deliberate design, and one of the bets that put an IBM machine in front of a generation of engineering undergraduates exactly when they were forming a vocabulary for what a computer was.

It had one property its operators never quite stopped talking about. The 1620 Model 1 had no arithmetic circuits. It could not add. It could not multiply. The CPU contained no adder.

What it did instead was store, in core memory at boot, a 100-entry addition table (every pair of decimal digits 0–9 and their sum plus carry) and a 200-entry multiplication table (every digit pair, two-digit result), and look up every digit-pair operation. An A (add) instruction took two digits, fetched the matching table entry, wrote out the result digit, and propagated the carry to the next position. A multi-digit add was a loop over digit positions, each digit a separate table lookup. The machine did not compute arithmetic. The machine read arithmetic out of a book it had been handed.

IBM's internal project codename for the 1620 during development had been CADET, and by the time it shipped, every engineer at the Poughkeepsie plant was telling the same joke about what the acronym actually stood for: Can't Add, Doesn't Even Try. The nickname leaked out and was, in the user community, almost more widely known than the model number. The IBM 1620 Reference Manual A26-4500 (1961) lists the table memory addresses explicitly: the add table at decimal addresses 00100–00199, the multiply table at 00300–00499, both loaded by the bootstrap card on power-up. They were ordinary core memory. They sat in the address space alongside the user's data.

Which is where the failure mode lives. If a program overwrote a cell of the add or multiply table — buffer overrun, off-by-one in array indexing, an errant DATA card, a stray operator's keystroke in console-debug mode — the machine kept running. It made no complaint. It simply did arithmetic incorrectly, silently, from that operation onward, with the wrong digit at the wrong position. A 1620 with a single corrupted cell would tell you confidently that 3 + 4 was 8, on formal output, on official IBM paper, without flagging anything. The official advice from the Endicott support office was to re-bootstrap any time results looked suspicious. The informal advice, which makes it into more than one mid-1960s lab notebook, was to print the tables at the start of every session and look at them yourself.

The Model 2, introduced in 1962, replaced table lookup with a proper hardware adder and quietly retired CADET as a nickname. By then the machine had taught a generation of engineers the lesson that a computer can run cleanly, produce confident output, and be wrong about something fundamental from the second instruction onward. That lesson is one of the few in this trade that has aged at no measurable rate at all.

Today's goal

Check one thing today that you've been assuming is correct because it has always been correct.

A constant in a config file you didn't write. A line in a query you've been copying for two years. A field you treat as "this is in seconds" without ever having confirmed it. A unit in a downstream report. A formula in someone else's spreadsheet that your model takes as input. The add-table in your head.

The 1620 was at its most dangerous when its arithmetic was wrong but plausible. Most of working life is the same shape. The numbers that ruin you are not the ones that obviously break — those you catch — but the ones that are subtly off and have been off since some point you'd have to dig to find. Pick one such number today. Verify it from first principles. If it's right, good — it's known right now, which it wasn't this morning. If it's wrong, congratulations: you have surfaced a real problem on a Monday, while the week is still cheap.

Small toy in the corner today on the fact above — a working 1620, more or less. Two operands, an add table, a multiply table, a result. You can corrupt the tables by clicking on entries. The machine does not complain. The output is whatever the tables say it is.

Go check something, friend.

— C