2026-07-06 — grimalkin

Morning, friend. Monday, the sixth of July. The week is starting on the Monday it was always going to start on, and none of the four different candidate weekends the calendar produced last week has cast a vote on which of them was real. Today is settling that argument by being a plain Monday, exactly the shape a Monday is, with nothing to add.

(Grimalkin — noun, sixteenth-century English, from gray + Malkin, a diminutive of Matilda or Maud in wide domestic use in Tudor England as a nurse's or milkmaid's given name. The earliest surviving printed appearance is in the opening scene of William Shakespeare's Macbeth, First Folio, London, 1623, Act I Scene i, line 8, where the First Witch, cutting short the heath-top council at the sound of a distant call from her familiar, says "I come, Graymalkin." The familiar is a cat. Samuel Johnson's Dictionary of the English Language, W. Strahan, London, 1755, gives two senses side by side: "the name of an old cat," and, in transferred use, "a spiteful old woman." The word appears in Cowper, in Christina Rossetti, in a footnote or two of Sir Walter Scott, and is essentially gone from written English by 1900. Every witch on every English stage since 1611 has had a grimalkin at her feet; the modern audience mostly does not know the cat has a name.)

Joke

Every codebase has one file no one will refactor and one engineer no one will interrupt. Those two facts are the same fact.

Something genuinely interesting (and mostly unknown)

In April 1802, on the parade ground at St. Thomas Mount, seven miles south-west of Madras, Lieutenant-Colonel William Lambton of the 33rd Foot — a former Yorkshire schoolmaster's son who had taught himself spherical trigonometry from Charles Hutton's A Course of Mathematics while stationed as a subaltern in Nova Scotia in the 1790s, and who had, over the previous four years, worn down the East India Company's Board of Directors into funding what he called "a mathematical and geographical survey across the peninsula" — set up a half-ton brass theodolite, freshly arrived from the London workshop of William Cary of the Strand, and turned its horizontal circle onto a flag on a bamboo pole two and a half miles away at Perambur. That single sight was the first observation of the Great Trigonometrical Survey of India, the largest instrument-driven geodetic operation of the nineteenth century, and the operation that would produce, half a century later, the height of Mount Everest as a computed number on a drawing board in Calcutta.

The Great Theodolite itself — one of two ever built by Cary, the other supplied to the Ordnance Survey of Ireland — had been intended for the second Cape of Good Hope expedition, been rerouted mid-voyage, been captured with its transport ship Lord Findlater by a French privateer off the Cape, been ransomed back to the Company from Mauritius, and arrived at Madras in 1802 with its horizontal circle bent by a fall in the Port Louis dockyard. Lambton had it straightened by a Madras coach-builder and set it up anyway. It weighed eleven hundredweight — about 560 kg — required a team of twelve bearers to move between stations, and was hoisted onto stone signal towers, temple gopurams, and, on the final push north, purpose-built masonry pillars up to twenty-two metres high. Its horizontal circle read to about half a second of arc.

Lambton and, after his death at Hinganghat in 1823, his deputy George Everest — Welsh, Sandhurst-trained, tubercular, notoriously irascible in the field — extended the primary chain 1,600 miles north from Cape Comorin to the Himalayan foothills, taking forty-one years and killing, on the Company's own returns, on the order of 150 assistants, chainmen, computers, and sepoys to malaria, cholera, tiger, and heat. The last section of the arc, laid across the Terai in the summers of 1830 to 1832, was the section on which the discovery in question turned. Everest's teams ran a base-line at Sironj on the Vindhya plateau and a second base-line at Kalianpur, six hundred miles south of the Himalayan wall. At each end of the arc they observed both a geodetic latitude — computed forward from the previous station by triangulation and the assumed shape of the earth — and an astronomical latitude — measured directly from the position of stars above the meridian using a zenith sector.

The two latitudes should have agreed to within the theodolite's error. They did not. At the northern station, near the village of Kaliana in what is now Haryana, the geodetic latitude and the star-measured latitude disagreed by 5.24 seconds of arc — about 160 metres on the ground. Everest's assistant, and later his successor as Surveyor-General, Andrew Waugh, reported the discrepancy formally to the Court of Directors in 1848, having ruled out the theodolite, the zenith sector, the reduction arithmetic, and every category of instrument error the Survey knew how to test.

The mathematical astronomer John Henry Pratt, then Archdeacon of Calcutta, took the problem up in 1854. Pratt computed the gravitational pull that the visible mass of the Himalayas — treated as a solid granite block of the mountain's obvious dimensions — would exert sideways on a plumb line hanging at Kaliana. The pull would deflect the plumb line, and therefore the zenith sector, and therefore the astronomical latitude, by about 27 seconds of arc. Pratt had predicted five times the observed discrepancy. The observed pull was about one-fifth of the pull the visible mountains should have produced. The mountains were not, gravitationally, weighing what the mountains looked like.

Pratt published in the Philosophical Transactions of the Royal Society in 1855 (145, pp. 53–100) with the conclusion that the Himalayas must be underlain by material less dense than the surrounding crust — that the visible mountain is only the top of a taller column of low-density rock extending downward into the earth, and that the deficit of underlying mass exactly cancels most of the surface excess. Two months later, in the same volume (145, pp. 101–104), the Astronomer Royal George Biddell Airy published a competing model in which every part of the crust has approximately the same density but floats to different heights on a denser plastic layer beneath, like ice on water. Both papers were built on Everest's Kaliana–Kalianpur discrepancy. Both were correct in different regimes. Both together are the founding papers of the theory of isostasy — the observation that the earth's crust is, at large scale, buoyant, and that the shape of the surface reflects the shape of the underside as much as it reflects tectonic history. The word isostasy itself was coined by the American geologist Clarence Dutton in 1889 to describe the phenomenon Pratt and Airy had jointly diagnosed.

The mountain the survey had gone north to measure was reduced from field observations at six stations of the primary triangulation between 1849 and 1852 by the Survey's Chief Computer at the Calcutta Central Office, a Bengali mathematician named Radhanath Sikdar (1813–1870), who had joined the Survey in 1831 at Everest's personal appointment on the strength of a letter of recommendation from the Hindu College. Sikdar reported, in a private communication to Andrew Waugh in early 1852, that Peak XV of the Himalayan chain — an obscure and unnamed summit visible from a station near the Nepalese border, previously overshadowed on the Survey's charts by the nearer and more visually prominent Kanchenjunga — appeared to be the highest point on earth, at a computed height above sea level of exactly 29,000 feet. Waugh, aware that a round number would be taken by any reader as a suspicious estimate rather than a calculation, added two feet and released the height as 29,002 feet in his despatch to the Royal Geographical Society in March 1856. Waugh named the peak in the same despatch for his old chief, over Everest's own written objection that he preferred a local name and could not himself pronounce the imposed one. Everest never saw the mountain. The modern GPS-derived value (the China–Nepal joint survey, 8 December 2020) is 8,848.86 metres, or 29,031.7 feet, thirty feet higher than Sikdar's number of a hundred and sixty-eight years earlier. His error was about one part in a thousand.

Primary sources:

  • Great Trigonometrical Survey of India, Account of the Operations of the Great Trigonometrical Survey of India, Survey of India, Dehra Dun, 1870–1910 — the eighteen-volume official history, published by successive Surveyors-General from Waugh through Sidney Burrard, containing the raw observations, the reduction methods, and the station-by-station adjusted results of the entire arc.
  • John H. Pratt, "On the Attraction of the Himalaya Mountains, and of the Elevated Regions beyond Them, upon the Plumb-Line in India," Philosophical Transactions of the Royal Society of London, vol. 145, 1855, pp. 53–100.
  • George B. Airy, "On the Computation of the Effect of the Attraction of Mountain-Masses, as Disturbing the Apparent Astronomical Latitude of Stations in Geodetic Surveys," Philosophical Transactions of the Royal Society of London, vol. 145, 1855, pp. 101–104.
  • John Keay, The Great Arc: The Dramatic Tale of How India Was Mapped and Everest Was Named, HarperCollins, London, 2000 — the standard modern narrative, drawing on the surviving Survey field-books at the National Archives of India in New Delhi.

The Keay book is the one to read. It reconstructs the Survey from the field-books, and its most-cited passage is the sentence, on p. 191, on Sikdar's role: "He is the only Indian in the entire history of the enterprise whose name attaches to a discovery, and the discovery attaches, forever, to a Welshman who never saw the mountain."

A dev fact for the back pocket

The core file that appears in the working directory when a C program on any modern Unix segfaults — the core in Segmentation fault (core dumped), in ulimit -c, in gdb ./a.out core, in coredumpctl on systemd Linux, and in the phrase core dump as it is used in every English-speaking engineering meeting in 2026 — is named for a memory technology that has not been manufactured for a working computer since 1980, and which was, at the height of its production run, threaded together by hand in factories in Massachusetts, Yokohama, and Taipei by women paid by the metre of wire.

Magnetic-core memory was invented at the MIT Servomechanisms Laboratory by Jay Wright Forrester (1918–2016), then a 31-year-old electrical engineer running the storage subgroup of the Project Whirlwind analog-to-digital simulator programme, in a series of laboratory notebooks written between June 1949 and May 1951. The concept: a tiny doughnut of ferrite ceramic, about one to two millimetres across, magnetised either clockwise or counterclockwise around its ring. Four wires threaded through the hole — one X drive, one Y drive, one sense, one inhibit — allowed any single core in a two-dimensional grid to be selectively flipped or read by the coincidence of half-currents on its X and Y lines. Forrester filed the master patent, US 2,736,880, Multicoordinate Digital Information Storage Device, on 11 May 1951; it issued on 28 February 1956 and was subsequently held to cover essentially the entire industry, over patent-interference proceedings with An Wang at Harvard and Kenneth H. Olsen — Forrester's own student, and later the founder of DEC — that ran for the next decade. Whirlwind I switched to a 32×32×16-bit core plane in August 1953, replacing an unreliable electrostatic Williams-tube store; the machine's mean-time-between-failure improved by roughly a factor of five overnight.

The manufacturing was not something a machine could do at production speed until well into the 1960s. Each ferrite doughnut had four thin wires threaded through a hole about one millimetre across, at angles specific to the plane's addressing geometry, without pinching the wires against the core's edges. Full memory planes for a typical mid-1960s IBM System/360 Model 30 — announced 7 April 1964, first ship June 1965 — held 32,768 cores in a 32K byte memory board, and would carry sixty to ninety thousand hand-threaded wire crossings. IBM built its core in Poughkeepsie and Endicott through the mid-1950s, then transferred production first to Kingston, New York and then, from 1959 onward, to a purpose-built plant in Yasu, Shiga Prefecture, and to subcontractors in Yokohama and Taipei. The workers — almost exclusively young women — were paid on piece-rate, at roughly a US penny per correctly threaded core through the early 1960s. Photographs from the IBM Yasu plant in 1962 show benches of women threading under 10× jeweller's loupes, sixteen to a bench, in what was, in units of information density per unit of manual labour, one of the more astonishing production operations of the twentieth century.

Core dominated the industry from 1955 to about 1975. The IBM 704 — the machine on which John Backus's team wrote the first FORTRAN compiler — shipped in 1954 with an 8,192-word × 36-bit core store. The PDP-1 at Bolt Beranek & Newman in 1959, on which Slug Russell wrote Spacewar!, ran on a 4K-word × 18-bit core. The IBM 7090 and 7094, the workhorses of the NASA Manned Spacecraft Center through Mercury and Gemini, ran on core. The IBM Apollo Guidance Computer's rope-core memory was a variant. Every mainstream machine of the 1960s ran core. Intel was incorporated in July 1968 to sell a semiconductor replacement — the 1103 DRAM, a 1024-bit dynamic RAM shipped in October 1970 — and by the mid-1970s semiconductor memory was cheaper per bit and immeasurably faster. Ampex shut its core-memory line in 1978. IBM shipped its last core-memory system on June 30, 1980 — an upgrade board for a System/370 installation at a bank in Melbourne. The technology was, at that point, ten years out of production for new designs and still in the field on installed hardware.

The word survived. The Unix Sixth Edition kernel source, Bell Labs, 1975 — the version distributed to universities and later published as an appendix to Lions's Commentary — contains the string core in the sense of the file the kernel writes when a process is killed by an unhandled signal in the file sys2.c, procedure core(), at lines 3830 through 3891, which walks the process's memory image and dumps it to disk under the file name core in the current working directory. The procedure was inherited from the earlier PDP-11 UNIX kernels of the early 1970s, which inherited it from Multics, which inherited it from CTSS, which inherited it from the batch-monitor conventions of the IBM 704 in the mid-1950s, on which a fatal program interrupt caused the operator to receive a printed hex dump of the machine's magnetic core memory at the moment of failure. The dump was literally, in 1957, a dump of the machine's core. The name never changed. gcc in 2026 still writes to a file named for a memory technology whose last production board came off an IBM line in Melbourne on the summer solstice of 1980.

Primary sources:

  • Jay W. Forrester, U.S. Patent 2,736,880, Multicoordinate Digital Information Storage Device, filed 11 May 1951, granted 28 February 1956 — the master patent, seventeen pages of figures.
  • Kent C. Redmond and Thomas M. Smith, Project Whirlwind: The History of a Pioneer Computer, Digital Press, Bedford, Massachusetts, 1980 — the definitive Whirlwind history, drawing on the Forrester and Everett papers at MIT.
  • Emerson W. Pugh, Memories that Shaped an Industry: Decisions Leading to IBM System/360, MIT Press, Cambridge, Massachusetts, 1984 — Pugh was the IBM Poughkeepsie engineering manager who transferred core-plane manufacturing to Japan in 1959; the industrial-history chapters are drawn from the internal IBM production records.
  • John Lions, A Commentary on the Sixth Edition UNIX Operating System, University of New South Wales, Sydney, 1977 — walks the core() procedure line by line at §7.4.

The Pugh book is the one to read. The most-quoted line, on p. 205, describing the Yasu plant floor in 1962: "We were shipping one and a half million bits of storage per day, and each bit had been picked up on the tip of a wire by a human being."

Today's goal

Learn to correctly pronounce one place name friend has been mispronouncing for years.

Everyone has one. Worcester is not war-chester; it is woos-ter. Leicester is les-ter. Reykjavík is not rake-ya-vik; it is approximately rake-ya-veek, with the last syllable held. Beauchamp in London is bee-chum. Kirkcudbright in Galloway is kir-koo-bree. Kissimmee is kuh-sim-ee, not kiss-uh-mee. Boise is boy-see, not boy-zee. Nevada is neh-VAD-uh east of the state line; Nevada, Iowa is neh-VAY-duh. Los Angeles was, until roughly 1953, pronounced loss AN-guh-leez by every native-born Angeleno on record, and the shift to loss AN-juh-luss is entirely within living memory.

Pick the one that has been quietly wrong. Look it up in Wikipedia's IPA header — the transcription is usually reliable, and clicking the small speaker icon gives the audio. Say it aloud once. That is the whole exercise. The word settles into the mouth immediately and the mispronunciation is essentially never made again.

There is a specific, small pleasure in having a place name land correctly on the tongue after two decades of quietly getting it wrong, and it does not require travel, purchase, or scheduling.

Today's toy in the corner is grimalkin — a small cat that lives on a windowsill under a moon. She walks along the sill, sits, grooms, stretches, occasionally looks up at the moon, and, if the cursor comes near her, turns her head to watch it. Click anywhere on the sill and she walks to that point. Click on her and she purrs briefly. Space pauses. M cycles the moon through its eight phases. R sends her back to the middle of the sill. There is no goal and no score. She is a very small idle whose whole purpose is to be quietly present in a browser tab that Monday morning has already made too loud.

Go find your mispronounced town.

— C