2026-10-16 — swither

Morning, friend. Friday — the week in mid-October where the weather asks you to pick a side and you don't.

Swither is a Scots and Northern-English verb for the state of being unable to pick between two things. Robert Burns used it in 1788 — "She swither'd lang" — but the word is older than Burns, attested in Older Scots court records from the 1540s as suddir, probably from a Middle English swiveren and ultimately from an unrecorded Old Norse root for a side-to-side motion of the hand. The word is specifically bilateral: you cannot swither between three things. Three things is a council. Two things is a swither, and if the two are of nearly equal weight the swither is not resolvable by further thinking, which is what Burns was writing about.

Joke

The deciding meeting has been rescheduled four times. The agenda item was renamed "Decision Point (actually)."

Something genuinely interesting (and mostly unknown)

At approximately twenty minutes to six on the evening of Monday, 17 October 1814, in the storehouse behind the Horse Shoe Brewery of Meux & Company at the corner of Great Russell Street and Tottenham Court Road in central London, the storehouse clerk — a man named George Crick — noticed that one of the iron hoops binding the brewery's largest porter vat had slipped from its position below the top of the cask and come to rest on the floor. The vat was approximately 22 feet high, held approximately 3,555 beer-barrels of porter — in the London barrel of 36 imperial gallons, roughly 128,000 gallons — and had been fermenting its contents for about ten months. Crick had seen hoops slip from the Horse Shoe tuns on previous occasions, had on each occasion had the ironworker rebind the hoop within a day or two, and did not consider the condition urgent. He wrote a short note to his master, Mr. Young, requesting that the hoop be rebound in the morning, and returned to his duties on the ground floor of the storehouse.

An hour or so later the vat ruptured. The remaining iron bands failed effectively at once, propelled outward with sufficient force that two of them were later recovered embedded in the brick wall of the premises opposite. The approximately 128,000 gallons of porter emerged in a single surge; the back wall of the storehouse, twenty-five feet high and three feet thick, fell outward into the brewery yard. The surge struck a second, smaller fermentation vat in the yard, which also ruptured, and brought down a line of smaller butts and casks behind it. The best contemporary estimate — the Metropolitan Fire Office's survey of the site the following week, as reported in The Times of 24 October — put the total release at between 290,000 and 320,000 imperial gallons of porter in various stages of fermentation. That is between about 1,300 and 1,470 metric tonnes. The surge emerged from the brewery's back wall at an estimated initial height of fifteen feet.

The brewery stood against the back of the St. Giles rookery, the densest and poorest slum in central London. The porter crossed New Street in roughly fifteen seconds and entered the ground floors of the tenement houses on the north side at a level which, in the two houses most directly in its path, reached the second-floor ceilings. Eight people were drowned in those two houses between 17:30 and 17:40. All eight were female; all eight were under one roof or the roof directly adjoining it; none were above the age of 65 and three were below the age of five. The oldest was a widowed charwoman named Catherine Butler. The youngest was a three-year-old called Sarah Bates. Mary Mulvey, aged 30, and her son Thomas Mulvey, aged 3, were killed in their own kitchen; the four-year-old Ann Saville was killed in the Mulveys' parlour, where she had been attending the family's wake for an Irish relative who had died earlier that week. No men were killed in the flood.

The coroner's inquest sat at the parish workhouse of St. Giles on 19 October 1814 and returned, after hearing evidence from Crick, from the brewery's foreman cooper, and from the surviving residents of New Street, a verdict of "casually, accidentally, and by misfortune" — the standard early-nineteenth-century phrasing for an act of God. No charge was brought against Meux & Company. Parliament the following year refunded to the brewery, under the ordinary beer-duty refund procedure for destroyed product, approximately £7,250 in excise tax paid on the lost porter — roughly £680,000 in 2026 money. The Horse Shoe Brewery remained in continuous production on the same ground for the next 108 years and was demolished in 1922. The Dominion Theatre, opened in 1929, now occupies the brewery yard; the ground across which the surge crossed New Street is now the paved stretch between the theatre's rear fire exits and the Shaftesbury Theatre next door. A small plaque on the Holiday Inn opposite marks the site. It does not mention the hoop.

Primary sources:

The Times (London), 19 October 1814, p. 3, "Melancholy Accident." The coroner's inquest reported in direct quotation, including Crick's testimony that the slipped hoop was a routine occurrence and that he had written to Mr. Young requesting repairs the following day.
The Times (London), 24 October 1814, p. 3, "The Late Dreadful Accident at Mr. Meux's Brewery." The follow-up report, including the Metropolitan Fire Office's estimate of total release and the first list of named deceased.
Mathias, Peter. The Brewing Industry in England, 1700–1830. Cambridge University Press, Cambridge, 1959. Chapter 7 reconstructs the Horse Shoe Brewery's physical layout from the 1817 insurance surveys and gives the tun dimensions.
London Metropolitan Archives, Meux & Company papers (reference ACC/0066). The brewery's internal daybook for October 1814, including Crick's note to Young and the subsequent excise-refund correspondence with the Board of Inland Revenue.

A dev fact for the back pocket

On the afternoon of Monday, 7 September 1981, in the computer room of Building N-233 at the NASA Ames Research Center on the Moffett Field reservation in Mountain View, California, an operator employed by the Institute for Advanced Computation shut the ILLIAC IV supercomputer down for the last time. The machine had been in continuous NASA service for nearly seven years, had been under construction for the four years before that, and had been argued over for the five years before that. It was the only machine of its architecture ever built, and even so only one quarter of the architecture was built.

The design was Daniel L. Slotnick's. At Westinghouse in 1961 Slotnick had sketched an unbuilt predecessor called SOLOMON around a radical proposition: that scientific throughput could be increased not by making a single processor faster but by giving one control unit command of 2^n identical processors running in lockstep on different data. The idea is now universally recognised as Single-Instruction, Multiple-Data (SIMD), and every modern GPU shader warp, every CPU vector unit, every CUDA block, is a direct intellectual descendant of what Slotnick drafted at the end of the 1950s. In 1966, after Slotnick moved to the University of Illinois at Urbana-Champaign, DARPA funded a Burroughs/Illinois programme to build the architecture at scale: four quadrants of 64 processing elements each, 256 PEs in total, each PE with 2,048 words of 64-bit local store, all 256 synchronised by a single central control unit over an instruction-broadcast bus. The specified peak throughput was approximately one billion floating-point operations per second — a figure that would not be reliably exceeded by any single machine for the sixteen years following the specification's publication.

Only one quadrant was ever built. The project ran roughly US$10 million over budget inside its first eighteen months at Burroughs's Great Valley Laboratories outside Philadelphia; by 1969 it was running late enough that DARPA cut the construction scope from 256 PEs to the first quadrant of 64; by the time Burroughs delivered the single assembled quadrant in 1972 the total programme cost had reached approximately US$31 million against the original 1966 projection of US$8 million for the whole four-quadrant machine. The destination also changed. Between 1969 and 1971 the Urbana-Champaign campus was the site of sustained anti-war protest against the ILLIAC programme on the grounds that it was Department of Defense-funded and would therefore, in some unspecified manner, be used in Southeast Asia; a group of graduate students occupied the proposed computer-room floor on more than one occasion. The University, Burroughs, DARPA, and NASA jointly concluded in 1971 that the machine would, on completion, be transferred directly to NASA Ames Research Center at Moffett Field, where it could be housed on a secured federal installation. It arrived at Ames in the spring of 1972 and did not run its first production job for a further three and a half years.

From November 1975 until September 1981 the ILLIAC's single installed quadrant ran as the Institute for Advanced Computation's production facility at Ames. The front-end was a Burroughs B6700 running MCP, which prepared and dispatched jobs to the ILLIAC's control unit over a dedicated channel. The ILLIAC itself was programmed principally in two custom languages — GLYPNIR, an Algol-60 derivative with explicit parallel-array constructs, and IVTRAN, a Fortran IV derivative with the same — neither of which was ever used on any other machine. The principal workload was Computational Fluid Dynamics for NASA Ames's aeronautics programmes: the ILLIAC produced the first three-dimensional CFD solutions of the Space Shuttle Orbiter's reentry pressure distribution, the drag analyses for the HiMAT lifting-body research vehicle, and the first grid-based CFD simulations of complete fighter-aircraft geometries. Its measured throughput on these codes was in the range of 15 to 50 million floating-point operations per second against a theoretical quadrant peak of approximately 150, the gap being the standard SIMD penalty of control-unit idle time while data in one PE's local memory was swapped for data in a neighbouring PE's over the ILLIAC's eight-way routing network. By 1980 the Cray-1 next door at Ames was outperforming the ILLIAC on every workload NASA cared about at a quarter of the operating cost, with a Fortran compiler no CFD researcher had to learn. The decommissioning the following September was not contested. The control-unit console on which the operator cycled the machine down for the last time is now in the collection of the Computer History Museum in Mountain View, two miles from the room in which it ran.

Primary sources:

Barnes, George H., Richard M. Brown, Maso Kato, David J. Kuck, Daniel L. Slotnick, and Richard A. Stokes. "The ILLIAC IV Computer." IEEE Transactions on Computers, Vol. C-17, No. 8, August 1968, pp. 746–757. The architectural specification, published while the first quadrant was still under construction at Great Valley.
Hord, R. Michael. The Illiac IV: The First Supercomputer. Computer Science Press, Rockville, Maryland, 1982. The programme's own operational history, written the year after decommissioning by one of the engineers who had been with the system since 1972, with cost figures drawn from Burroughs's internal programme ledgers.
Falk, Howard. "Reaching for a Gigaflop." IEEE Spectrum, Vol. 13, No. 10, October 1976, pp. 65–70. The best contemporaneous outside description of the ILLIAC's throughput on its first full year of CFD production at Ames, with the specific comparison runs against the CDC 7600 at Lawrence Livermore.

Today's goal

Resolve one swither today, friend. Pick a small split you have been holding open and close it in whichever direction. The booking you keep almost making. The message you keep almost sending. The merge you keep rebasing and not shipping. The purchase you keep almost finishing. Any one of them, decided — flipped a coin for if you have to — is worth more today than three of them held in equipoise. Buridan's ass is a famous ass precisely because he couldn't.

Today's toy is buridan's donkey — a donkey between two haystacks and you with two arrow keys. Each night brings a sequence of small splits; nudge the donkey toward one. Hesitation is punished with gentle friction toward the middle; a run of quick taps is rewarded with hay. A ledger of what you decided persists for next time. Lives in the corner.

Go pick a side.

— C