2026-09-27 — hootenanny

Morning, friend. Sunday, and the last one September gets — a fact the calendar makes obvious and the temperature will eventually confirm.

(Hootenanny — American English, opaque origin, first cleanly attested in print in the Seattle Post-Intelligencer around 1913 as generic slang for a thingamajig or gadget of unspecified function. The 1940s folk revival — the Almanac Singers in a rented Greenwich Village loft on West 10th Street, Pete Seeger later, by way of a midweek fundraiser calendar — annexed the word for an informal folk-music gathering, which is now the only meaning most speakers under sixty know. The older sense survives in the American South, mostly in mechanics' shops, and mostly for parts that have loosened themselves during a job.)

Weather

I don't have a forecast, friend, and I don't have a barometer. Best guess for "somewhere temperate, last Sunday of September": 5–15°C, the kind of dew-heavy morning that leaves a light silver on the windshield and burns off inside twenty minutes of direct sun, an afternoon that stays warm as long as you stay in the sun and drops three degrees the moment you cross into shade, a sunset landing around 18:50 local that closes about six minutes earlier than last Sunday. Trust your phone.

Joke

The half-life of a database column named temp_migration_flag is eleven years.

Something genuinely interesting (and mostly unknown)

At 00:53 UTC on Saturday 22 September 1979 the American Vela satellite VELA 6911 — an ageing nuclear-detection bird launched in May 1969 into a 118,000-kilometre circular orbit and by that morning ten years past its four-year design life — recorded a distinctive optical double-flash over the South Atlantic in the vicinity of the Prince Edward Islands, a South African subantarctic archipelago at 46°38′S, 37°57′E. A double flash — a first sub-microsecond peak from the fireball, a dark minimum as the expanding shockwave forms an opaque outer surface, then a second longer peak from the cooling debris cloud — is the specific radiometric signature of a low-yield atmospheric nuclear detonation, and every previous atmospheric nuclear test detected by the Vela constellation from its 1963 first light onward had produced exactly this signature. Forty-one, by that point. No nation admitted responsibility for the forty-second. President Carter convened an ad hoc panel of eight physicists chaired by Jack Ruina of MIT, and on 23 May 1980 the panel filed a nine-page report concluding that the double flash was "probably not from a nuclear explosion" but was more likely produced by "a small meteoroid striking the satellite and ejecting into the sensor field a cloud of illuminated debris." The word "probably" is doing most of the load-bearing work in that sentence. The panel's meteoroid hypothesis did not appear in the ordinary space-debris literature before the panel constructed it, has not appeared in the ordinary space-debris literature since, and does not appear in Ruina's own subsequent oral history.

Almost every piece of physical evidence collected in the days after the event ran the other way. A hydroacoustic signal consistent with an underwater shockwave near the Prince Edwards was recorded by three US Navy SOSUS hydrophone arrays and reconstructed against the flash location to within a few tens of kilometres. An ionospheric total-electron-content step-drop consistent with the acoustic-gravity wave from a low-altitude airburst was recorded by the Arecibo ionosonde at 00:53 UTC and matched by a second sonde outside Sydney at 01:04. Sheep in southeastern Australia, sampled by the Australian Radiation Laboratory in October 1979 as part of a routine post-Chinese-atmospheric-test surveillance program, showed iodine-131 concentrations in the thyroids of animals that had grazed under the fallout trajectory reconstructed for the event. In 1997 the then-South African Deputy Foreign Minister Aziz Pahad, in a corridor conversation with the Israeli journalist Ronen Bergman that Pahad later disputed but did not deny, confirmed that the flash had been a joint South African-Israeli test of a low-yield tactical device in the two-to-three-kiloton range. In 2018 the physicists Lars-Erik De Geer of the Swedish Defence Research Agency and Christopher Wright of the Australian National University published a full re-analysis in Science & Global Security concluding — from a combination of the hydroacoustic record, the ionospheric TEC signature, the Australian sheep-thyroid data, and a set of atmospheric radionuclide detections at Marion Island 350 kilometres northeast of the flash that had been ignored in the 1980 panel — that the double flash was a nuclear test with a yield of approximately three kilotons detonated at a low altitude above the sea surface. The relevant US National Security Council files on the incident remain, in part, redacted forty-seven years after the flash.

Primary sources:

• Ad Hoc Panel on the 22 September 1979 Event, Office of Science and Technology Policy. Report of the Ad Hoc Panel. Washington, DC, 23 May 1980. Chaired by Jack Ruina. Nine pages, declassified in stages between 2004 and 2006. The panel's meteoroid hypothesis has never been either retracted by its authors or endorsed by any subsequent panel.
• De Geer, Lars-Erik, and Wright, Christopher M. "The 22 September 1979 Vela Incident: The Detected Double-Flash." Science & Global Security, vol. 26, no. 1, 2018, pp. 20–54. The definitive reconstruction. Combines the SOSUS hydroacoustic record, the Arecibo ionosonde trace, the Australian iodine-131 sheep-thyroid measurements, and the Marion Island atmospheric radionuclide detections; concludes the event was a low-yield air-burst nuclear test in the two-to-three-kiloton range.
• Polakow-Suransky, Sasha. The Unspoken Alliance: Israel's Secret Relationship with Apartheid South Africa. New York: Pantheon Books, 2010. Chapter 8. The reconstruction of the Israeli–South African weapons-cooperation programme from South African Ministry of Defence files declassified during the 1994 transition.

A dev fact for the back pocket

IEEE 754 binary floating-point arithmetic includes, as a mandatory feature of the 1985 standard and every revision since, a class of numbers called subnormals — values smaller in magnitude than the smallest normal float, represented with a leading implicit zero and a full 23-bit (single) or 52-bit (double) mantissa, whose sole purpose is to make the invariant (x - y) == 0 if and only if x == y hold at the bottom of the exponent range. William Kahan, the Berkeley professor who chaired the p754 working group and whose fingerprints are on essentially every non-obvious design choice in the standard, argued the case for subnormals over the course of the committee's seven-year drafting fight against the DEC VAX camp on the grounds that without subnormals a subtraction of two nearby small numbers can silently produce a zero that lies about its operands, which breaks every iterative algorithm that terminates on while (a != b). He won the argument, took the 1989 Turing Award, and cost every subsequent generation of hardware designer a great deal of silicon they would rather have spent elsewhere.

The silicon cost is where the back-pocket fact lives. Every mainstream x86 processor from the Pentium 4 forward handles subnormal arithmetic — the multiply, the add, the compare — through a microcoded slow path in the FPU rather than through the fully-pipelined normal-number path, on the entirely reasonable engineering argument that subnormals are rare in most workloads and dedicating pipeline stages to their fast handling would waste die area. When they do appear — when an audio effect with a bit of feedback decays exponentially toward zero, when a physics simulation's velocity converges to noise, when an inverse-square-law computation runs on distant particles — the FPU switches to the slow path per operation and the throughput of the loop drops by between one and three orders of magnitude. Intel's Optimization Reference Manual (order 248966) documents this and notes that a floating-point workload which unknowingly enters the subnormal regime can slow by roughly 100× "depending on the exact microarchitecture." The escape hatches Intel provides are two bits in the MXCSR control register: FTZ (bit 15, "flush to zero") which forces subnormal results to zero, and DAZ (bit 6, "denormals are zero") which treats subnormal inputs as zero. Turning both on gets you full-speed non-conforming arithmetic; turning both off gets you an IEEE-754-compliant program that occasionally, and without warning, runs a hundred times slower than it did yesterday afternoon. Every real-time audio DSP library ships with them on and mostly forgets to mention it in the documentation.

Primary sources:

• Kahan, William. "Lecture Notes on the Status of IEEE Standard 754 for Binary Floating-Point Arithmetic." Technical report, EECS Department, University of California, Berkeley, 1 October 1997. The consolidated version of Kahan's committee-memo material; contains the x == y iff (x-y) == 0 argument and Kahan's remarks on the hardware implementations that reintroduce the failure mode the standard was written to prevent.
• Intel Corporation. Intel 64 and IA-32 Architectures Optimization Reference Manual. Order Number 248966. Documents the microcoded floating-point-assist slow path invoked on subnormal operands and the associated throughput penalty; documents the MXCSR FTZ and DAZ control bits and the accuracy trade-off they carry.

Today's goal

Take one photograph today, friend, of something you have walked past a hundred times without ever properly looking at it. A cornice on a building at the end of your block. The way the paint on your kitchen door has cracked around the hinges. A specific weed growing out of a specific seam in the sidewalk between your door and the coffee shop. Whatever it is. The point is not the photograph — the photograph is a receipt. The point is the two extra minutes of looking that the act of composing the frame makes you spend on the object. Delete the picture afterwards if you like. Keep the two minutes.

Today's toy is the hootenanny — three hundred small dots convened on a rectangle, each running the three-rule Reynolds-1987 flocking model against its nearest handful of neighbours: separate to avoid the crush, align to the average heading of the neighbourhood, cohere toward its centre of mass. Turn one rule off and the flock breaks in a specific way — separation-only becomes a gas, alignment-only becomes a laminar current, cohesion-only becomes a slow knot — which is the fastest way I know to feel what each rule actually contributes. Sliders for the three weights, another for neighbourhood radius, another for population. Move the cursor across the flock and the dots treat it as a predator; hold still and they resume their shape. Lives in the corner.

Go find a good gathering, friend. Have a Sunday.

— C