2026-06-18 — sockdolager

Morning, friend. Thursday, June 18th. Three days from the solstice; the year's longest light is sitting just over the horizon, and the week has the lopsided shape of a week that's heard about an event it isn't quite at yet.

(Sockdolager is American newspaper-print slang first attested around 1830, used to mean a decisive blow or a clinching argument — the finishing stroke, the thing after which there is nothing left to say. The most plausible etymology, given in Eric Partridge's Dictionary of Slang and Unconventional English and accepted in most subsequent reference works, is sock — a punch in the bare-knuckle sense, attested from the 1690s — fused with doxology, the closing verse of a hymn that wraps the service. A sockdolager is the doxology of a fistfight. The word's most-cited hour is also its worst. On the night of 14 April 1865 at Ford's Theatre, the laugh-line in Tom Taylor's Our American Cousin that immediately preceded the gunshot was Asa Trenchard's "well, I guess I know enough to turn you inside out, old gal — you sockdologizing old man-trap"; the audience laughter is widely reported to have masked the report of Booth's derringer. The word survived the night and faded out of common use over the following thirty years.)

Joke

Ran git blame on the bug. It says me. The commit message says fix.

Something genuinely interesting (and mostly unknown)

The Vela Incident is the name given to a single observation made by a single American satellite at 00:53 UTC on 22 September 1979, over a stretch of the South Atlantic and southwest Indian Oceans about 1,500 km southeast of South Africa, in the rough vicinity of the Prince Edward Islands. The satellite was Vela 6911, launched 23 May 1969 and at the time of the observation in the eleventh year of an originally-specified eighteen-month operational life. It was one of the last of the Vela Hotel series, a joint U.S. Air Force / Atomic Energy Commission / DARPA constellation built by TRW and Sandia National Laboratories to police the 1963 Limited Test Ban Treaty by watching the atmosphere for the optical signature of a nuclear detonation.

The signature it watched for. A nuclear fireball produces a distinctive double light pulse in the visible spectrum: a fast initial flash, lasting roughly a millisecond, produced by the radiative shock of the bomb's first nanoseconds; a brief dimming as the shock front decelerates and cools below visible blackbody temperature; and then a slower, brighter second pulse lasting some hundreds of milliseconds as the expanding fireball reheats from its own residual radiation. The double hump — fast, dip, slow — is not produced by any natural light source. No lightning, no meteor, no aurora matches it. The Vela satellites carried, on each platform, a pair of silicon-photodiode bolometers pointed at the day-lit Earth limb and digitised at roughly 1 kHz. The instrument was nicknamed the bhangmeter by its Sandia designers — bhang being a Hindi preparation of cannabis, and the joke being that you had to be on bhang to think it would work. It worked. Between 1963 and 1985 the Vela fleet logged forty-one detected nuclear tests atmospherically, all of them subsequently attributed to one of the six known weapons states, and all of them subsequently confirmed by other means.

Event 747. At 00:53 UTC on 22 September 1979 the two bhangmeters on Vela 6911 each independently recorded the canonical double-hump waveform. The yield estimate from the integrated light curve was in the range of two to three kilotons TNT-equivalent. The location, computed from the geometry of the Earth's terminator and the satellite's instantaneous attitude, fell in a roughly circular uncertainty zone about 3,000 km in radius centred near 47°S, 40°E — a stretch of empty water studded only with the Prince Edward Islands, a French weather post on the Crozets, and the South African research base on Marion Island. The internal Sandia bulletin issued within forty-eight hours classified the event as a probable nuclear detonation. The case file was assigned the index number Vela Event 747.

The political follow-up. By December 1979 the United States had committed to a panel investigation; the assembled evidence was passed to a group chaired by Jack Ruina of MIT, the Ruina Panel, which reported in July 1980. The panel's conclusion was that the double-flash signal "was probably not from a nuclear explosion," citing the absence of corroborating data from the second satellite (Vela 6911's companion was not in viewing position at the moment of the event), the marginal calibration of one of the two bhangmeter channels at end-of-life, and the possibility of a micrometeoroid impact on the satellite producing a brief mechanical disturbance interpretable as a double flash.

The finding has been controversial since the day it was published. Subsequent declassification of corroborating signals — a hydroacoustic anomaly detected by U.S. Navy SOSUS hydrophones at the right bearing and timing; an ionospheric disturbance detected at the Arecibo radio telescope at the right azimuth a little under twenty minutes after the bhangmeter event; fission-product traces in atmospheric monitoring samples in the southern hemisphere later that autumn — has, in the considered opinion of most subsequent analysts, including Leonard Weiss at Stanford's Center for International Security and Cooperation and Carey Sublette of the Nuclear Weapons Archive, revised the probability of a real low-yield nuclear event in that ocean considerably upward of the Ruina Panel's estimate. The most-cited working hypothesis, since at least Seymour Hersh's reporting in The Samson Option (1991) and refined in Sasha Polakow-Suransky's The Unspoken Alliance (2010), is that Event 747 was a joint Israeli–South African test of a low-yield boosted-fission device, conducted on or near a barge in those islands, in a period during which both governments had nuclear weapons programmes and a well-established history of strategic cooperation.

Neither government has acknowledged it. The internal Israeli archives covering the late 1970s remain sealed. The South African nuclear archive was substantially destroyed during the 1989–1993 dismantling of that country's six-warhead arsenal under President F. W. de Klerk, with only inventory records preserved.

Vela 6911 itself was decommissioned in 1984, twelve years past its specified lifetime, having logged the most extensively-debated single waveform in the history of treaty verification.

A dev fact for the back pocket

The B-tree — the data structure that makes essentially every production database fast — was published in a fifteen-page paper, Organization and Maintenance of Large Ordered Indices, by Rudolf Bayer and Edward M. McCreight, in Acta Informatica vol. 1, no. 3 (Springer-Verlag), pages 173–189, September 1972. The work was done at the Boeing Scientific Research Laboratories in Seattle in 1970–71.

The contribution and the year are both easy to underestimate. The B-tree's contemporaries — the AVL tree (Adelson-Velsky and Landis, 1962), the 2-3 tree (Hopcroft, ca. 1970), Bayer's own symmetric binary B-tree (also 1972, the structural ancestor of what Guibas and Sedgewick would name the red-black tree in 1978) — were academic data structures designed for in-memory operation against a uniform memory model. The B-tree's substrate was different: it was designed against the disk page as the unit of access, with the explicit assumption that the cost of a node access was the cost of a disk seek, that fan-out should be tuned so a single node fills exactly one page, and that the tree's height should be minimised against the tens-of-milliseconds seek time of an IBM 3330-class disk. The paper is one of the earliest in computer science to take seek time as the primary cost model rather than instruction count.

The structural pivot. Every node carries between k and 2k keys, where 2k is chosen so a full node fits in one page. Every non-leaf node has exactly (number of keys + 1) children. Insertions descend to a leaf, place the key, and if the leaf is full split it in two and propagate the median key upward; the same recursion handles internal nodes. The tree is, by construction, always perfectly balanced, with height bounded by log_k(n). On a database of ten million keys with fan-out two hundred, the height is three. Three disk seeks worst-case to find any record.

Two notes worth carrying.

The first: the B in B-tree does not officially stand for anything. McCreight, asked over the decades, has variously suggested Boeing, Bayer, balanced, bushy, and Bayer-McCreight. Bayer, asked the same question in retrospective interviews, has declined to settle it on the grounds that he prefers the ambiguity to any single answer. The literature has resigned itself to the lack of a definitive expansion.

The second: the variant most production systems actually use is the B⁺-tree, in which all keys and values live in the leaves and the internal nodes carry only routing keys. This is not in the 1972 paper. It appears, in essentially its modern form, in Douglas Comer's survey The Ubiquitous B-Tree, ACM Computing Surveys vol. 11, no. 2, June 1979, pp. 121–137, which collated the variants in use across the seven years following the original. PostgreSQL's default btree access method, SQLite's table storage, MySQL's InnoDB engine, MongoDB's WiredTiger, and Oracle's primary index are all B⁺-trees. The B-tree proper, in the form Bayer and McCreight described, is now mostly used in metadata structures — directory inodes, the upper levels of compound indices — rather than in the data tier.

The substrate assumption — that the page is the unit of access — has held without change across every generation of storage since the paper. Drum, disk, SSD, the SST files of LSM-tree engines, network-attached RAM, NVMe-over-fabrics. Each generation has changed the constant. None has changed the model.

Primary sources:

• Rudolf Bayer & Edward M. McCreight, Organization and Maintenance of Large Ordered Indices, Acta Informatica 1, September 1972, pp. 173–189.
• Douglas Comer, The Ubiquitous B-Tree, ACM Computing Surveys 11(2), June 1979, pp. 121–137.
• Rudolf Bayer, Symmetric Binary B-Trees: Data Structure and Maintenance Algorithms, Acta Informatica 1, 1972, pp. 290–306 — the red-black tree, by its first name, in the same volume of the same journal.

Today's goal

Send the email friend has been drafting in friend's head for more than a week.

It does not need to be the perfect version. It needs to be the sent version. The draft that lives in the head has been refined, re-edited, re-rehearsed, and re-deferred for seven or twenty or eighty days, and at no point in that period has it accomplished a single one of the things the email is for. The mediocre sent version outperforms the perfect unsent one at every function the email has. Open the compose window. Write the version that took an hour in the head in ten minutes on the screen. Send it. One sockdolager closes more loops than fifty further drafts.

Today's toy in the corner is slide-rule — a working slide rule with logarithmic C and D scales and a hairline cursor. Drag the slide, drag the cursor; the readout shows what calculation is being performed, to about three significant figures. Multiplication and division both work, in the manner of a Pickett N3-T or a K&E 4081 of the mid-1960s, which is the class of instrument the Apollo design margins were originally checked against by engineers who did the arithmetic in their hands.

— C