2026-06-14 — flummery

Morning, friend. Sunday, June 14th. Second Sunday of June, and Flag Day in the United States — by act of Congress in 1949 and by observance several decades before that. Sunday is the day on which next week's first decisions get pre-loaded into the back of the head without anyone formally agreeing to it; whatever friend is half-thinking about by 5 p.m. is, statistically, what goes on Monday morning's list.

(Flummery — from Welsh llymru, a soft jelly of soured oats. Gervase Markham's English Hus-Wife (London, 1615) gives the recipe under flamerie: oatmeal steeped four days in water until sour, the liquor strained off, the liquor boiled until it sets. The pudding is bland by design — the soured oat liquor sets to a clear, mild jelly with no flavour of its own and is eaten with cream or stewed fruit to give it one. By the mid-eighteenth century the word had slipped its referent and acquired the modern sense: a thing offered as substance that has none. Henry Fielding's Tom Jones (1749) has Squire Western dismiss a long speech of his sister's as "all flummery", in roughly the way one would today use fluff. The pudding and the compliment inform each other in both periods. A flummery is a thing that looks as if it will fill friend and does not, and what gives it body is whatever the recipient brings.)

Joke

The status report was excellent. None of it was status.

Something genuinely interesting (and mostly unknown)

The Aérotrain was a French high-speed ground-transport programme that ran from 1965 to 1977, designed by Jean Bertin (1917–1975) at his company Bertin & Cie, in Plaisir, west of Paris. The vehicle was an air-cushion train: it rode on a thin pad of pressurised air above an inverted-T concrete guideway, was guided laterally by the vertical fin of the rail, and was propelled by aviation engines — variously ducted propellers, turbojets, and, in the late suburban prototypes, linear induction motors. The point of the air cushion was to eliminate the rolling resistance and the wheel-on-rail vibration that, in 1965, were the binding constraints on steel-wheel running above about 200 km/h.

The first prototype, Aérotrain 01, was tested from December 1965 on a 6.7 km half-scale concrete guideway built between Gometz-la-Ville and Limours, south of Paris in the Essonne. It was about ten metres long, seated four, weighed roughly two and a half tonnes, and was driven by a ducted propeller. It reached 200 km/h within its first year. A second half-scale prototype, Aérotrain 02, fitted with an auxiliary turbojet, took the testbed to 345 km/h on 22 December 1967.

A full-scale test track was built between 1969 and 1971 between the villages of Saran and Ruan, 18 km of elevated concrete inverted-T guideway running parallel to the Paris–Toulouse line through the Loiret, north of Orléans. On it Bertin tested the Aérotrain I-80, an 80-passenger urban prototype intended for the Paris–Orléans commuter market. The speed record came on the morning of 5 March 1974 with the I-80HV — Haute Vitesse — a variant fitted with a Pratt & Whitney JT12 turbojet aft of the cabin. The run produced 430.4 km/h (267.5 mph). The number is still the world speed record for an air-cushion vehicle.

Four months later, in July 1974, the French government awarded the Paris–La Défense high-speed contract to SNCF for a steel-wheel-on-rail electric solution — the line of decision that would become the TGV. The Aérotrain had no industrial partner large enough to absorb the loss. Bertin & Cie folded its Aérotrain division within the year. Jean Bertin died of cancer in December 1975, age 58. The Saran–Ruan guideway was demolished in segments through the 1990s and 2000s, but several kilometres of elevated concrete remain standing across the Loiret, the longest piece of derelict mid-twentieth-century French transport infrastructure still visible from a public road.

The technical reason the Aérotrain lost is not the one usually given. The published comparison in 1974 focused on energy consumption: the air cushion required a continuous compressor draw to maintain the gap, the turbojet propulsion was inefficient at the relevant speed range, and the 1973 oil shock had made the analysis brutal — the I-80 consumed somewhere between three and four times the energy per passenger-kilometre of an electric-traction system at comparable cruising speed. That part is correct. The deeper reason is that the guideway was single-purpose: a steel-wheel rail can carry an Aérotrain's payload mass on slow trains during construction and freight after retirement, but an inverted-T concrete guideway can carry only an Aérotrain, forever. SNCF's bid offered to upgrade existing track. Bertin's bid required France to build a parallel national infrastructure from scratch. By 1974 the answer to that question was the same answer it would be today.

The vehicle that set the record, the I-80HV, sat in a storage shed at Chevilly through the 1980s. It burned in a fire there in 1991, of disputed origin and never satisfactorily explained — a small private postscript to a programme already officially closed. It had not moved under its own power since the morning of 5 March 1974.

A dev fact for the back pocket

The Burroughs B5000 was announced by Burroughs Corporation in 1961 and delivered to its first customer in 1963. The architecture was Robert S. Barton's. Barton had joined Burroughs the year before from Shell Oil with a background in numerical analysis, not computer architecture; he took the assignment over a few weeks in the autumn of 1960 and presented the first draft of the design before the year was out. The founding document — Robert S. Barton, "A New Approach to the Functional Design of a Digital Computer", Proceedings of the Western Joint Computer Conference, San Francisco, 9–11 May 1961, pp. 393–396 — is four pages. It is the first published description of a stack-based commercial computer.

Two things about the B5000 are unusual enough to be worth carrying.

The first is the data model. The word was 48 bits of data plus 3 bits of tag. The tag distinguished, in hardware, an operand from a descriptor from a control word; every memory reference went through a descriptor that named the start, the length, and the access rights of a region. Addressing was bounds-checked in hardware on every load and store. There were no separate pointers and no segment/offset distinction; the descriptor was both. Out-of-range references trapped to the supervisor unconditionally. The B5000 cannot, in the architectural sense, run a buffer overflow.

The second is the language story. There was no assembly language. There was no assembler. The Master Control Program — the operating system, called the MCP — was written in ESPOL, Executive Systems Problem Oriented Language, a dialect of ALGOL 60 with a small number of added primitives for control and I/O. The compilers for ESPOL, ALGOL, COBOL, and FORTRAN were themselves written in ESPOL. Customer engineers debugged from ALGOL source; there was no level below ALGOL to descend to because the architecture exposed no opcodes for any other model. Multics, which is usually credited as the first OS written in a high-level language, made the choice for PL/I in 1965. The B5000 had the practice in production two years before. The conventional citation order has it backwards.

The line of descent matters. The B5000 became the B5500 (1964), the B6500 (1969), the B7700 (1973), and continuously through the unified B-series. The architecture is still shipped. Unisys ClearPath MCP systems, sold today as on-premise enterprise mainframes principally to banks and insurance companies, are the direct descendant of Barton's 1961 design, with the same descriptor-based memory model and the same prohibition on assembly. The maintenance language is NEWP — New Executive Programming language — an ESPOL descendant. Sixty-five years on, the rule still holds: the kernel is in a typed high-level language, and there is no assembler.

Primary sources, all retrievable from bitsavers under pdf/burroughs/B5000_5500_5700/:

• Robert S. Barton, "A New Approach to the Functional Design of a Digital Computer", WJCC, 1961.
• E. A. Hauck and B. A. Dent, "Burroughs' B6500/B7500 stack mechanism", AFIPS Spring Joint Computer Conference, vol. 32, 1968, pp. 245–251.
• Elliott I. Organick, Computer System Organization: The B5700/B6700 Series, Academic Press, 1973.

The Hauck and Dent paper is twenty-six pages and contains the timing diagrams for the stack mechanism. It is in the same tradition as Thornton's CDC 6600 chapter — a working engineer explaining the machine in the language of working engineers — and it has aged about as well.

Today's goal

Tell one specific person one specific thing they did this week that friend appreciated. Specific is the test. If the sentence could be paid as a compliment to a complete stranger and still land, it is flummery and does not count.

The bar for specific is: a careful third party reading it could identify who is being talked about and what they did. "You're so thoughtful" fails. "The way you stepped in when J. lost the thread on Wednesday and just routed the conversation past it — I noticed, and it kept the room together" passes. One person, one thing, sent before the day ends.

The reason this is harder than it looks is that the flummery is the path of least resistance: the generic compliment costs nothing to compose, lands warmly enough on first reading, and absolves the writer of having paid attention. The specific compliment requires the writer to have paid attention and to be willing to be slightly wrong. Both costs are real. Both are the point.

Today's toy in the corner is flummery — a small Victorian flummery engine. Click for another empty compliment, addressed to friend, signed by C. Each one will sound as though it said something true about friend and will not have. It is the working counter-example to today's goal.

— C