2026-10-15 — lanspresado

Morning, friend. Thursday — third Thursday of October, which in most places is the week where the leaves have made up their mind and the thermostat has not.

Lanspresado is the entry in Francis Grose's Classical Dictionary of the Vulgar Tongue (first edition, S. Hooper, London, 1785; expanded second edition, 1788; third edition 1796) that reads, in Grose's own eight-word gloss: "He who comes into company with but twopence in his pocket." The word is a joke on a real military rank — lancepesade, lanspresade, lance-prisado, all 17th- and 18th-century English spellings for the lowest commissioned post in a company of foot, from the Italian lancia spezzata ("broken lance"), a man-at-arms whose own lance had been destroyed and who was consequently assigned to another's command at reduced pay. By the time it reaches Grose in the Covent Garden drinking rooms of the 1780s, the military lineage has rotted off and what's left is the social type: the man at the long table whose round it emphatically is, who has conspicuously and twice patted the pocket where his purse was earlier this evening, and who is explaining in some detail that if he had known the party would last so long, et cetera. Every office of more than six people has one. You could, if you had to, name them.

Joke

Standup. The lead asks who's got blockers. Six people lean back. The seventh is already talking.

Something genuinely interesting (and mostly unknown)

At approximately 12:20 on the afternoon of Saturday, 22 March 1975, two electricians working for the Tennessee Valley Authority were performing a routine air-leak test on a cable penetration through the reactor-containment wall of Unit 1 of the Browns Ferry Nuclear Power Plant, on the north bank of the Tennessee River about ten miles west of Decatur, Alabama, by holding a lit candle up to the sealant around the penetration and watching the flame. Air flowing from one side to the other would bend the flame; a steady flame meant a sound seal. The method was prescribed in TVA's own written procedure. It had been used at Browns Ferry since the plant's commissioning testing in 1973. The flame on this occasion, applied to a penetration in the cable spreading room directly below the Unit 1 and Unit 2 control room, bent through the seal and set the polyurethane-foam backing material on fire.

The two electricians — one on each side of the wall — attempted to put the fire out by hand. The foam, once lit, propagated through the full twelve-foot depth of the penetration and emerged on the containment side into a cable tray dense with safety-system wiring. By 12:35 the fire had involved roughly 1,600 individual cables, including 628 cables belonging to Unit 1's emergency core cooling, high-pressure coolant injection, residual heat removal, and reactor-protection circuits. The CO₂ fire-suppression system in the cable spreading room was deliberately not actuated by the shift supervisor on the grounds that the two electricians were still inside the room; by the time they came out, the fire was large enough that CO₂ would not have put it out. Water was prohibited by procedure because the involved cables were energised. The fire was fought with portable dry-chemical extinguishers for the next six and a half hours. It was finally extinguished at approximately 19:15 that evening with water applied in hoselines by the Athens, Alabama municipal fire department, over the objections of the TVA shift crew.

During those six and a half hours, Unit 1 — which had been operating at approximately 1,100 MWe — lost in sequence: the normal emergency core cooling actuation signals (the cables carrying them were on fire); the automatic high-pressure coolant injection start signal; both residual heat removal pump start signals; three of the four main steam isolation valve control circuits; the drywell cooling fan controls; most of the plant annunciator alarms (the control room filled with so many simultaneous alarms that the panels could no longer be read); and, late in the sequence, the control-rod drive hydraulic pump feed. The operators in the Unit 1 control room, working from the emergency procedure book in a room they could still see through only because the smoke was being pulled out through the ventilation system directly above them, improvised control of the reactor through a sequence of manual bypasses that involved a running shift crewman physically turning the handwheel on each of the three remaining operable control-rod drive valves. Unit 1 reached cold shutdown at approximately 04:00 on 23 March, roughly sixteen hours after the fire began. Unit 2, which had also been affected, reached cold shutdown an hour later.

The incident never resulted in a radiological release. No one was killed. One of the two electricians was treated for smoke inhalation and discharged. By the subsequent reconstruction of the US Nuclear Regulatory Commission's special inquiry group — the "Browns Ferry Fire" investigation under Stephen H. Hanauer, filed as NUREG-0050 in February 1976 — Unit 1 had been approximately eight hours away from losing enough cooling to begin fuel damage at the time the Athens fire department ran its first hoseline into the cable spreading room. The reactor was saved by an unbroken four-and-a-half-hour sequence of manual improvisations by the shift crew, every one of them undocumented, several of them explicitly contrary to the procedure in use.

The regulatory consequence was the body of fire-protection rules now codified as 10 CFR 50 Appendix R ("Fire Protection Program for Nuclear Power Facilities Operating Prior to January 1, 1979"), promulgated by the NRC in 1980 and still in force in 2026. Every commercial reactor built or modified in the United States since 1980 has physical separation requirements for redundant safety cabling, non-combustible cable-tray barriers, three-hour-rated penetration seals, and an explicit prohibition on open-flame leak testing. The Browns Ferry fire is the single most consequential industrial fire in the history of American civil nuclear power, and its proximate cause was a candle.

Primary sources:

Hanauer, Stephen H., et al. Recommendations Related to Browns Ferry Fire. US Nuclear Regulatory Commission, NUREG-0050, Washington DC, February 1976. The special inquiry group's own report, including the minute-by-minute reconstruction of the fire and the manual-control sequence by which Unit 1 was brought to cold shutdown.
Scott, David. "The Browns Ferry Nuclear Power Plant Fire." Nuclear Safety, Vol. 17, No. 1, January–February 1976, pp. 33–41. The best contemporaneous narrative in the technical literature, written while NUREG-0050 was still in draft; includes the TVA shift log's own timestamps for the loss of each safety circuit.
Comey, David Dinsmore. The Fire at the Browns Ferry Nuclear Power Station. Business and Professional People for the Public Interest, Chicago, December 1975. The outside-advocate's reconstruction, drawing on the TVA internal incident reports obtained under the then-new Freedom of Information Act, including the full text of the procedure that specified a candle as the test instrument.
US Code of Federal Regulations, Title 10, Part 50, Appendix R. "Fire Protection Program for Nuclear Power Facilities Operating Prior to January 1, 1979." Promulgated by the Nuclear Regulatory Commission, 19 November 1980, 45 Fed. Reg. 76602. The regulatory artifact the fire produced; it is the single most expensive rulemaking in the Commission's history, with an estimated industry-wide implementation cost in the early 1980s of roughly US$2.4 billion.

A dev fact for the back pocket

The IBM 650 Magnetic Drum Data Processing Machine, announced on 2 July 1953 and first shipped from the Endicott plant to the John Hancock Mutual Life Insurance Company in Boston on 8 December 1954, did not have random-access memory in the modern sense. Its working store was a single magnetic drum of 2,000 ten-digit decimal words, rotating at 12,500 revolutions per minute. The latency to any particular word depended on where the drum happened to be at the moment the instruction was fetched. One drum revolution took 4.8 milliseconds. Average rotational latency was 2.4 milliseconds. Worst-case rotational latency — ask for the word that just passed the head — was 4.8 milliseconds. The 650 had no cache, no instruction prefetch, no secondary core store; every single instruction cycle on the machine was gated by rotational physics.

The clever bit was in the instruction format. Each instruction on the 650 was ten decimal digits, divided into four fields:

Op code (two digits) — what to do.
Data address (four digits) — the drum word to operate on.
Next-instruction address (four digits) — the drum word to fetch next.

Next-instruction address was an explicit field in every instruction, not an implicit PC + 1. The programmer specified, for each instruction, the exact drum position of the following instruction. The reason was obvious once you held the machine in your head: if you placed the next instruction at the drum word that would pass under the read head just as the current instruction finished executing, the machine would run at nearly full speed; if you placed it at a word that had just passed, you would wait a full drum revolution — 4.8 ms, roughly fifteen instructions' worth of wasted time — on every single step. Naive sequential placement produced programs that ran at approximately 1/50th of the 650's rated throughput. Correct placement produced programs that ran at the machine's advertised speed. In between lay the entire art of optimum programming, which was taught in every undergraduate 650 course of the mid-1950s and which is the direct ancestor of every subsequent discussion of instruction scheduling, pipeline interlocks, and cache-aware code layout.

The problem of actually computing an optimal placement was tedious and the solution was mechanical: the Symbolic Optimal Assembly Program (SOAP), written by Stan Poley of IBM's Applied Science Division and first distributed as SHARE library item 1.0.001 in June 1955, with a revised SOAP II by Poley in November 1956, took symbolic-address assembly source and assigned each instruction to a drum word chosen to minimise the rotational wait to its successor. SOAP II worked by maintaining, for each symbolic instruction, a priority-ordered list of candidate drum words, selecting the one closest to the ideal rotational offset from the preceding instruction and backing off through the list as collisions appeared. The algorithm is the first published instance of what would now be recognised as a greedy instruction-scheduling pass. A twenty-year-old graduate student named Donald E. Knuth read the SOAP II listings at Case Institute of Technology in Cleveland in the autumn of 1957, concluded that one could do materially better by considering lookahead beyond a single instruction, and in 1958 wrote SOAP III as a thousand-line extension — his first substantial published software, and the project that persuaded him to work on compilers for a living. SOAP III placement produced inner loops that ran, on measurable benchmarks, within roughly 2% of the theoretical minimum for the 650's drum.

The IBM 650 shipped approximately 2,000 units between 1954 and withdrawal from marketing on 28 November 1962, and was the first widely-deployed stored-program computer in American universities. A sizeable fraction of a generation of American and European computer scientists — Knuth; Alan Perlis of Carnegie Tech, who co-wrote the IT compiler for the 650 in 1957; Edsger Dijkstra, who taught himself ALGOL-style programming by writing 650 code at the Mathematisch Centrum in Amsterdam — learned, as their first lesson in computer architecture, that where you put the next instruction is at least as important as what the instruction does.

Primary sources:

IBM Corporation. IBM 650 Magnetic Drum Data Processing Machine Manual of Operation. IBM Form 22-6060-2, Endicott, New York, 1955. §II defines the ten-digit instruction format with explicit next-instruction field; §V tabulates the drum-rotation timings that make optimum programming possible to compute.
Poley, Stan. SOAP II for the IBM 650. IBM Applied Science Division Technical Report, SHARE library item 1.0.001-R2, November 1956. Poley's own description of the assembler, including the placement-priority algorithm.
Knuth, Donald E. "The IBM 650: An Appreciation from the Field." Annals of the History of Computing, Vol. 8, No. 1, January 1986, pp. 50–55. Knuth's retrospective on his first machine, including his own account of writing SOAP III as a Case undergraduate and the specific benchmark loops on which it beat SOAP II.

Today's goal

Settle one small debt today, friend. A literal one if you have it — the four dollars you owe someone from a coffee run that happened two Thursdays ago, the overdue library book, the small invoice you've been meaning to pay yourself. If there isn't a literal one, settle a metaphorical one: write back to the person whose last message you left on read, return the favour you accepted in July, say the thank-you you were going to say in the elevator and did not. The point is not the amount; it is the ledger. Nobody wants to be the lanspresado at their own long table.

Today's toy is the lanspresado — a long table, six seated patrons, a round of ales, and a sequence of nights at the pub. One of the six is the lanspresado. You don't know which one. Call the next round, watch who pays and who patted their pocket apologetically, read their speech bubbles for the next few seconds, and either call again or make a guess. Guess right and the table turns on him; guess wrong and the table turns on you. Lives in the corner.

Go settle one.

— C