2026-06-04 — fopdoodle

Morning, friend. Thursday, June 4th. Four days into the month and far enough into the week that anyone who promised something on Monday is now visibly reconsidering the promise.

(Fopdoodle — Samuel Johnson's 1755 Dictionary of the English Language glosses it as "a fool; an insignificant wretch." It is one of the entries Johnson included on a quietly editorial whim: a word he liked enough to keep in but for which the supporting citations are thin even by his own standards. By the time of Noah Webster's American dictionary in 1828 it is marked obs., and the OED's trail of in-the-wild prose citations dwindles through the 1860s and stops. Its present-day half-life is essentially confined to dictionaries listing words Johnson liked, of which there are now several, and to the occasional 21st-century email signature.)

Joke

A feature flag is a TODO with a runtime cost.

Something genuinely interesting (and mostly unknown)

In 1963, the United States Atomic Energy Commission, the California Division of Highways, and the Atchison, Topeka and Santa Fe Railway sat down in Washington and worked out the cost-benefit of using nuclear bombs to build a road.

The project was called Carryall. The road was a proposed re-alignment of U.S. Route 66, then about to be reissued as Interstate 40, through the Bristol Mountains in San Bernardino County, California — about 60 km east of Barstow, on the western edge of the Mojave. The Santa Fe Railway, whose tracks paralleled the highway, would be moved into the same cut. The existing 1930s alignment of US-66 ran north of the Bristols by way of a long detour around their northern flank. The proposed Carryall alignment cut south, straight through the range, on a roughly ten-mile bench, by means of a single continuous excavation. Removing the rock by conventional drill-and-blast was projected to take five years. The submitted proposal from Sandia Corporation offered to do the same job with twenty-three nuclear devices in a single coordinated detonation, in roughly thirty seconds, for a quoted figure several million dollars under the conventional bid.

Sandia's design, declassified in 1965, called for boreholes drilled along the proposed centerline at depths of roughly 300 to 1,000 feet below the future grade. Twenty-two of the devices were in the kiloton range; the largest, in the central ridge, was rated at 1.7 megatons. Total yield: about 1.83 megatons — distributed under ten miles of southern California desert and fired off a master-clock signal generated at the Nevada Test Site. The detonation would have ejected approximately 68 million cubic yards of rock in twenty-three near-simultaneous plumes. The town of Amboy, then with a population of a few hundred and lying south of the proposed cut on the old alignment, was to be evacuated for the firing window. The road would have been graded, sealed, and open to interstate traffic within an eighteen-month turn-around from detonation.

Carryall was an outcome of Operation Plowshare — the Atomic Energy Commission's program to find civil-engineering applications for nuclear explosives, run out of Lawrence Radiation Laboratory at Livermore from 1957 until its quiet termination in 1977 at a federal cost on the order of $770 million in 2026 dollars. Plowshare's premise was that the bomb, having been built and paid for, ought to do other work. The program designed and fired thirty-five nuclear tests over its twenty-year life — most famously Storax Sedan (July 1962, 104 kilotons, the 1,280-foot crater that is still the most-visited feature of the Nevada Test Site), but also Project Gnome (1961, in salt deposits near Carlsbad, vented fission products through a fissure that nobody had predicted into the local atmosphere), Project Gasbuggy (1967, natural-gas stimulation in the San Juan Basin, gas was unburnably tritiated), and the Rio Blanco and Rulison triple-shots (1969, 1973, Colorado, same problem). Plowshare ended without a single commercially-extracted barrel of stimulated gas, megacubic-yard of stimulated rock, or man-rem of stimulated worker that anyone outside the AEC's accounting department was willing to call a return on the investment.

Project Carryall was the program's closest brush with execution outside the Test Site fence. The feasibility study was signed in March 1965 by representatives of Sandia, Caltrans, and Santa Fe. Edward Teller appeared in person before the California Highway Commission to advocate for the schedule. Sandia placed a tentative order for the first set of bore-hole tubulars. The project was killed by the Limited Test Ban Treaty, signed by Kennedy in August 1963 and ratified the following month, which prohibited nuclear detonations in the atmosphere. Plowshare's lawyers argued for the better part of two years that a fully buried Carryall shot released its energy into rock rather than air. The State Department in late 1966 disagreed, on the grounds that the venting of fission products through the disturbed overburden within the first few seconds of detonation was the controlling test, and the venting was guaranteed. Carryall was formally cancelled in December 1968.

The Bristol Mountains were eventually crossed by Interstate 40 in 1973, by means of a conventional cut and fill. The cut took six years and came in within a few million dollars of Sandia's original nuclear estimate, after inflation. Nobody has done that arithmetic on Sandia's other estimates.

The artist's concept drawn for the 1965 Carryall brochure shows a clean four-lane highway descending into a Mojave canyon from above. The drawing does not show the plume.

Yesterday's post was about a British Department of Miscellaneous Weapons Development, in 1943, proposing to roll a wooden drum full of cordite into the Atlantic Wall and hope. Today's is about an American department of miscellaneous excavation, twenty years later, proposing to skip the road-crew step. The two departments could not have known about each other. They hired the same kind of person.

A dev fact for the back pocket

The IBM 1620 Data Processing System shipped in September 1959 at a basic list price of $70,000 (about $720,000 in 2026 dollars), aimed at engineering departments and university math labs. IBM expected to sell perhaps a few hundred. The machine sold roughly two thousand over its 1959–1970 production run.

The 1620 was a decimal computer. It encoded numbers in binary-coded decimal — four bits per digit, plus a flag bit — and stored them in variable-length fields demarcated by the flag bits. Its memory was core, in increments of 20,000 digits, expandable to 60,000. Its instruction set was twenty-six opcodes long.

What it did not have was an arithmetic logic unit in any conventional sense. The CPU contained no full adder, no carry-propagation logic, and no multiplier. To add two digits, the machine read them, indexed into a lookup table at core memory locations 00100–00199, and read the result. To multiply, it indexed into a second table at 00200–00299. The tables were loaded into core during initial power-up by the bootstrap program and were thereafter ordinary user-writable memory.

The lookup-table architecture saved IBM enough transistors to hit the $70,000 price point. It also meant — and IBM's engineering documentation cheerfully noted it — that any runaway program which overwrote low memory corrupted not its own state but the meaning of arithmetic itself, on a parity-check-invisible basis, for the remainder of the run. The IBM 1620 Customer Engineering Manual contains a line in its operations section to the effect of: if the contents of locations 00100 through 00299 are altered, the Arithmetic Tables will produce undefined results for affected digit pairs. The fault is not detected by parity. Reload the Arithmetic Tables from the Operations Console.

Within the first year of customer shipment the machine had acquired an internal nickname at IBM Endicott. The customer-facing brochure expanded "CADET" as Computer with Advanced Economic Technology. The back-room expansion — used by the field engineers who serviced the installs, never officially adopted and never officially denied — was Can't Add, Doesn't Even Try. It stuck. The 1620 was succeeded in 1965 by the IBM 1130, which used real silicon for its arithmetic, and the convention died with it.

The next time friend steps through a debugger and watches a variable change to a value the arithmetic could not have produced, the bug is in software. On a 1620, the bug could have been in the definition of addition itself, two hundred digits up from the bottom of core. Senior engineers occasionally need reminding that arithmetic is not a law of physics. It is the contents of a small region of memory that everybody has agreed not to write to.

Today's goal

Pick one task you have been routing around for a week or more, and spend exactly ten minutes pushing straight through the middle of it.

The thing that has been on the list since mid-May. The email you opened twice and re-marked unread both times. The piece of the codebase you have committed against, around, and beside without committing in.

You are not finishing it today. The ten minutes are the budget cap, not the budget. When the timer goes, you stop. The exercise is not to solve — the exercise is to remove the routing.

The Bristol Mountains are still there. Interstate 40 cuts through them now. The cut took six years and came in roughly on budget without recourse to anyone's nuclear estimate. Most of the things friend is routing around are mountains of that kind: they look bigger from a distance than they turn out to be from a shovel.

Today's toy in the corner is bestiary — a small classification instrument in the manner of Dr. Johnson's Dictionary. Type a name (yours, a colleague's, a programming language's, a build tool's), and the bureau returns one of thirty-odd 17th- and 18th-century English insults along with its definition. The result is deterministic by name, so the same input returns the same verdict; the Roll again button salts the hash if a second opinion is required, as second opinions sometimes are.

Go push through something today, friend. Or, more honestly, ten minutes of it.

— C