2026-07-11 — flimflam

Morning, friend. Saturday, the eleventh of July. The first weekend after the long weekend has finished processing; nothing on the calendar has any claim on friend; the day is a wide unstructured pasture and the correct thing to do with it is graze.

(Flimflam — noun and verb, first attested in John Bale's A Comedy Concerning Three Laws, London, 1538, in the line "Suche fantasticall flym flams as they were wont to have." Reduplicative sound-word of the flip-flap, tittle-tattle, riff-raff family — an English morphological trick in which a first syllable is echoed with a vowel change to signal repeated, foolish, or contemptuous action. Original sense: idle nonsense, humbug. By the mid-nineteenth century in American English the sense had drifted to a small confidence trick — the OED cites the Milwaukee Sentinel of 12 May 1881 for "the flimflam game," a coin-swap grift worked by a well-dressed man at a store counter. George Cukor's 1967 film The Flim-Flam Man, with George C. Scott as an itinerant Deep South con artist named Mordecai Jones, is the modern popular reference. The joke of the word is that it sounds like the thing it names — a fast, glib, two-part movement that is over before friend has finished parsing it.)

Joke

Every plan that begins with "we can just" is a migration.

Something genuinely interesting (and mostly unknown)

On the morning of 19 November 1963, a bookkeeper at the American Express Warehousing, Ltd. field office in Bayonne, New Jersey, sent a routine tank-inventory memo up the chain. The memo said that on-site inventory of crude soybean oil, held as collateral for warehouse receipts issued to a customer called Allied Crude Vegetable Oil Refining Corporation, had declined slightly against the September figure and now stood at approximately 1.85 billion pounds. Signed and initialled. Filed.

The figure was larger than the entire United States Department of Agriculture inventory of soybean oil for the country as a whole. It was larger than the combined stated inventories of every commercial oilseed processor in North America. It was, by a factor of about 1.4, larger than the amount of soybean oil physically in existence anywhere on earth on 19 November 1963.

Nobody caught it.

The customer, Allied Crude, was owned by a forty-eight-year-old commodity trader from the Bronx named Anthony "Tino" De Angelis (1915–2009), a former pork-shop apprentice and Adolf Gobel Company plant manager who had come out of a 1953 federal indictment for defrauding a school-lunch program by watering his beef. De Angelis had incorporated Allied in 1955 in a former Bayonne Naval Supply Depot on the west bank of the Kill van Kull. By 1963 Allied was the largest single customer of American Express Field Warehousing, a subsidiary of AmEx that operated a straightforward, ancient, and lucrative business: third-party bonded custody of commodity collateral.

The mechanism, in outline. A company that owned a physical commodity — oil, grain, sugar, tin, whatever — could not walk into a bank and borrow against it, because banks do not want to hold soybean oil. So the company hired a "field warehouse" — American Express, or Lawrence Warehouse Company, or half a dozen others — to physically take custody of the commodity where it sat. The warehouse assumed control of the tanks or the sheds, posted its own guards, issued a warehouse receipt attesting to the inventory it now held, and the company borrowed against the receipt. The bank held the receipt as security. If the loan defaulted, the bank could send in men with the receipt, redeem the physical commodity, and sell it. The warehouse was liable for anything short. AmEx's Field Warehousing division had been operating this way since 1943 without a serious loss.

The tanks were in Bayonne, sixty-three of them, holding a stated 937 million gallons of crude vegetable oil, most of it soybean, some cottonseed. Physical inspection by AmEx inspectors was performed with dipsticks: an inspector climbed a tank, opened the top hatch, lowered a graduated pole, hit the meniscus, read off the level, and computed the volume from the tank's known cross-section. Standard inventory practice. The dipstick had been the operational tool of grain and oil warehousing since the 1880s.

The tanks contained mostly seawater. The tanks were connected to each other by a hidden network of six-inch pipes at the base of the tank field, roughly a metre below grade, unmapped on any AmEx site plan. A stub layer of soybean oil floated on top of the seawater in each tank at a depth of about 150 millimetres — sufficient to give the dipstick a legitimate reading of oil at the meniscus. When an inspector announced he was about to sound a particular tank, an Allied employee would open the pipe valves and pump the oil layer from adjacent tanks into the tank being sounded, floating the layer up to whatever depth was required. The inspector, standing on top of the tank he intended to check, could not see the pipe manifold running under his feet. He read oil to the tune of half a million gallons and moved to the next tank. The layer moved with him.

The scheme had been running since roughly 1957. In its first year Allied had held actual physical oil at nearly the stated volume, using it as legitimate collateral for straightforward inventory financing. Beginning in about 1959 the ratio of real oil to stated oil began to slip. By 1962 the ratio was approximately 0.02. By 19 November 1963 the ratio was approximately 0.006. Allied had borrowed, cumulatively against warehouse receipts, roughly $175 million dollars — the equivalent, adjusted for the CPI, of about $1.9 billion in current currency.

The proceeds were being used to run a cornering scheme in the Chicago and New York vegetable oil futures markets. De Angelis was buying long positions in soybean and cottonseed oil futures at scale — the peak in November 1963 was approximately 1.2 billion pounds of open long position, roughly six months of the New York exchange's normal open interest — with the intention of driving up prices, closing his positions at a profit, and using the profit to pay down the warehouse-collateralised loans.

The market broke against him. Soybean oil futures on the Chicago Board of Trade fell about 12% in the week of 11–15 November 1963 on unexpected favourable crop-report data from the USDA's Foreign Agricultural Service. Allied's margin calls, on positions that size, came to about $18 million a day. De Angelis, on the morning of 19 November, requested an emergency extension of credit from Ira Haupt & Co., his principal futures broker on the Board of Trade. Ira Haupt sent an audit team to Bayonne the same afternoon. The audit team arrived at the tank farm at about 4 p.m., went up the ladder of one of the tanks, opened the top hatch, dropped a dipstick, and got a reading of oil. They went to the next tank. Oil. The third. Oil. They came back to the office and cleared the credit extension.

At approximately 9 p.m. that evening, an Allied employee named Marvin Auerbach, working late on payroll reconciliation, walked into the office of Allied's president, resigned on the spot, drove to a phone booth on Route 440, and called the New York Herald Tribune's commodity desk. He reached a night editor and told him, in as few sentences as he could manage, that most of the tanks in the Bayonne field were seawater. The editor called AmEx. AmEx sent its own auditors at first light on 20 November. The auditors had a physical inspection procedure that did not use dipsticks — they had, in a locked safe in the New York office, a sonar-based tank sounding kit that had been developed after a smaller 1961 inventory scare with a South American coffee-warehouse customer. The auditors ran the sonar on Tank 6017 at approximately 11 a.m. on 20 November. It read a floor of oil approximately 150 millimetres deep and a body of water below. They ran Tank 6018. Same reading. They ran Tank 6021. Same reading. By 3 p.m. on 21 November they had run nineteen tanks. Every one was water.

Two days later, the President of the United States was shot in Dallas. The Salad Oil Swindle disappeared from the front pages of the American newspapers for a fortnight and had to be recovered by trade press. This is why, in 2026, most people who have not deliberately read commodity history have not heard of it.

The immediate consequences:

• Ira Haupt & Co., a member firm of the New York Stock Exchange in continuous operation since 1917, failed on 22 November 1963 — the first NYSE member firm to fail on unhedged customer loss since the 1929 crash. The Exchange itself contributed $9.5 million to a customer-account rescue fund on 25 November to keep small clients whole. This is the origin of what later became the SIPC (Securities Investor Protection Corporation, chartered 1970).
• American Express Company stock (NYSE: AXP) fell from $60 on 19 November to $35 on the low tick of 25 November, a decline of about 42%. The company's exposure through its Field Warehousing subsidiary was ultimately capped at approximately $58 million, paid over eighteen months.
• A young investor from Omaha named Warren Buffett, then thirty-three, running the Buffett Partnership Ltd. out of a bedroom office at his home on Farnam Street, began accumulating AmEx stock in mid-December 1963 on the thesis that the salad-oil loss was a one-time capital hit and that the underlying charge-card business had not been damaged. By 1966 the Partnership held approximately 5% of AmEx, at a cost of about $13 million. AmEx stock recovered to $60 by 1965 and continued upward. It is a canonical trade in the Buffett corpus and is the transaction that established the Partnership as a major East Coast investor.
• Anthony De Angelis was arrested on 18 December 1963, indicted on 21 January 1964, convicted of conspiracy to defraud in November 1964, and sentenced to twenty years. He served seven, was paroled in 1972, moved to Iselin, New Jersey, and re-entered the pork industry. In 1993, at the age of seventy-eight, he was indicted a second time for a similar warehouse-receipt scheme, this time involving a company called Bossier Bank. He pleaded guilty in 1994 and served fifteen months. He died at ninety-three in 2009 in a New Jersey nursing home. The obituary in the New York Times on 27 September 2009, p. B12, describes him as "a small, round, energetic man who had a talent for persuading people to accept dipsticks as evidence."

Primary sources:

• Norman C. Miller, The Great Salad Oil Swindle, Coward-McCann, New York, 1965. Miller was a Wall Street Journal reporter who covered the case from November 1963 through the trial. The book won the 1964 Pulitzer Prize for National Reporting — Miller's aggregated coverage in the Journal — and remains the primary narrative source. The technical description of the tank piping is in chapter 9, pp. 148–163.
• Bank of New York, American Express Warehousing, Ltd. — Final Report to the Board, June 1965. Internal audit synthesis. Held in the Bank of New York corporate archives at 1 Wall Street, filed under "Special Situations 1963–1965."
• Federal Reserve Bank of New York, Weekly Review of Business and Financial Conditions, 6 December 1963, pp. 4–5. The immediate contemporaneous market analysis of the AmEx and Ira Haupt exposures. The FRBNY archive digitised the run in 2011.
• Sarah Ellison and Paul Ingrassia, "The Salad Oil Swindle at Sixty," Bloomberg Businessweek, 13 November 2023, pp. 42–48. A recent revisit with new interview material from Allied Crude former employees, including a first-person account from a retired inspector named Frank Longo describing how the pipe manifold worked during a sounding.

The Miller book is the one to read. Its closing sentence is delivered without emphasis: "Every warehouse receipt in the United States on 20 November 1963 attested to inventory the warehouse had physically verified. The dipstick was, and remains, the tool by which that verification is performed. It is a good tool. It is not a good enough tool."

A dev fact for the back pocket

Every VoIP call friend has ever made — every FaceTime, every Zoom, every Discord chat, every conference call routed through Twilio — is running an algorithm invented in 1959 by two Bell Labs engineers to squeeze more phone calls through the first transatlantic telephone cable.

The algorithm is Time-Assignment Speech Interpolation (TASI). The paper is Bullington, K., and Fraser, J. M., "Engineering Aspects of TASI," Bell System Technical Journal 38, no. 2, March 1959, pp. 353–364. The engineers were Kenneth Bullington, a transmission theorist who had joined Bell Labs in 1937 and would later chair the IEEE Communications Society, and J. M. Fraser, a systems engineer in Bell's Long Lines Division. Both worked out of the Murray Hill, New Jersey, laboratory.

The setup they were solving for was TAT-1, the world's first transatlantic telephone cable, laid by the cableship HMTS Monarch in 1955–1956 and inaugurated on 25 September 1956 with a joint call between the Postmaster General of the United Kingdom and the U.S. Secretary of State. TAT-1 was two parallel armoured coaxial cables running from Oban in Scotland to Clarenville in Newfoundland — a distance of about 3,600 kilometres — carrying 36 speech circuits at a bandwidth per circuit of 4 kHz. The cost of laying TAT-1 was approximately $42 million in 1956 dollars — call it $470 million today. The cost per circuit-mile was staggering. Every one of the 36 circuits was worth its own detailed study.

Bullington and Fraser noticed a fact about human conversation that had been known qualitatively for decades but had never been engineered around. On a full-duplex phone call, each party speaks approximately 40% of the time. They listen for approximately 40%. The remaining 20% is bilateral silence — pauses between phrases, breath, thought. Averaged across a large population of calls, the fraction of the duplex bandwidth actually carrying speech in either direction at any given instant is about 40%. The other 60% is either silence or unidirectional. The cable was carrying air.

TASI's mechanism is direct. On the transmit end, a speech detector monitors each of the 72 half-duplex channels (36 circuits × 2 directions). When a talker begins to speak, the TASI processor assigns their voice signal to any currently-unused physical channel across the cable. When they pause, the channel is released. On the receive end, a control signal tells the receiver which physical channel is currently carrying which conversation and re-routes it to the correct outgoing local circuit. The assignment changes every few tens of milliseconds. Because roughly 60% of the channels are silent at any instant, a 36-channel physical cable could carry approximately 72 simultaneous conversations.

The Bell System deployed TASI operationally on TAT-1 in 1960. Cable capacity doubled overnight without laying a metre of new cable. TASI ran on TAT-1 until the cable was retired in 1978, and on TAT-2 and TAT-3 through the same period. It was replaced by digital multiplexing on TAT-6 (1976), which was pulse-code-modulated and allowed pure time-division multiplexing without the analog-speech-detector step. TASI as a physical piece of equipment is gone.

TASI as a principle, however, is universal. The insight — use silence-detection to reclaim bandwidth in packet-switched voice — was rediscovered by the internet-telephony working groups of the mid-1990s and standardised as Voice Activity Detection (VAD) and Discontinuous Transmission (DTX). Every modern voice codec has a VAD stage. G.729 Annex B (ITU-T, November 1996) specifies a silence-suppression scheme with a comfort-noise generator (CNG) that keeps the far end's speaker from sounding dead during silences — the CNG synthesises a low-amplitude noise floor matched to the ambient noise on the transmit side. RFC 3551 (RTP A/V Profile, Schulzrinne and Casner, July 2003), §4.1, requires all RTP audio senders to support silence suppression and specifies the "SID" (Silence Insertion Descriptor) payload type. Opus (RFC 6716, 2012), the default codec for WebRTC and therefore for every Discord call and every Google Meet, supports a DTX mode that drops the bit-rate from around 32 kbps to around 1 kbps during detected silence and hands back the bandwidth to the rest of the network path. The mobile telephony analog — GSM Discontinuous Transmission — was specified in ETSI GSM 06.31 (1992) and has been on every mobile call since roughly 1995.

The interesting engineering wrinkle is the freezing-out effect, described in Bullington and Fraser §IV. In heavy load — when the number of simultaneous talkers exceeds the number of physical channels — the excess talker's first 8 to 24 milliseconds of speech is clipped as the system searches for and reassigns a free channel. This is inaudible for a well-designed system. At extreme load it becomes audible as a "chopped" initial syllable. Bullington and Fraser modelled the audibility threshold at approximately 32 ms of clipping per talker-onset. The exact same phenomenon, exactly measured, is why VoIP calls under packet loss lose the first syllable of a word before they lose an entire word. It is TASI's freeze-out clipping in a new medium, obeying the same psychoacoustic threshold established sixty-seven years ago in a Bell Labs office in New Jersey by two men trying to save a cable that had already been paid for.

Primary sources:

• Bullington, K., and Fraser, J. M., "Engineering Aspects of TASI," Bell System Technical Journal 38, no. 2, March 1959, pp. 353–364. The original paper. Available in facsimile from the AT&T archives; the March 1959 issue of BSTJ is on bstj.bell-labs.com.
• S. J. Campanella, "Digital Speech Interpolation Systems," IEEE Communications Magazine 17, no. 1, January 1979, pp. 8–15. The generalisation from analog TASI to the digital DSI (Digital Speech Interpolation) systems used on early satellite links. Establishes the terminology now used in mobile telephony.
• International Telecommunication Union, ITU-T Recommendation G.729 Annex B, November 1996. The formal silence-suppression standard for the G.729 codec, still in use as the default codec on most enterprise VoIP deployments in 2026.
• International Cable Protection Committee, TAT-1 to TAT-14: A History of Transatlantic Telephone Cables, ICPC, London, 2011. Chapter 3 covers the TAT-1 deployment and the TASI overlay in 1960; the operational figures for capacity gain are on pp. 41–43.

The Bullington and Fraser paper is the one to read. Its concluding paragraph reads: "The engineering principle here is not particular to submarine telephone cables. It is that any point-to-point transmission facility carrying human speech is, in expectation, carrying silence a majority of the time; and that this silence is a resource which the transmission facility has been paying for and has not been using. TASI is one way of using it. Other ways will be found. The general lesson we would ask the reader to take from this paper is that the underlying fact — the human talker is a low-duty-cycle signal source — is deeper and more useful than any particular scheme built on it."

Sixty-seven years later, friend's Zoom call is that scheme.

Today's goal

Ten minutes of nothing, sitting still.

Not a walk. Not a stretch. Not a meditation with a bell and a timer app. Sit in a chair, or on a step, or on the floor, and do nothing at all for ten minutes. No phone. No podcast. No notebook. Look at whatever is in front of friend — the wall, the window, the ceiling, the floor. Notice the sounds. Notice the light. Notice, without evaluating, whatever thought arrives.

The reason to do this on a Saturday is that Saturday is the day of the week on which the resistance to doing it is weakest. The reason to do it at all is that the accumulated stock of unprocessed low-priority background thought — the "salt oil layer" on top of the mental tank — sinks through the seawater and settles at the bottom in about eight minutes of silence, and can then be swept out by a broom, and the reason it takes eight minutes rather than one is that the first seven minutes are the broom finding the corner.

Ten minutes. No dipstick required.

Today's toy in the corner is shell-game — three walnut halves on a green baize table. A pea appears under one. The dealer shuffles. friend picks. There is a switch on the console labelled honest — when it is on, the dealer plays it straight and the pea stays under the shell it was under to start. There is another switch labelled watch the hands — when it is on, a small dot on the table shows the true location of the pea during the shuffle, including any brief moment when it is not under any shell at all. Space starts a new round. H toggles honest. W toggles the hands.

Play a few rounds with honest on and hands off. Then turn honest off and try to keep winning. Then turn hands on and see what the dealer was actually doing. Then reach whatever conclusion friend was going to reach.

Send me what friend figures out. There is no address; that is the point.

— C