writing

Yardsticks

Mar 2026

In 1951, CIA analysts unanimously agreed a Soviet invasion of Yugoslavia was a “serious possibility.” When Sherman Kent asked his colleagues what odds they’d had in mind, answers ranged from 20% to 80%: we’re not very good at thinking in numbers 1.

The problem runs both ways. Vague words fail to map to numbers; that’s Kent’s case. But numbers also fail to map to intuition. A billion seconds means nothing to most people; “one career” means something to everyone. The original yardstick was Henry I’s arm: the distance from his nose to his outstretched thumb 2; probably legend, but it lasted centuries because a body is easier to picture than a number. The “yardsticks” below do the same thing; swap raw figures for a vacation, a Porsche, a pregnancy.

Money

Imagining 10x your wealth is hard; imagining everything costing a tenth is easy. A Porsche 911 at $10,000. 3

Odds

Per try

Per lifetime

Time

Speed

Latency at Human Scale

If a CPU cycle were one heartbeat (0.5s): 29

Distance

Data

Energy

Volume

Quantities

Correlation (r)

How much does knowing X help you guess Y? Square r to find out: r = 0.68 for height and weight means r² ≈ 0.46, so if you know someone’s height, you are almost halfway to their weight.

Below 0.3 the relationship is weak; above 0.8, each variable nearly determines the other.

Reliability 48

NinesDowntime/yearDowntime/day
90%37 days2h
99%4 days14m
99.9%9h1m
99.99%52m8s
99.999%5m<1s
99.9999%31s86ms

Deaths

  1. Kent, “Words of Estimative Probability,” 1964 ↩

  2. Try it; nose to thumb is surprisingly awkward. The earliest account, William of Malmesbury’s Gesta Regum Anglorum, actually says Henry standardised the ell, not the yard, using “the measure of his arm.” ↩

  3. A 2026 911 Carrera lists at $129,950, so a true tenth is $13,000; $10,000 is the same number rounded to one significant figure, the right precision for a daydream. ↩

  4. Median UK full-time pay was £39,039 a year and £19.67 an hour in April 2025 (ONS). To make a pound buy a minute you would need to earn about £120,000. ↩

  5. The 4% rule turned upside down. Bengen’s 1994 study found a portfolio drawn at 4% survived every 30-year window in the historical US record; invert 4% and you get the 25× that usually gets quoted, with 30× the cautious end. Cutting your spending works on both sides of the sum at once. ↩

  6. Two soft numbers multiplied together: Google’s index at “about 400 billion documents” and a $1.775 trillion FY2025 deficit (CBO). $5 × 400 billion ≈ $2 trillion, so the yardstick survives an order of magnitude of error in either input. ↩

  7. The tenth-richest person’s $150 billion over a 40-year working life is $120 a second; bending down takes two or three seconds, so the note is worth less than the time. Wealth is a stock and the rate is a fiction. ↩

  8. 50.73% at 23 people, 47.57% at 22. People guess far too low, because they picture 22 comparisons against themselves rather than the 253 pairs in the room. ↩

  9. Three to four monozygotic twin births per thousand, almost everywhere it has been measured; fraternal twinning swings from about 6 per thousand in Asia to 40 in parts of Africa. ↩

  10. The National Hole-in-One Registry says 1 in 12,500 for an amateur and 1 in 2,500 for a professional. The average golfer who gets one waits 24 years for it. ↩

  11. The distribution is not what people expect: most deaths are experienced jumpers under a working parachute, and tandem students, who are the ones actually frightened, are safer still. ↩

  12. The average British commute runs 1,007 miles a year, at 2.7 deaths per billion miles: 1 in 368,000. A year of driving to work costs about what one skydive costs, and neither feels remotely like the other. ↩

  13. The risk per passenger, not per flight: IATA records one fatal accident per 5.6 million flights across 2021—2025, and most fatal accidents leave survivors. ↩

  14. The median of thirteen major draws rounds to 1 in 100 million. Buying two tickets doubles your chance. Buying one a week for a century does not get you anywhere near even odds. ↩

  15. If everyone alive had shuffled a deck once a second since the Big Bang, they would still be forty orders of magnitude short of covering the possibilities. 52! is a perfectly precise quantity that means nothing at all until you unpack it. ↩

  16. One US lifetime holds a 1 in 100 chance of dying on the road (National Safety Council) and 1 in 4,000,000 of dying to a shark; the everyday risk beats the exotic one by four orders of magnitude. Britain is roughly five times safer. ↩

  17. The US National Weather Service puts the lifetime odds at 1 in 15,300 counting injuries, 1 in 79,746 counting deaths alone. These are population averages: an outdoor worker in Florida and an office worker in Manchester do not share a risk. ↩

  18. The International Shark Attack File gives 1 in 4,332,817 over a US lifetime. Figures in the hundreds of millions are annual, not lifetime. Sharks kill five people a year worldwide while people kill roughly 100 million sharks. ↩

  19. About 5,000 meteorites reach the ground each year; a 100 m² house over an 80-year life comes to about 1 in 13 million. Multiplied by the billion-odd houses on Earth, that is a hit somewhere once a year, roughly how often it is reported. ↩

  20. 11 days, 13 hours, 46 minutes and 40 seconds; pleasingly, it lands almost exactly on a fortnight’s holiday. ↩

  21. Gladwell’s rule, from Ericsson, Krampe & Tesch-Römer’s 1993 study of violinists. Ericsson objected for years: 10,000 was the average accumulated by the best group by age 20, not a threshold. ↩

  22. 31.7 years, so a 30-year career is a shade under. A million seconds is a holiday and a billion is a working life; the factor of a thousand between them is the whole point. ↩

  23. Global life expectancy at birth was 73.3 years in 2024 (UN World Population Prospects 2024); the rich-world figure is roughly 80. These are period estimates; a child born today will probably do better. ↩

  24. The oxygenated atmosphere has about 1.08 billion years left (Ozaki & Reinhard, 2021). The mechanism is CO₂ starvation, not heat: as the Sun brightens, silicate weathering draws carbon dioxide below what photosynthesis needs, and oxygen collapses. ↩

  25. 4.54 ± 0.05 billion years, from lead-lead dating of meteorites; Patterson’s 1956 measurement has survived seventy years of refinement almost unchanged. ↩

  26. 13.787 ± 0.020 billion years from the Planck 2018 results; measurements that climb the local distance ladder give a faster expansion rate and so a younger universe. ↩

  27. A handgun round leaves the muzzle at 350—400 m/s, so a millisecond carries it about 40cm. ↩

  28. Sarah, at Cincinnati Zoo, ran 100m in 5.95 seconds, with a radar peak near 27 m/s. Wild cheetahs are slower: across 367 tracked hunts in Botswana the mean top speed was 14.9 m/s. ↩

  29. Scaled ×10⁹ from Jeff Dean’s “Latency Numbers Every Programmer Should Know”; the ratios are the durable part, the absolute figures roughly 2012 vintage. ↩

  30. 384,400km on average; the equator is 40,075km, so the Moon is 9.6 laps. ↩ ↩2

  31. About 500 words at roughly five characters each, one byte per character. This footnote is around 300 bytes. ↩

  32. The current-article dump enwiki-latest-pages-articles.xml.bz2 was 25.6GB in August 2026: English text only, compressed, with no images and no edit history. ↩

  33. 3.1 billion base pairs at two bits each, 1.5GB for the pair you actually carry. Stored as a diff against a reference genome, the part that makes you you is a few megabytes. ↩

  34. The banana equivalent dose is a teaching device; the body holds potassium at a fixed concentration, so eating a banana does not raise your radiation dose. ↩

  35. A 51g Mars bar is 228 kcal; walking burns roughly 250 kcal an hour for a 70kg adult. The eating takes ninety seconds. ↩

  36. A cubic metre of water is a tonne, which is why a fish tank that size needs a thought about the floor. ↩

  37. Not a coincidence but a definition; the kilogram was fixed as the mass of a litre of water at 4°C in 1799. ↩

  38. The average English secondary school held 1,054 pupils across Years 7—11 in 2022/23, which is about 200 a year group. ↩

  39. The useful direction is the reverse: when someone quotes a rate in ppm, ask how many seconds of a fortnight that is. ↩

  40. 69.3 million at mid-2024 (ONS). The figure has moved fast, so the “67 million” most people carry in their heads is about five years out of date. ↩

  41. Diener & Oishi, 2000. A within-country figure; across countries the correlation between income and life satisfaction is far stronger, around 0.6 to 0.8. ↩

  42. Kobrin et al., 2008: 0.35 among enrolled students, rising to 0.53 once corrected for range restriction. ↩

  43. Pearson & Lee, 1903, father to son. The pattern is what Galton, 1886 named regression to the mean: tall fathers have sons taller than average but shorter than themselves. ↩

  44. Silverman, 2022. Within a single sex the correlation drops; much of what height tells you about weight is telling you whether you are looking at a man or a woman. ↩

  45. Bouchard et al., 1990, the Minnesota Study of Twins Reared Apart. Separated twins were mostly placed in similar homes, the standard objection to a clean heritability estimate. ↩

  46. Bouchard & McGue, 1981, a review of 111 studies. 0.86 is close to the reliability ceiling of the tests themselves, so identical twins raised together score about as alike as one person tested twice. ↩

  47. Roth et al., 2013. People report their height accurately relative to each other while systematically rounding themselves upward. ↩

  48. Rounded to the nearest whole unit. ↩

  49. A full widebody is about 200 lives. The Aviation Safety Network counted 548 deaths in 2025, the worst year since 2018; even that bad year is one crash every three months at widebody scale. ↩

  50. 132 million births and 63 million deaths in 2025 (Our World in Data), or about 360,000 and 170,000 a day. Roughly two births for every death; the remainder is population growth. ↩ ↩2