Why a Match Is Always Available

The half of the claim you can check

A numerological claim has two parts. First, a match: these letters total that number, these two lives share this detail, this date reduces to that figure. Second, almost always implied rather than stated, and a match like this is unlikely.

Only the second part can be checked, and checking it needs a number the claim never supplies. How many chances did the search get before a hit was declared?

That isn't a matter of opinion. It's the product of everything free to vary. How many ciphers were available. How many ways the name could be written. How many ways the date could be expressed. How large the set of numbers counting as significant was. Lay those out and the arithmetic is secondary-school arithmetic, which is what the calculator above does.

The answer is usually that a match was close to certain. Not likely. Certain.

The target gets drawn after the shot

The formal name is the Texas sharpshooter fallacy. Fire at the barn, then paint the target around the tightest cluster. Every hit is a bullseye and the marksmanship can't be falsified.

Particle physics has the same problem under a drier name, the look-elsewhere effect, and an expensive institutional answer to it. Scan a spectrum for a bump and you'll find bumps, because the spectrum is long. That's why the field won't call anything a discovery below five sigma, a threshold corresponding to about one chance in 3.5 million. Physicists aren't more sceptical by temperament. They just count how many places they looked, and set the threshold to survive it.

Nobody reading letter-sums counts how many places they looked. The cipher gets chosen after the word, the spelling after the cipher, and the significance of the resulting number after all three.

Intuition is badly calibrated even with nobody cheating

Before any of that there's a cleaner problem. People are simply wrong about how often coincidences happen when nothing at all is being manipulated.

Put twenty-three people in a room and it's more likely than not that two share a birthday. Most people guess the answer is somewhere near a hundred and eighty. The gap opens because the question isn't does anyone match me, it's does any pair match, and twenty-three people make 253 pairs. The number of chances grows much faster than the number of people. Same mechanism as the calculator, with no bad faith in it anywhere.

J. E. Littlewood, the Cambridge mathematician, put the general version as a joke with real arithmetic behind it. Define a miracle as a one-in-a-million event. A person is alert for maybe eight hours a day and experiences roughly one distinguishable thing a second. That's about 30,000 events a day, so a million events take about thirty-five days. On that accounting each of us should expect a miracle roughly once a month, and the striking thing about the estimate is how well it matches how often people actually report one.

The list everybody has seen

The Lincoln and Kennedy parallels first appeared in print in 1964, months after the assassination, and have circulated ever since. Elected to Congress in '46, to the presidency in '60. Both succeeded by a Southerner named Johnson. Both shot on a Friday, in the head, beside their wives.

Martin Gardner worked through the list in Scientific American and reprinted the result in The Magic Numbers of Dr. Matrix. Two things come out of it. The first is that several items are simply untrue. The famous one about a secretary named Kennedy has no basis at all; Lincoln's secretaries were John Nicolay and John Hay. The second is more interesting, because it applies to the items that are accurate.

Two American presidents have, between them, some tens of thousands of recorded facts. Dates, places, family members, staff names, injuries, speeches, physical descriptions. A parallel counts if it shows up in any of those categories, and nothing was specified in advance. Under those conditions a list of a dozen or so matches isn't an anomaly needing explanation. It's the expected yield. The test is whether the same exercise works on other pairs of presidents, and it does, which is the load the list can't carry.

The time it was run as a controlled experiment

The best demonstration on this subject wasn't an argument. It was a control.

In 1994 Statistical Science published a paper by Witztum, Rips and Rosenberg reporting that the names of later rabbinic figures appeared in Genesis, read at fixed letter intervals, close to their dates of birth and death, at odds they put at about one in sixty thousand. In 1999 the same journal published an examination by Brendan McKay, Dror Bar-Natan, Maya Bar-Hillel and Gil Kalai showing the result turned on choices made during the analysis.

Then came the part that settled it. Michael Drosnin, whose 1997 bestseller popularised the method, said publicly that he'd believe his critics when they found an assassination encoded in Moby-Dick. McKay applied the same method to Melville and produced encoded assassinations for Indira Gandhi, Martin Luther King Jr, John F. Kennedy, Abraham Lincoln, Yitzhak Rabin, and later Princess Diana. McKay's team also found that War and Peace yielded results as good as Genesis.

This is what a control text is for, and it's the first thing to ask of any method of this kind. Run it on something that cannot contain the signal. If it finds the same density of meaning in a novel about a whale, it's measuring itself rather than the text.

What this doesn't establish

The arithmetic shows a match is expected. It doesn't show any particular claim is false, and reading it that way is just the opposite error.

Real patterns produce matches too. Sometimes the number does mean something. The archive's own entry on gematria is careful about this, because the practice has a documented history in rabbinic exegesis where the technique is doing genuine interpretive work rather than prediction. A coincidence calculation can't tell a real signal from a manufactured one. All it can tell you is that the presence of a hit isn't informative by itself.

What does separate them is one thing, and it's cheap. Say what would count before you look. Name the cipher, the spelling, the date form and the target number in advance, then check. The first preset in the calculator is what that looks like, and the odds it returns are the odds a claim would have to beat.

Almost nobody does this, which is itself the most useful fact about the field. A method that can't be stated in advance isn't being tested. It's being fitted.