In 1994 a peer-reviewed statistics journal published evidence that the names of medieval
rabbis were encoded in Genesis as equidistant letter sequences — take every
50th letter, read the words. By 1997, The Bible Code was finding assassinations.
It is the perfect closing statistics case, because the source text is
actually real and remarkable, the statistical method family is
genuinely useful elsewhere, and the prophecy is an
overclaim — and the whole dispute turns on one computable
question: how many hidden words should any text contain by chance?
Michael Drosnin issued the kill test himself: "When my critics find a message about the
assassination of a prime minister encrypted in Moby Dick, I'll believe them."
Below is Moby Dick. Let's oblige him again.
Fig. 1 — The decoder, pointed at 48,000 letters of Moby-Dick (chapters 1–9, Project Gutenberg text). Wrap the letters at the skip distance and the "code" stands up as a column — the same presentation the Bible-code books use. Below: words found vs. words chance predicts.
The method is real and the search below is genuine. An equidistant letter
sequence (ELS) starts at any letter and takes every $s$-th letter after it;
Bible-code "messages" are words assembled this way, displayed by wrapping the text at
the skip so the word stands up as a column. Nothing is faked here: the text is
Melville's, the search is exhaustive, and the finds are real.
The question was never whether hidden words exist. It's how many should. For a
$k$-letter word with letter frequencies $p(c_i)$, in a stream of $L$ letters, over
skips $2 \le s \le S$ in both directions, pure chance predicts:
Run any word and compare the count against the formula: they agree to within Poisson
noise, every time — OMEN appears 849 times against a predicted 831; TRUTH 39
against 39. Then press famous finds: RABIN runs backwards through
Ishmael's "coffin warehouses… the rear of every funeral I meet"; COMET pierces "the
horrid spell would be broken"; TRUMP crosses FAMINE seventeen letters apart. All real,
all in one whaling novel — all at exactly the rate the alphabet requires.
That is the entire Bible-code argument, run in reverse. Brendan McKay answered Drosnin's
challenge in 1997 by finding nine assassinations in Moby-Dick, and
with Bar-Natan showed the celebrated "famous rabbis" result rode on flexible name
spellings — freedom that conjures equal miracles from War and Peace.
The same claim, read by parallax — from the manuscript to the prophecy industry.
actually real
The text is extraordinary — and so was the paper.
Signal. The Masoretic text really is a marvel of transmission: scribes counted letters, marked midpoints, and discarded scrolls over single errors — a checksummed manuscript tradition a millennium deep. And Witztum–Rips–Rosenberg (1994) was a real, peer-reviewed Statistical Science paper; taking it seriously was the correct scientific reflex.
Kill test. Reproduce the transmission claim: collate surviving codices. The fidelity survives. The prophecy claim gets its own test — two cards down.
genuinely open
What can letter statistics prove about a text?
Signal. Statistical text analysis is a live, legitimate field: stylometry settled the disputed Federalist Papers, flags forgeries, and de-anonymizes authors today. Counting letters genuinely extracts truths — about authorship, not the future.
Unresolved. Where the method's edge sits — which attribution problems are solvable, at what text lengths, against adversarial writing — is an active research frontier.
Kill test. Blind attribution trials on known-author corpora, run continuously by the field. The honest version of "hidden signals in text" is thriving; it just finds Hamilton, not Armageddon.
real kernel · overreached
"The rabbis' names are encoded in Genesis."
Signal. The kernel is real: the WRR experiment found the famous-rabbis correlation at a reported p ≈ 0.00002, with serious mathematicians behind it, and it survived four years of review.
Noise. The names and dates had spelling freedom — many defensible variants each — and the chosen variants happened to be the ones that worked. McKay & Bar-Natan re-ran the experiment with equally defensible spellings tuned toward War and Peace: Tolstoy prophesies the rabbis just as well.
Kill test. Fix the spellings in advance (independent experts, sealed) and re-run. Done in the 1999 Statistical Science rebuttal and after: the signal evaporates. Run, and failed.
Signal. Drosnin's Rabin find felt uncanny — and the feeling is data about pattern-hungry minds reading an unbounded search space.
Noise. Every "prediction" was found after the event, with free choice of words, spellings, skips, and adjacency — a search protocol that cannot miss. A code that only reads yesterday forbids nothing about tomorrow.
Kill test. Drosnin's own: assassinations in Moby-Dick. McKay found nine (Kennedy, Lincoln, Rabin among them); this page's 48,000-letter excerpt volunteers RABIN beside Ishmael's funerals unprompted. The prophecy rate is the letter-frequency rate.
The sift, one more time: the manuscript survives its audit; the honest statistics moved to
stylometry and thrives; and the prophecy dies by its own challenge, issued in print and
answered in whales. A message that appears in every book was written by the
alphabet, not the author.