Coincidence is cheap

A name has a number. How surprising is it that the Torah contains a word with the same number? We counted for every value from 1 to 1500 — and the answer is: hardly surprising at all.

That many values from 1 to 1000 have at least one content word with the same sum in the Pentateuch under the strict rule. Under the loose rule, 937 of 1000. A match is the norm, not an event. What has value is not the match but who recorded it, and when.

Two rules, two answers

Any such figure depends on what counts as a word. Strict: consonants only, prefixes stripped, nouns, verbs and adjectives only. Loose: every word form as written, any part of speech. We publish both, because a reader who checks must find exactly what we said.

By size of the number

Values with no word at all

Under the strict rule, among values 1–861 there is no word for 56: 1, 2, 229, 279, 369, 379, 384, 533, 549, 574, 579, 583, 587, 593, 597, 629, 634, 638, 639, 642, 644, 664, 669, 694, 699, 721, 727, 729, 733, 739, 754, 765, 767, 771, 773, 774, 779, 783, 787, 795, 797, 803, 813, 818, 821, 823, 827, 831, 833, 837, 839, 841, 843, 847, 853, 861. Under the loose rule, only for 13: 1, 2, 583, 599, 729, 773, 779, 801, 827, 833, 839, 847, 859.

The number 861, which gave the project its name, has no content word under the strict rule; under the loose rule it has two: ונתתה (Exodus 21:23) and באמתחתי (Genesis 42:28). We do not hide this.

Values present in every portion

Under the strict rule only six values have a word in all 54 weekly portions: 26, 50, 56, 62, 140, 311. Under the loose rule, eighteen: 26, 30, 31, 36, 50, 52, 58, 62, 74, 76, 80, 90, 100, 115, 120, 170, 401, 501.

What follows

If almost any name will find a Torah word with the same number, the match itself cannot prove anything. That is why our readings place something else next to the match: who recorded the equality, and in what century. The attestation database is here: /research/kolel.

What the null model gave

We permuted every consonant of the Pentateuch, and separately regenerated each word from a first-order Markov chain fitted on the Torah itself — a thousand replicates for the first model, three hundred for the second. The random text covered more: 947 ± 5 and 938 ± 5 of the first thousand values against our 867. Not one replicate fell to the observed level. The reason is plain: real language repeats itself, so it yields fewer distinct sums than a random letter stream of the same size. Coverage is therefore not a mark of design — it is below what chance alone delivers. Still missing: a comparison against other Hebrew corpora, a surrogate beyond first order, and an independent re-derivation of the attestation dataset. Model, code and the per-value table: https://doi.org/10.5281/zenodo.22848828

Download the table (CSV): value, portions under the strict rule, portions under the loose rule

The other study: One in four equalities needs a tolerance · /method · /research/spec