The rabbis experiment

What was done

Doron Witztum, Eliyahu Rips and Yoav Rosenberg took the 32 people whose entries in Margalioth's Encyclopedia of Great Men in Israel ran from 1.5 to 3 columns. For each they took names and titles, and the Hebrew day and month of birth or death. Each name paired with each date form gave 298 word pairs. They measured how close each pair stood as letter sequences in Genesis, using a corrected distance they had designed for that purpose.

Then they scrambled which rabbi got which dates, a million times less one, and asked where the true pairing ranked. The method page gives every step and formula.

What they reported

The rank of 4 in a million, multiplied by 4 for having four statistics, gives 0.000016, or 1 in 62,500. That is the number you will see quoted. The authors concluded that the closeness of related sequences in Genesis was not due to chance. They asked readers to treat their P figures as ordinal indices, not as probabilities. The journal's editor called the paper a challenging puzzle.

What happened when others tried

In 1999 Brendan McKay, Dror Bar-Natan, Maya Bar-Hillel and Gil Kalai confirmed the arithmetic. They argued that the source of the number was the freedom in choosing names and date forms. Using a new expert, with lists he drew up on his own, they ran the same test and got 0.233 for the same 32 rabbis and 0.404 for 26 new ones.

• The result depends on a few items. Removing 4 of the 32 rabbis takes it from 1 in 60,000 to 1 in 30.
• The date forms left out of the experiment do poorly alone.
• Other editions of Genesis weaken it. On the Leningrad text, which we search, the best rank is 0.0013 for this list.
• In the rest of the Torah this list ranks from 0.10 to 0.85.
• A list with changed names scores about as well in War and Peace.

Witztum disputes these findings. He argues that the War and Peace list breaks the rules that governed the original, and in 2022 he argued that the permutation test itself is the wrong tool. Others in his circle report further positive results. The critique page has the full exchange.

The committee's test

The closest thing to an agreed replication came from a Hebrew University committee formed in 1996, chaired by Robert Aumann. It tested a related claim that used rabbis' communities in place of dates. Both experiments landed in the middle of the pack, at 463,406 and 578,075 out of a million. Two members disagreed with the report. Our entry on that test has the details.

What you can do here

You cannot re-run the whole experiment in a browser, since a full test needs a million scrambles of about 300 pairs. You can do the next best thing. The laboratory computes the WRR corrected distance for any pair of 5 to 8 letters within a bounded skip range, ranks it among controls, and shows you how it moves when you change a setting. Try one pair, then change a spelling, and see for yourself how sensitive a single value is.

We show this entry as a record of what was published. The figures above belong to the Koren text and to their authors. No sequence is displayed here, and none is claimed for our text. If you want the paper's own account, its appendix sets out each step so that a reader can repeat the work.