Luke Ford

Princeton, the kosher dining hall, the academic year 1979-80. Moshe Koppel (b. 1956) had finished a doctorate in mathematics at NYU's Courant Institute a few months earlier, under Martin Davis (1928-2023), and had come down to the Institute for Advanced Study for a postdoctoral year. He was in his early twenties. He had grown up in New York, in yeshiva, and had spent a year at Yeshivat Har Etzion in the Judean hills. He knew Hebrew, Aramaic, Talmud, and recursion theory. He did not know what to say to the woman across the table.

She was Jewish, secular, a graduate student. In the book he wrote forty years later he called her Heidi and acknowledged that the character was partly composite. Her parents kept a kosher home on Long Island and belonged to a Conservative synagogue. At Princeton she had made friends from everywhere. She asked Koppel how he justified his tribal loyalty to the Jews. The lesson of the Holocaust, she said, was that Jews above all should never put their own interests ahead of anyone else's.

Koppel later described his own reaction in three words: “slack-jawed, staring at her uncomprehendingly.”

He had no argument ready. He had grown up around people for whom the question could not arise. His grandfather prayed in a Gerrer shtiebel in Manhattan with a small band of Polish Hasidim who had survived the camps. One of them, a merchant Koppel calls Shimen in the book, had lost his wife and two children in the Holocaust; he came out of Lodz and Auschwitz, and after the war he picked up a handgun and went looking for Jewish children who had been hidden with gentile families. Elie Wiesel (1928-2016) davened in that shtiebel from time to time. Koppel wrote of those men that “one thing they had no patience for was high-minded pieties.” They were funny in a hard way. They took liberties with God because they were on close terms with Him. They would not pour boiling water over a tea bag on Shabbat.

Shimen could not have answered Heidi either. He would not have understood why she thought an answer was owed.

The rest of Koppel's career is, among other things, a forty-year project to supply the answer, using tools Shimen would have found unrecognizable and Heidi would have found congenial.

The dissertation work was in decidability. In 1979 he published a solution to a problem posed by Davis, Yuri Matiyasevich (b. 1947), and Julia Robinson (1919-1985) about which Diophantine problems can be settled by mechanical procedure. A later paper asked how an investigator chooses among an ordered hierarchy of hypotheses that all fit the observed facts. The recurring question in both: what can be decided from the evidence available, and what remains open no matter how long you compute.

He moved to Israel in 1980 and joined the computer science department at Bar-Ilan University, where he spent his academic career.

The problem he became known for has the same structure as the ones he trained on. A text is visible. Its author is hidden. What in the visible thing survives the author's control?

A historian writing about Napoleon and a physicist writing about superconductivity use different nouns for reasons that have nothing to do with either man. Koppel and his collaborators, chiefly Shlomo Argamon and Jonathan Schler, worked instead on the parts of language a writer does not choose: pronouns, articles, prepositions, conjunctions, suffixes, punctuation, the rate at which a given grammatical form appears. A man can decide to stop using a favorite adjective. He cannot hold three hundred small grammatical habits under conscious guard across ten thousand words.

In 2003 the group reported that the same statistical signals could predict a writer's sex, and later age and native language. The Boston Globe and The New York Times both ran pieces on it that summer. The findings said that social experience deposits measurable residue in prose the writer is not aware of supplying. That was enough to generate a news cycle.

Koppel and Schler then built the method that carries his name furthest. Train a classifier to separate two bodies of writing. Find the features doing the most separating. Delete them. Retrain. Repeat. If the two bodies come from the same hand, accuracy collapses fast once the surface differences are stripped; the classifier runs out of things to grab. If they come from different hands, deeper differences remain and accuracy degrades slowly. They called it unmasking. It measures how far down the difference goes.

The first serious application was a rabbi who had hidden his own authorship.

Yosef Hayim of Baghdad (1835-1909), known by the title of his best-known work as the Ben Ish Chai, was the leading halakhic authority of Iraqi Jewry. In 1878 he published a collection of 524 responsa called Torah Lishmah, presenting it as an old manuscript he had come across rather than as his own composition. Historians had long suspected he wrote it himself, for reasons ranging from modesty to a wish to have the rulings judged on their merits rather than on his signature.

Koppel and Schler compared Torah Lishmah against 509 responsa Yosef Hayim acknowledged, in Rav Pe’alim, and against collections by four other rabbis of the period. The comparison authors fell away. The distance between the disputed collection and the acknowledged one behaved the way same-author distances behave. This supplied a new kind of evidence for a conclusion scholars had reached by other means, and it did so on an unusually hard case, since a man writing under a pseudonym has every incentive to change his habits on purpose.

He later turned the methods on the Kherson Genizah, a trove of letters presented as copies of correspondence among the Baal Shem Tov, Shneur Zalman of Liadi, and other early Hasidic masters, which surfaced in Ukraine around 1917 and which most scholars had already judged fraudulent. Koppel's analysis found that the supposed differences among the various attributed authors did not behave like differences between real independent writers. He published it in Hebrew, in a rabbinic journal, in 2010.

Then came the Bible.

In 2011 Koppel published an ACL paper with his student Navot Akiva and with Nachum Dershowitz (b. 1951) and Idan Dershowitz on splitting a document into stylistic components without any labeled examples. The method builds a list of synonym sets, roughly two hundred of them for biblical Hebrew, generated automatically by finding Hebrew roots that the King James renders with the same English root. For each chunk of text it records which synonym from each set the chunk uses. Chunks that make similar choices cluster together.

To test it, they took two books by two authors, shuffled them, and asked the method to unshuffle them. Jeremiah and Ezekiel. About seventeen percent of the verses could not be classified for lack of any distinguishing word. Of the roughly 2,200 that were classified, 26 landed on the wrong side. A 1.2 percent error rate.

Then they ran it on the Torah. Split into two families, the output matched the standard critical division between Priestly and non-Priestly material for about ninety percent of the verses where the standard authorities agree with each other.

The wire services got hold of it. An algorithm sheds light on the authorship of the Bible. By the time the story returned to Israel, Haaretz readers learned that the head of the project had announced proof of multiple human authorship of the Torah, which was news to the head of the project.

Koppel wrote a correction and published it on the Seforim Blog, a site read by the people who would care. It is the clearest thing he has written about the limits of his own instruments.

The number of families, he explained, is not discovered by the method. It is supplied by the operator. Ask it for two and it gives you two. Ask it to split Moby-Dick into thirteen and it will. That the Torah splits in two therefore says nothing whatsoever about how many authors it had. He also declined to let his readers take too much comfort: with better methods one probably could estimate an optimal number, and the correspondence with the critics' division suggests the split is not an artifact. Ask three families and the correspondence degrades. Ask four or more and the critics' J and E do not appear at all.

On the theological question he was blunter. The work could have no bearing on divine authorship, because inferring multiple authors from multiple styles rests on assumptions about human cognition and human performance, and nobody has a theory of what a divinely written text is supposed to look like. He noted the obvious objection, that this makes the claim unfalsifiable, and said the objection deserved a longer discussion than he was going to give it there.

Of the question that had opened every lecture he ever gave on authorship, “have you analyzed the Bible,” he wrote: “I never really understood the question.”

His Talmudic work runs on the same problem from the other side. Meta-Halakhah: Logic, Intuition, and the Unfolding of Jewish Law, published in 1996, argues that Jewish law works like a natural language. A grammar can describe a language without generating every judgment a competent speaker makes. Native speakers know a sentence is wrong before they can name the rule it breaks. Accomplished halakhists recognize that an argument does not fit before they can say which principle it violates. The intuition is competence built by immersion in cases.

The position sits between two accounts he rejects. One treats halakhah as a closed formal system from which every answer can be derived. The other treats it as an open field where a sufficiently clever reader reaches whatever result he already wanted. A language constrains without being exhaustively codifiable, and it changes through use. So does the law.

He also wrote Seder Kinnim, a mathematical commentary on the Mishnah‘s tractate about sacrificial birds that get mixed among groups of different legal status, generally reckoned the hardest tractate in the Mishnah. He founded and co-edited Higayon, a journal on rabbinic logic, with Ely Merzbach. He wrote on probability in halakhah and on rules for uncertainty in Talmudic reasoning with the logician Dov Gabbay.

Formalize the tacit; do not claim the formalization replaces it. That is the move, repeated across four decades and five fields.

Dicta, the Israel Center for Text Analysis, is the largest thing he has built that has a test set.

Hebrew is hard for tools designed around English. Prefixes carrying conjunctions, articles and prepositions attach to the front of words. Suffixes carry possession and objects. Ordinary Hebrew drops the vowels, so one written string maps to several pronunciations, several parses, several meanings. Biblical, rabbinic, medieval and modern Hebrew differ in vocabulary, syntax and spelling. Rabbinic texts add Aramaic, dense abbreviation, and orthography that never stabilized.

Dicta's tools do vocalization, abbreviation expansion, morphological analysis, source identification, parallel-text discovery, manuscript reconstruction, authorship analysis. Most are free.

One paper shows the house style. Finding every parallel passage in the Babylonian Talmud by brute force would require something like 1.6 trillion edit-distance calculations, which Koppel and his coauthors Avi Shmidman and Ely Porat estimated at roughly twenty years on one machine. Instead they represented each word by its two rarest letters. This works because the most common Hebrew letters are yod, aleph and vav, which are also the vowel markers and the common prefixes, so throwing them away throws away most of the spelling noise. Then they looked for four- and five-word strings differing by at most one word, and clustered the matches. The run found 4,602 parallel pairs across 1.8 million words in eleven seconds.

That is the pattern: find the smallest signal that survives the corruption you care about.

The models followed. BEREL in 2022, a language model trained on rabbinic Hebrew, on the premise that rabbinic Hebrew is a domain in its own right and not defective modern Hebrew. DictaBERT in 2023, with fine-tuned versions for prefix segmentation, morphological tagging and question answering. DictaLM the same year, DictaLM 2.0 in 2024, adapted from Mistral with a Hebrew evaluation suite. Dicta-LM 3.0 in December 2025, open-weight models at 1.7, 12 and 24 billion parameters, 65,000-token context, trained on roughly 100 billion Hebrew tokens and 30 billion English.

The phrase in the technical report is “sovereign Hebrew.” A society whose inheritance sits in a low-resource language becomes a tenant when the models through which it reads and writes are built abroad. Koppel's answer is to build the datasets, the benchmarks, the morphology, the models, and to release enough that others can check the work.

Through 2023, through the war, through Kohelet's collapse in funding, that work continued.

Judaism Straight Up: Why Real Religion Endures appeared in 2020, forty years after the dining hall. It began as a blog called Judaism without Apologies.

The book puts Shimen and Heidi side by side, though they never meet. Shimen inhabits an inherited world of ritual, kinship, food restrictions, memory and specific obligation. He cannot give a philosophical defense of most of what he does. His Judaism is a first language. Heidi's is a second language, learned through stated rules in formal settings, and she assumes that a practice earns allegiance only after it justifies itself by independently established principles.

Koppel's argument is that practice precedes belief. Nobody selects a theology and then deduces a life. People acquire habits, loyalties, reflexes and expectations by growing up among other people, and their stated beliefs come along afterward as explanation and repair. A tradition survives because its institutions keep producing people willing and able to carry it, not because each generation re-derives its propositions.

The argument is Hayekian in structure, though Koppel draws on economics, game theory, anthropology and moral psychology rather than applying Friedrich Hayek (1899-1992) to religion. Knowledge can be distributed across a community and stored in practices. A rule that looks pointless in isolation may be holding up a system that produces trust and restraint. The reformer sees the inconvenience. He may not see the network.

Koppel does not claim that survival vindicates a practice. Slavery survived. His claim is that inherited practices deserve interpretation before demolition, because they may contain information the reformer cannot see, and the burden falls partly on whoever proposes to dismantle a system whose coordinating functions nobody has mapped. He builds his account of moral foundations on Richard Shweder (b. 1945), arguing that Shimen's world runs on fairness, loyalty and restraint together while Heidi's runs on fairness alone.

Concrete obligations to family, neighbors and Jews are the institutions through which moral responsibility becomes something a person can act on. A man who claims equal concern for everyone may find he owes no one anything in particular. A man embedded in thick relations knows whose children need help this week.

The critics found the construction unfair, and they had a case. Heidi is thin where Shimen is thick, and the coercion and cruelty available inside traditional communities get less attention than the sterility of cosmopolitan life. Benjamin Schvarcz, writing from Princeton’s James Madison Program, pressed a different point: Shimen thanks America the way a guest thanks a hotel. If patriotism is a form of loyalty, Shimen has less of it than Heidi does, and Koppel’s own Israeli argument depends on the political loyalty his hero lacks. The Fathom reviewer pressed the sharpest one. Shimen’s children were murdered. Whether his Judaism transmits across generations is what the case cannot show.

Koppel entered Israeli constitutional politics before he built anything. He sat in on the Knesset Constitution, Law and Justice Committee under Michael Eitan (b. 1944), drafted the sections on religion and state, and later co-authored two complete draft constitutions, one of them with Eitan. He also wrote the 2011 law requiring Israeli nonprofits to disclose funding received from foreign governments. Supporters called it transparency about foreign state intervention. Opponents said it stigmatized European-funded human rights groups while leaving private foreign donations to right-wing organizations in the dark. That objection would come back.

He founded the Kohelet Policy Forum in February 2012, in offices at Am Ve'olamo 8 in Givat Shaul, Jerusalem. It ran on Jewish nationalism, market economics, individual liberty, and a theory of representative government hostile to unelected authority. Avraham Diskin, Avi Bell and Eugene Kontorovich were among the principals.

The innovation was the delivery. Israeli parties and members of Knesset have thin research staffs. A think tank that arrives with a finished bill, a memorandum answering the objections, comparative material from six countries, a witness for committee, and a lawyer who will stay through the amendments has leverage that an essay does not have. Kohelet supplied the right with what it had never had: a bench. Its scholars contributed to the 2018 Basic Law declaring Israel the nation-state of the Jewish people, and Ayelet Shaked (b. 1976) drew on its network during her years as justice minister. A former Kohelet researcher became legal adviser to the Knesset's Constitution Committee. A Kohelet fellow became director general of the Education Ministry.

Koppel told the Jerusalem Post in 2024 how he scores this kind of work. A think tank wins when it converts a nonissue into an issue. Judicial reform, he said, had been nothing at all. Kohelet made it something.

On January 4, 2023, six days after Benjamin Netanyahu (b. 1949) assembled his coalition, Justice Minister Yariv Levin (b. 1969) announced a program in five parts. The attorney general's advice would stop binding the government, which could hire its own counsel. Administrative decisions could no longer be struck down on reasonableness alone. Basic Laws would not be subject to judicial review. Statutes could be struck only by a supermajority of a full panel, and the Knesset could override. The judicial selection committee would be reweighted toward elected officials.

Kohelet had been publishing on four of those five for a decade. Levin named Aviad Bakshi, head of Kohelet’s legal department, as one of the scholars he had consulted.

Koppel disliked the override from the start, and said so where he thought he was among friends. Sixty-one members of the Knesset could pass a law and then use the same sixty-one to immunize it. In February he told a closed meeting of young immigrants in Tel Aviv that it was a “stupid idea.” In March, Channel 13 aired leaked remarks in which he told a small crowd that Kohelet's advice to Levin and to Simcha Rothman (b. 1980) had been that the override is completely idiotic, and that it was in the package for coalition reasons.

By then the country was in the street, and Kohelet was on the map.

On March 9, 2023, about a hundred demonstrators from Brothers in Arms, the reservists’ organization, blocked the Givat Shaul entrance with sandbags and barbed wire. On March 15, protesters followed him through Jerusalem as he met a delegation from the Jewish Federations of North America. On March 22, Israeli expatriates and American Jews filled a street on the Upper West Side of Manhattan, a few blocks from where the Gerrer men had prayed, to protest his appearance at an Aish New York event. On April 2, six women from a feminist group entered the offices, and were arrested; the group said it was protesting Kohelet‘s positions on sexual assault, domestic violence, the poor and the LGBTQ community, and a Kohelet employee wrote that a woman had run at her and tried to hit her. On June 1, demonstrators came to his home in Efrat, in the Judean hills south of Jerusalem, over the Green Line, and posted flyers on his fence: “Here lives a collaborator with a dictator.” A neighbor turned a garden hose on them.

In the essay he wrote that autumn, Koppel recorded two more. A retired chief education officer of the IDF posted a call for his elimination, specifying that he meant the word “in the military sense.” A retired defense minister told him there would be blood in the streets, and that although the minister's own friends would be spilling it, the blood would be on Kohelet's hands.

His account of the meetings is the part of the record that will interest historians most, and it comes from one side only. He and his colleagues met pilots, magnates, former security chiefs, former ministers. He says his interlocutors almost never wanted to discuss the reform's contents or how to improve them. He says he could not make sense of what they were saying until he found the premise that made all of it coherent: that the wrong people had won the election and would keep winning, and that unless the court and the legal bureaucracy retained unlimited power, Israel's unelected institutions would fall into unfit hands. He named the premise the Barbarians at the Gates axiom.

He also produced a scene. Four leaders of the Kaplan Street protests came to Kohelet to demand that the forum renounce the reforms. Koppel and his colleagues said policy should be set by the elected government and legislature. One of the men, a magnate who led the chant of de-mo-krat-ya at rallies, asked whether Koppel seriously imagined his own political preferences should count no more than those of a taxi driver from Beit Shemesh. Koppel replied with the line from The Princess Bride about a word not meaning what the speaker thinks it means.

The story is his, told in Tablet in June 2026, with no name attached to the magnate and no corroboration offered. Take it as what it is: the memory of a participant with an argument to win. It is also the single most useful thing anyone has published about what those rooms were like, and nobody on the other side has published a competing account of the same meeting.

Meanwhile the compromise track ran. Koppel says opposition politicians refused to sit with Levin, and that he tried repeatedly to get them into a room. He and his counterparts in rival policy shops then negotiated privately for two months and reached detailed agreement on everything except the composition of the selection committee. The package went to President Isaac Herzog (b. 1960), who was hosting the national compromise effort. Herzog, Koppel says, was in close contact with sitting and retired justices, and returned with a proposal much closer to the hard opposition's position. It went nowhere.

Koppel's diagnosis of why compromise failed is the one honest thing in the essay that cuts against his own side's interest. So long as both camps feared a worse outcome, a deal was available. Once pressure on the government not to move without broad consensus made unilateral action unlikely, the opposition had no reason to trade. The people demanding consensus destroyed the conditions for it.

He finished the essay on October 6, 2023.

He did not look at it again for three months. When he did, he left the text alone and added a postscript.

His policy view had not moved. His sense of proportion had. He described it to the Jerusalem Post the following April, in his house in Efrat on a rainy day, as a married couple in the middle of a fight who get hit with a family health crisis and cannot believe what they had been shouting about.

“I don't think I was wrong, but it wasn't that important,” he said.

He said the reform had been far from perfect, that the opposition had had perfectly good reasons, that there was blame to go around for how the political process ran and that he took his share. His son and son-in-law had spent months in Gaza. Most of Kohelet's senior staff had been there too. He noted in the essay that his interlocutors' sons and his own were serving together, and that a little more grace on both sides would have gone a long way.

He argued in the same postscript that legislation had been the wrong instrument all along, because the court had by then struck down the one piece of the reform that passed and could strike down any future limit on itself. He argued that the war had wrecked the prestige of the institutions he opposed, and that prestige, not statute, is what determines who defers to whom. He borrowed Peter Turchin's (b. 1957) elite overproduction to explain the fury: a generation of Israeli magnates, retired generals and former ministers had been produced in numbers the country's top jobs could not absorb, and Netanyahu's longevity had blocked the rest.

The money went first. Arthur Dantchik, the Philadelphia investor who was Kohelet's principal benefactor, announced on August 4, 2023 that he had stopped giving to Israeli think tanks, saying the country needed to “focus on healing and national unity.” Protesters had been outside his house in Philadelphia; Koppel said afterward that he understood a man not wanting to pay for the privilege of being harassed. Kohelet's income fell from 30 million shekels in 2022 to 24 million in 2023 to 13.6 million in 2024. It let go dozens of employees, cut research, and stopped funding allied groups. Of the roughly four million dollars it took in during 2024, about three and a half million came through the Central Fund of Israel. In January 2025 the Knesset Finance Committee left it off the list of 138 nonprofits approved for donor tax credits while the Tax Authority examined claims that it engaged in partisan political activity; the authority found no basis for the claims, and the committee restored the status in March, with ten coalition members turning up to make sure the vote carried. Kohelet said more than ten thousand Israelis had recently given it money.

The organization that had once been sustained by one American now advertised a tax refund to small domestic donors.

Shani Granot-Lubaton, a New York-based leader of the protests and of the campaign against Dantchik, was given a right of reply by the Jerusalem Post. She said the Israeli public was troubled by Koppel's attempt to damage Israeli democracy and the rights of women, minorities and everyone else; that Dantchik had faced legal, determined and effective protest and nothing beyond it; and that Koppel's delicacy about being demonized had not extended to the campaigns run against liberal organizations in Israel. She said the protesters were defending Israeli democracy for Koppel's children too.

Then the thing he wanted arrived.

In January 2026 the coalition advanced a bill placing ministry legal advisers under the directors general appointed by ministers rather than under the attorney general. On July 15, 2026, after eleven hours of filibuster and nearly seventy committee sessions run by Rothman, the Knesset passed the central measure: the attorney general's written legal opinions cease to bind the government, which may declare that an opinion does not state the law, and may take its own position in court and hire private counsel if the attorney general will not argue it. Netanyahu was absent from the vote. Bezalel Smotrich (b. 1980) congratulated Rothman on a decision of enormous significance for Jewish identity, Zionism, security and settlement.

That is item one on Levin's list of January 4, 2023, close to word for word, and it is the item Koppel argued for most consistently and most publicly for a decade.

Note how it came. It passed by roughly a dozen votes as part of a bundle assembled with the Haredi parties, who supplied their votes after the coalition advanced a Basic Law on Torah study, froze arrests of draft evaders, and reversed changes to kosher certification. The vote nearly collapsed anyway when Moshe Gafni (b. 1952) held out over funding for ultra-Orthodox kindergarten teachers, and Rothman stretched a speech to buy time while Netanyahu put the funding item on the cabinet agenda. The provision splitting the attorney general's role was dropped. Within thirty days of the law taking effect the government must decide how future attorneys general will be appointed, which critics read as a route to removing Gali Baharav-Miara (b. 1959) without the professional committee that blocked the last attempt.

The law takes effect on January 1, 2027. The election is expected in late October 2026. The coalition that passed it may not be the government that lives under it.

Deputy Attorney General Gil Limon told the committee that once the bill passes the government will decide for itself what the law is. Petitions reached the Supreme Court the same night.

Koppel's own standard for this class of decision, stated in the essay he finished on October 6, 2023 and sharpened in a Kohelet piece in November 2025, is the veil of ignorance: constitutional arrangements should be designed without knowing which camp will hold power when they are used, because nobody knows what the coalition and the court will look like in thirty years. He applied the test to his opponents and asked whether they would grant the same powers to unelected institutions if Oranges ran them and Blues formed the government. He said we all knew the answer.

The test is available to be run in the other direction on a law passed by sixty-odd votes, in the last weeks of a Knesset, as part of a package traded for draft exemptions and kindergarten money, coming into force after the election. Koppel has not published on it as of this writing.

At Dicta, every claim has a held-out set. Vocalize the text and a scholar can check the vowels. Find the parallels and a Talmudist can open the two pages. Split Jeremiah from Ezekiel and count the twenty-six verses that went wrong. When the wires said his algorithm had proved the Torah had many authors, Koppel wrote three thousand words explaining that the number of authors was an input he had chosen, that the method could not have discovered it, and that the whole exercise was silent on the question everyone actually cared about. He gave away the strongest claim available to him because it was not supported.

At Kohelet there is no held-out set. A constitutional arrangement is run once, on one country, by the people who wanted it, against opponents who will not concede the result. There is no Jeremiah to shuffle with Ezekiel. The output cannot be scored and the error rate cannot be estimated, and the operator's choice of how many families to look for goes uncorrected, because nobody can run the experiment again with the other assumption.

His philosophy holds that a community stores knowledge in practices its members cannot state, and that the man who redesigns the practice from a diagram is likely to break something he never saw. His career has consisted of writing the diagrams. He built classifiers to extract what writers do not know they do, formal logic to describe what halakhists know before they can say it, language models to carry Hebrew that no grammar exhausts, and a policy shop to hand finished statutes to men in a hurry.

He argued that Israel's problem was a small, self-selecting body of credentialed people who converted professional expertise into political authority and could not be voted out. He was right that such a body existed. He answered it by building a small body of credentialed people, funded for a decade by one foreign billionaire, that converted professional expertise into political authority and could not be voted out.

Shimen would have had no idea what any of it was for. Heidi got her answer forty years late, in book form, from a man who had spent the interval learning to speak her language well enough to tell her why he would not adopt it. The state is the third party to the argument, and it does not read.

Notes

 Biography and career. Wikipedia, Moshe Koppel (born 1956; NYU Courant 1979 under Martin Davis; IAS postdoc; aliyah 1980; Bar-Ilan; Har Etzion; YU). Kohelet’s own bio page, Prof. Moshe Koppel. Yeshivat Har Etzion, Dr. Moshe Koppel.

The Princeton scene and Heidi. Benjamin Schvarcz, Acton Institute, Tradition: A Guide to Social Survival in the 21st Century. This is the source for the kosher dining hall, Heidi’s challenge, and the “slack-jawed” self-description, and also for the patriotism objection I quote. The reviewer is writing from Princeton’s James Madison Program, which he notes sits at the address of the former kosher dining hall.

Shimen. Book of Books review (Lodz, Auschwitz, wife and two children murdered, the handgun and the hidden children, Wiesel in the shtiebel, the “high-minded pieties” passage). First Things, “Mere Judaism”. Jewish Action review by Rabbi Yitzchok Adlerstein. Fathom review for the objection that Shimen has no surviving children. The blog that became the book was Judaism without Apologies; Elli Fischer’s Lehrhaus review coined “the Torah of the Kishkes.”

Authorship attribution. Koppel, Schler & Argamon, Computational Methods in Authorship Attribution, JASIST 2009. Koppel, Schler & Bonchek-Dokow, Measuring Differentiability: Unmasking Pseudonymous Authors, JMLR 2007 (the Torah Lishmah case, 524 vs 509 documents). Argamon, Koppel, Pennebaker & Schler, Automatically Profiling the Author of an Anonymous Text, CACM 2009. The 2003 press cycle: Charles McGrath, “Sexed Texts,” New York Times, August 10, 2003; Clive Thompson, “He and She: What’s the Real Difference?“, Boston Globe, July 6, 2003.

The Bible episode. Koppel’s own correction is the key document: Attribution and Misattribution: On Computational Linguistics, Heresy and Journalism, Seforim Blog, July 11, 2011. All the numbers I cite (200 synonym sets, 17% unclassified, 26 of ~2,200 misassigned, ~90% agreement with the Driver/Friedman consensus, no J/E at four or more families) are his. The underlying paper is Koppel, Akiva, I. Dershowitz & N. Dershowitz, Unsupervised Decomposition of a Document Into Authorial Components, ACL 2011. The Kherson Genizah paper is in Hebrew, Yeshurun 23 (2010), 559-566.

Talmudic work. Meta-Halakhah: Logic, Intuition, and the Unfolding of Jewish Law (Jason Aronson, 1996). Seder Kinnim (Hebrew). Higayon, co-edited with Ely Merzbach: journal page “Uncertainty Rules in Talmudic Reasoning,” with Dov Gabbay, History and Philosophy of Logic 32:1.

Dicta. Shmidman, Koppel & Porat, Identification of Parallel Passages Across a Large Hebrew/Aramaic Corpus (4,602 pairs, 1.8m words; the arXiv abstract says “just over 30 seconds,” the v2 paper text says 11 seconds, and I used 11 with the twenty-year brute-force estimate that appears in the paper). BEREL, DictaBERT, DictaLM, DictaLM 2.0, Dicta-LM 3.0. Organizational overview: European Association for Jewish Studies showcase.

Constitutional work and Kohelet’s founding. Wikipedia, Kohelet Policy Forum (founded 2012, Am Ve’olamo 8, Givat Shaul, Diskin/Bell/Kontorovich). The 2011 foreign-government funding disclosure law and the two constitutional drafts are documented in the Wikipedia biography with links to the Knesset committee archive.

The 2023 fight. Koppel’s own long account, which is where the compromise negotiations, the Herzog proposal, the Barbarians axiom, the Blues and Oranges, the Turchin argument, the “elimination” post, the “blood in the streets” threat, and the demonstration calling him a foreign implant all come from: The Long Road to Democracy, Kohelet, March 30, 2025, originally in Hashiloach. Note one small discrepancy: he writes there of demonstrators in New Jersey and of “44 years” in Israel; the documented March 2023 protest was on the Upper West Side of Manhattan (Jewish Press), and 1980 to 2023 is 43 years. They may be separate events; I kept the Manhattan one, which is sourced, and paraphrased the “foreign implant” line without a location.

The override remarks: Times of Israel, February 23, 2023 (“stupid idea,” closed meeting with young immigrants in Tel Aviv) and Times of Israel, March 26, 2023 (Channel 13 leak, “completely idiotic,” advice to Levin and Rothman). Levin’s citation of Aviad Bakshi is in the same pieces.

The protests: Times of Israel, March 9 blockade and April 2 office entry; JNS on the six arrests; Haaretz on the Breaking Walls action; Times of Israel, June 1, 2023, on the Efrat home protest (the flyer text and the garden hose).

The taxi driver scene. Moshe Koppel, “De-Mo-Krat-Ya!”, Tablet, June 5, 2026, part of a series tied to Tablet Studios’ The Battle for Israel’s Soul.

October 7 and after. Eliav Breuer, Kohelet’s Moshe Koppel reflects on the judicial reforms a year later, Jerusalem Post, April 13, 2024. This is the source for the couple-fight analogy, “I don’t think I was wrong, but it wasn’t that important,” the sons in Gaza, the think-tank-victory remark, the Dantchik harassment comment, and Shani Granot-Lubaton’s reply, which I paraphrased rather than quoted. Also useful: 18Forty interview, November 2024; Daniel Gordis’s two-part interview.

Money. Times of Israel on Dantchik, August 4, 2023; AP/PBS; Shomrim, July 21, 2025 for the 30m/24m/13.6m shekel figures and the Central Fund of Israel share; Times of Israel, January 20, 2025 and JNS, March 3, 2025 on the Section 46 fight and the ten thousand donors.

The 2026 endgame. Times of Israel, January 7, 2026 on the legal-advisers bill; Times of Israel, January 12, 2026 on the wider package; Ynet, July 15, 2026 for the law’s contents, the Haredi bundle, the Gafni holdout, the January 1, 2027 effective date, the thirty-day appointment clause, and Gil Limon’s line; Times of Israel, July 15, 2026 for the eleven-hour filibuster and Smotrich’s remarks.

The veil of ignorance. Stated in The Long Road to Democracy and sharpened in Koppel’s November 2025 Kohelet piece, “Restoring Balance to Government Requires the Veil of Ignorance.” Also Jerusalem Post op-ed, February 2025.

Jurispathic: Moshe Koppel and Robert Cover

In 1983 the Harvard Law Review asked Robert Cover (1943-1986) to write its annual Foreword on the Supreme Court’s term. He was forty, a professor at Yale, an observant Jew who read Talmud and taught law, and he had already written a book about antebellum judges who enforced the Fugitive Slave Act against their own convictions. He produced ninety pages called “Nomos and Narrative.” He died three years later, at forty-two.

The argument runs like this. We inhabit a normative universe, a nomos, and we hold it together with stories. Rules do not stand on their own. No set of legal prescriptions exists apart from the narratives that locate it and give it sense. Strip the story and the rule is noise.

Communities make law constantly. Cover called this jurisgenesis, and his claim was that it happens everywhere people share a narrative and take on obligations, in monasteries and unions and families and yeshivas. The state is late to this. It does not manufacture normative meaning. It arrives after meaning is already in production, in surplus.

He distinguished two patterns. The paideic is world-creating: a common body of texts, a shared discourse about how to read them, and interpersonal commitment strong enough to make the reading binding. Its natural home is the small group. It is thick, it educates, it obligates, and it fragments, because every serious reading generates further readings, and the community can hold them together only by teaching harder or by expelling people. The imperial is world-maintaining: thin, universalist, procedural, concerned with keeping distinct worlds from colliding. Its natural home is the state. Cover’s judgment on the imperial mode is severe. He called its virtues weak forces, system-maintaining, and said the organizing principle it supplies cannot itself produce the meaning that is life and growth.

Then the move that gave the essay its shock. Hart and Kelsen and Dworkin had all assumed a shortage of law, and had spent their careers explaining how judges make more of it in hard cases. Cover said the opposite. Law is in oversupply. The problem is fecundity. A hard case is hard because two litigants arrive with two valid nomoi, each intelligible inside its own narrative, and the court must pick. When it picks it does not create law. It destroys the other one.

“Judges are people of violence,” Cover wrote, and the office they hold is the jurispathic office. The law-killing office. He did not mean this as an accusation of bad faith. He meant it as a description of the job. Somebody has to reduce the multiplicity, and the court is the instrument, and what makes the court’s version stick is not that it is more correct but that it commands men with guns. His term for the state’s law is the social organization of law as power, which he set against the organization of law as meaning, and he thought the gap between the two was the permanent tragedy of the enterprise. In “Violence and the Word,” three years later, he put it without any cushioning: legal interpretation takes place in a field of pain and death.

The case he built the Foreword around was Bob Jones University v. United States, decided that term, in which the Court upheld the revocation of tax-exempt status from a Christian university whose religious doctrine forbade interracial dating. Cover was not defending the university’s racial doctrine. What held him was the amicus briefs from Mennonite and Amish groups who had no stake in the racial question at all and who filed anyway, because they understood what the case was about: whether a community may live by its own law when the state’s law says otherwise, and what the state may charge for the privilege. Cover’s complaint against the opinion was that it did the killing without saying so. He called it uncommitted, unwilling to put much on the line, an expression of the passing will of the state rather than a constitutional commitment the state was prepared to own.

Two ways a community can meet the state, in his scheme. Insular: turn inward, ask to be left alone, pay the price of separation. Redemptive: go out, try to remake the whole order in the image of your own nomos. His examples of the second were the abolitionists, Garrison against Douglass on whether the Constitution was a slave document or an antislavery one, both of them reading the same text out of a narrative the state did not share.

And the test that runs under all of it. A legal interpretation “cannot be valid if no one is prepared to live by it.” Meaning is cheap. Commitment is the thing that converts a reading into law.

I have not found an English-language treatment that runs it over the Israeli constitutional fight of the last three years, which is strange, because the fight is the best specimen the theory has been offered since Cover wrote, and because the man at its intellectual center has spent forty years describing jurisgenesis without ever using the word.

Moshe Koppel published Meta-Halakhah in 1996. Its thesis is that Jewish law works the way a natural language works. A grammar describes a language without generating every judgment a competent speaker makes. Native speakers know a sentence is wrong before they can name the rule it breaks, and accomplished halakhists recognize that an argument does not fit the tradition before they can say which principle it violates. The intuition is competence, built by immersion in cases, and it exceeds any codification of itself.

That is a theory of jurisgenesis. The law lives in a community’s practice and its stories, the written corpus underdetermines it, and the people who carry it know more than the code contains.

Judaism Straight Up, twenty-four years later, is the same claim scaled up to a society. Koppel sets a Gerrer Hasid he calls Shimen, a survivor of Lodz and Auschwitz who prayed in a Manhattan shtiebel with Koppel’s grandfather, against a Princeton graduate student he calls Heidi. Shimen inhabits a thick weave of ritual, food restrictions, kinship rules, commercial norms, communal memory and specific obligation. His Judaism is a first language and he cannot give a philosophical defense of most of it. Heidi holds a single overriding principle, avoid harm to any person, learned as a rule and applied as a rule. Koppel’s argument is that Shimen’s community can reproduce itself and Heidi’s cannot.

Read that through Cover and it stops being a polemic about Jews and becomes a proposition about law. Shimen’s shtiebel is a paideic nomos: corpus, discourse, commitment. Heidi’s harm principle is imperial: a world-maintaining rule that tells you how to avoid collisions without telling you what to build. Koppel’s charge is that the imperial mode cannot generate normative meaning on its own and is parasitic on paideic communities it did not make and cannot make. That is Cover’s own position, stated on page sixteen of the Foreword, in 1983, by a Yale law professor who would have found Koppel’s politics uncongenial.

Neither man cites the other. Cover was dead before Koppel wrote a word on any of this. Koppel does not seem to have read him.

The machines belong here too. Koppel’s authorship attribution works by finding the residue that a community deposits in a writer without the writer’s knowledge: pronouns, prepositions, suffixes, the rate at which grammatical forms recur. His method for locating parallel passages across the Babylonian Talmud, which found 4,602 of them in 1.8 million words in eleven seconds, is a device for tracing how a nomos propagates itself through repetition and rephrasing. When his group’s clustering algorithm split the Torah along roughly the same line the biblical critics had drawn, he wrote three thousand words on a blog explaining that the number of authors was an input he had chosen and the method could not have discovered it, and that the result was silent on divine authorship because nobody has a theory of what a divine text looks like. He gave away the strongest claim available to him because his data did not support it.

The doctrines Koppel spent a decade attacking were forged in a single case, and the case was about whether the state may let a community live by its own law.

In 1986 Yehuda Ressler, an attorney and reserve officer, petitioned the High Court of Justice against the defense minister’s blanket deferment of military service for full-time yeshiva students. Earlier petitions on the same subject had been thrown out twice over, for lack of standing and for non-justiciability. Ressler came with an affidavit from a former head of the army’s manpower planning branch stating that he and his fellow petitioners served longer because the yeshiva students did not serve at all.

Aharon Barak (b. 1936) wrote the judgment, delivered in 1988. He granted standing to a petitioner with no personal grievance beyond membership in the public, and he declared that everything is justiciable, including matters of war and peace. Those two holdings are the foundation of everything that followed: the court’s reach into appointments, security, immigration, budgets, religion. On the merits Ressler lost. The deferment survived, within the zone of reasonableness, with a warning attached that if the numbers kept climbing the answer might change.

Cover would have recognized the structure at once. A paideic community had an arrangement with the state, an arrangement that had never been legislated. It began as an understanding between David Ben-Gurion (1886-1973) and the rabbinic authority known as the Chazon Ish, Avraham Yeshayahu Karelitz (1878-1953), covering roughly four hundred students, so that a Torah world destroyed in Europe could be rebuilt. It was a covenant between two nomoi and it existed nowhere in the statute book. The whole subsequent litigation, from Ressler through Rubinstein in 1998, the Tal Law in 2002, the Movement for Quality Government in 2006, Ressler again in 2012, has been the state’s effort to decide whether that unwritten covenant counts as law, and if so at what price.

The whole subsequent litigation, from Ressler through Rubinstein in 1998, the Tal Law in 2002, the Movement for Quality Government in 2006, Ressler again in 2012, has been the state’s effort to decide whether that unwritten covenant counts as law, and if so at what price.

Barak‘s later work extended the jurispathic office in every available direction. Reasonableness, from 1981, let the court substitute its own weighting of considerations for a minister’s whenever it judged the weighting wrong. In 1993 he held that the attorney general‘s advice binds the government and that the attorney general is free to argue against the government’s position in court, which put a jurispath inside the executive. In 1995 the court held that ordinary statutes could be struck for conflicting with Basic Laws. In 2019 it agreed to hear a challenge to the validity of a Basic Law, and on January 1, 2024, sitting with all fifteen justices for the first time in Israeli history, it voided one, eight to seven, with twelve of fifteen affirming the power in principle. The amendment it voided was the one that had abolished reasonableness.

Read Koppel’s program against that list and it resolves into a single project. Levin’s five items of January 4, 2023 are, item by item, a contraction of the jurispathic docket. Narrow standing: fewer occasions to kill. End reasonableness: remove the instrument that does most of the killing. Immunize Basic Laws: place certain meanings beyond killing. Supermajority for striking statutes: raise the cost of each killing. Unbind the attorney general: disarm the jurispath inside the government.

Barak called the system substantive democracy, and the phrase claims that the court is not choosing among valid nomoi at all, that it is finding the single correct one, that its readings follow from democracy’s nature rather than from the court’s preference. That claim is what Cover set out to demolish. Cover’s target was the pretense that a court’s interpretation is uniquely correct rather than uniquely enforced, and that what a judge does when he prefers one reading is discovery.

Koppel is thus half a Coverian without knowing it. He has the diagnosis. He has spent a career documenting the paideic. What he does not have is Cover’s conclusion, and the reason is that Cover’s conclusion cuts the other way.

Cover’s communities were weak. Amish, Mennonites, a small Bible college. They asked the state to leave them alone. His preferred settlement was an imperial power modest about its own paideia, holding the ring, cashing out its virtues in freedom of association, declining to impose a story of its own. He worried about a statist paideia because a state committed to its own meaning would use violence to destroy the germ of meanings alternative to those of the power wielders. His famous close was an injunction to stop circumscribing the normative universe and to let new worlds in.

Robert Post (b. 1947), reading the Foreword twenty years later, put his finger on the gap. A community can be insular and turn from the state, or redemptive and try to capture it. But once it holds the levers of government, it seems to lose its association with nomos, because the law it now speaks emanates from a field of pain and death. Cover never worked this out. He wrote as if the paideic communities would always be the ones being crushed.

Koppel’s community is redemptive and it won.

The traditionalist coalition he calls Orange, the amalgam of Revisionists, Sephardi traditionalists, national religious and Haredim, has been winning Israeli elections for most of half a century. It did not ask to be left alone. It went for the state. Koppel built it the instrument it lacked, a policy shop that arrives with finished statutes rather than complaints, and the instrument worked. And on the far side of that victory the roles invert. A nomos that holds the army and the tax authority and the cabinet is no longer the insular community pleading for room. It is the imperial power, and the jurispathic office does not disappear when a court is disarmed. It moves.

The week of July 13, 2026 shows where it went.

On Monday the Knesset passed Basic Law: Torah Study, sixty-three to fifty-two, enshrining Torah study as a foundational value of the state. In Cover’s terms this is a paideic community writing its narrative into the constitutional layer, so that a court reaching for its practice must reach past a Basic Law to get there. On Tuesday the Knesset passed an amendment to the Defense Service Law, fifty-eight to fifty-four, suspending arrest, investigation and criminal proceedings against qualifying yeshiva students through November 30, 2026, and directing the army to set up a committee to decide who qualifies. This is a suspension of the state’s violence against one community, granted by the state, on the community’s terms. On Wednesday the Knesset passed the attorney general law, ending the binding force of the office’s legal opinions and letting the government determine for itself what the law is and how it will be argued in court. Nearly seventy committee sessions had gone into it. The opposition filibustered for eleven hours and so, at the end, did the coalition.

The three were one package. The Haredi parties supplied their votes for the third in exchange for the first two, and the vote nearly collapsed anyway when Moshe Gafni held out over funding for ultra-Orthodox kindergarten teachers, so Simcha Rothman (b. 1980) stretched a speech at the podium while Benjamin Netanyahu put the kindergarten item on the cabinet agenda.

Within days Justice Ofer Grosskopf issued a provisional order freezing the arrest suspension and a conditional order requiring the Knesset and government to say why it should not be struck down, citing the implications of halting enforcement against only certain sections of the population. The jurispathic office, still holding the pistol, in the middle of the transfer.

Cover’s test is commitment, and the draft fight is the only place in this story where it is being paid at scale.

The unanimous nine-justice ruling of June 25, 2024 held that with the exemption framework expired there was no legal basis for the blanket deferment or for subsidizing the institutions whose students claimed it, and that the state must act to conscript them. That is the jurispathic office at full extension: the court killing an arrangement that had governed a community’s life for seventy-six years.

By early June 2025, of some nineteen thousand summonses issued, about nine hundred and ninety-six men presented themselves at induction centers and two hundred and thirty-two were conscripted. Roughly one in a hundred. Tens of thousands of young men, most of them poor, most of them with no political office and no lawyer, declined to answer the state, and their families absorbed the cut in yeshiva funding, and a portion of them accepted arrest.

By Cover’s standard that is law. Cover’s test asks who is prepared to live by an interpretation, and it registers that a reading nobody will suffer for is not yet law whatever the reporter says. On that measure the yeshiva world’s reading of its own obligation has more purchase in the world than the reading handed down by fifteen justices sitting en banc.

Which is Cover’s point about courts, made from the inside. The court commands violence. It does not command belief. When the two come apart the court can go on issuing orders into a nomos that will not receive them, and this is the position the Israeli Supreme Court has occupied since June 2024.

What Koppel achieved, then, is a change of venue.

A government empowered to declare that the attorney general’s written opinion does not state the law has acquired the office Cover described. It can now select among competing legal meanings and enforce the one it prefers.

A court that kills a nomos must publish reasons. It must hear the other side first. It must write in a form that binds it next time. It must do the killing in daylight, under a name, in a document its critics can quote back. Cover said these were thin protections. He was writing about a court whose violence wore the costume of interpretation, and the costume was what disturbed him, because it let judges believe they were reading when they were choosing.

A cabinet resolution declaring what the law is wears less. Under the new statute the government reports its declaration to a Knesset committee.

Cover’s worry about a statist paideia was that a state committed to its own meaning would use force to crush the meanings that compete with it. He assumed the meaning in question would be the secular liberal one, because in 1983 in New Haven that is what a state paideia looked like. The Israeli case runs the experiment with the other side holding the machinery. The law takes effect on January 1, 2027. The election is set for October 27, 2026. Whichever coalition wins will inherit an executive that can decide for itself what the law is, and there is no reason internal to the statute why that coalition should be the one that wrote it.

Koppel calls this the veil of ignorance: institutional arrangements should be designed without knowing which camp holds power when they are used, and he applied the test to his opponents and asked whether they would grant the courts these powers if the Oranges ran them and the Blues formed the government. He said everyone knew the answer. He has not, as of this writing, published the same test on a law passed by roughly a dozen votes in the last days of a Knesset, traded for draft exemptions and kindergarten money, coming into force after the election.

Cover has no solution. He could not explain how plural nomoi coexist once you concede that all of them are jurispathic, that every community sustains its world by excluding the others. Post’s objection stands: Cover would not admit that liberalism is a nomos with its own pieties, because admitting it would give liberalism a priority he did not want to grant, and the price of the refusal was that he could never say how the worlds are supposed to live together beyond the weak virtues of an empire he had already declared sterile. He offered no theory of persuasion. He filled the space between communities with conflicts settled in blood, and then he died, and the Foreword ends on an invitation rather than an argument.

That gap is the size of the Israeli problem. Two nomoi, both real, both jurisgenerative, both willing to pay, sharing one army and one tax base and one strip of land, with no umpire either will accept.

The court under Barak killed law, which was its office, and dressed the killing as discovery, which was its vice. Koppel documented a paideic world, built machines to make its jurisgenesis visible, and then spent his authority moving the killing office from one set of hands to another. The yeshiva world, having lost every case, kept its law by declining to appear.

Cover’s last instruction was to stop drawing the boundary so tight and let new worlds in. Israel has let two of them in, and they are standing on the same floor.

Notes

 The frame. Robert M. Cover, “The Supreme Court, 1982 Term — Foreword: Nomos and Narrative,” 97 Harv. L. Rev. 4 (1983). Free copy in Yale’s Faculty Scholarship Series: digitalcommons.law.yale.edu. Page cites for what I used: the nomos and the inseparability of law and narrative at 4-5; jurisgenesis at 11 following; paideic and imperial at 12-13; the imperial virtues as weak, system-maintaining forces at 12 and as incapable of producing meaning at 16; law as power against law as meaning at 18; the oversupply of law and the court as its solution at 40; commitment and the validity of an interpretation nobody will live by at 44; judges as people of violence and the jurispathic office at 53; the critique of Bob Jones at 66-67; the warning about a statist paideia at 61-62; the closing invitation at 68. Also Robert M. Cover, “Violence and the Word,” 95 Yale L.J. 1601 (1986), for the field of pain and death at 1601-02.

Reading Cover. Robert C. Post, “Who’s Afraid of Jurispathic Courts? Violence and Public Reason in Nomos and Narrative,” 17 Yale J.L. & Human. 9 (2005). This is the source for the insular-versus-redemptive point I lean on hardest, that a community which captures the state seems to lose its association with nomos, and for the objection that Cover would not concede liberalism is itself a nomos. The 2005 Yale Journal of Law and the Humanities symposium “Rethinking Robert Cover’s Nomos and Narrative” also contains Judith Resnik, “Living Their Legal Commitments: Paideic Communities, Courts, and Robert Cover,” 17 Yale J.L. & Human. 17 (2005). On Cover’s Jewish sources: Samuel J. Levine, “Halacha and Aggada: Translating Robert Cover’s Nomos and Narrative,” 1998 Utah L. Rev. 465, and Levine, “Law and Redemption,” 34 Yale J.L. & Human. 253 (2023). Cover’s “Obligation: A Jewish Jurisprudence of the Social Order,” 5 J. L. & Religion 65 (1987), argues that Jewish law starts from obligation where American law starts from rights, which is Koppel’s Shimen-and-Heidi thesis stated by a Yale professor thirty-three years earlier; I left it out of the body to keep the essay on the Foreword, but it belongs in any longer treatment.

Bob Jones. Bob Jones University v. United States, 461 U.S. 574 (1983). On the Mennonite and Amish amicus briefs Cover discusses, see Peter Leithart’s summary at Theopolis, which quotes the brief directly.

Novelty. I searched for English-language work applying Cover to the 2023-2026 Israeli fight and found none. I found Cover applied to Kiryas Joel, to Ghanaian courts and Islamic law, to environmental law, to political constitutionalism generally, and to religious liberty in New York, and I found the Foreword anthologized in The New Jewish Canon (2020) with a commentary by Christine Hayes. That is not proof of absence. Someone should check Hebrew-language scholarship, and Menachem Mautner’s Law and the Culture of Israel (Oxford, 2011) is the obvious place where a Coverian reading might already exist under another name.

Ressler and the jurispathic expansion. HCJ 910/86 Ressler v. Minister of Defence, 42(2) PD 441 (1988). English translation at the Cardozo Israeli Supreme Court Project, which is the best free English archive of Israeli constitutional decisions and worth bookmarking. The subsequent line: HC 3267/97 Rubinstein v. Minister of Defense (1998); the Tal Law, 2002; HCJ 6427/02 Movement for Quality Government v. Knesset (2006); HCJ 6298/07 Ressler v. Knesset (2012). Koppel’s own account of the doctrinal history, which I checked against the cases and found accurate on the facts if not on the tone, is in “The Long Road to Democracy”.

The reasonableness ruling. HCJ 5658/23 Movement for Quality Government v. Knesset, decided January 1, 2024, translated at Versa. Fifteen justices en banc for the first time; twelve of fifteen affirmed the power to review Basic Laws; eight of fifteen voided Amendment No. 3. Commentary from both directions: IDI and Verfassungsblog.

The draft. HCJ 6198/23 Movement for Quality Government v. Minister of Defense, June 25, 2024, nine justices, unanimous. Summary from the Law Library of Congress and from IDI. The enforcement figures I cite, roughly 19,000 summonses, 996 reporting, 232 conscripted by early June 2025, are from IDI’s one-year review, which also gives the roughly 480 million shekels in annual yeshiva funding affected. Treat these as one organization’s count; the Defense Ministry publishes its own numbers and they differ.

The July 2026 week. Basic Law: Torah Study passed 63-52 on Monday, July 13 (Jerusalem Post). The Defense Service Bill (Amendment No. 28) suspending arrests through November 30, 2026 passed 58-54 on Tuesday, July 14 (Israel365, Unpacked). The attorney general law passed Wednesday, July 15 (Ynet for the contents, the coalition package, the Gafni holdout, the January 1, 2027 effective date and the thirty-day clause on future appointments; Times of Israel for the eleven-hour filibuster). Justice Ofer Grosskopf’s provisional and conditional orders freezing the arrest suspension: Jerusalem Post. Election set for October 27, 2026.

Vote counts, unresolved. Ynet reported the attorney general law at 61-51, the Times of Israel at 65-51.

Koppel. Meta-Halakhah: Logic, Intuition, and the Unfolding of Jewish Law (Jason Aronson, 1996); Judaism Straight Up: Why Real Religion Endures (Maggid, 2020); “Attribution and Misattribution: On Computational Linguistics, Heresy and Journalism,” Seforim Blog, July 11, 2011; Shmidman, Koppel & Porat, “Identification of Parallel Passages Across a Large Hebrew/Aramaic Corpus,” J. Data Mining & Digital Humanities (2018). His veil-of-ignorance argument is in “The Long Road to Democracy” and in his November 2025 Kohelet piece on restoring balance.

Cover’s insular communities were powerless and his sympathy for them is doing more work in the Foreword than his analysis is. Applying him to a community that wins elections requires the extension Post makes rather than anything Cover wrote, and I have marked that in the essay but a hostile reader will press on it. Second, Cover’s refusal to treat liberalism as a nomos means the Israeli court comes out of this frame looking more purely instrumental than it is; a fair account would concede that the court’s justices believe things and that their beliefs constitute a narrative world, which is Post’s correction and which I have absorbed rather than argued. Third, everything I know about the negotiating rooms of 2023 comes from Koppel.

The Elite That Built the Court, and the One That Unbuilt It

In 2004 Harvard published a book by a comparative constitutional lawyer named Ran Hirschl arguing that the global spread of judicial review has almost nothing to do with a commitment to rights. The book is Towards Juristocracy: The Origins and Consequences of the New Constitutionalism, and it studies four cases: Canada, New Zealand, South Africa, and Israel. His hegemonic preservation thesis goes like this.

Constitutionalizing rights transfers power from parliaments to courts. Politicians do not ordinarily give away power. They give it away under one condition: when they can see that they are going to lose control of the electoral process. At that moment a ruling group can calculate that its policy preferences will survive better inside an institution it already staffs than inside a legislature it is about to lose. So it entrenches. It writes a bill of rights, hands enforcement to a court whose appointment process it controls, and calls the result a triumph of universal values.

Hirschl builds the argument around three groups acting together, each for its own reasons. Political elites facing erosion from peripheral constituencies. Economic elites who understand that a jurisprudence of individual rights and negative liberty protects property whether or not property is named in the text. And judicial elites, whose interest requires no elaborate account: prestige and jurisdiction. The three converge, the constitution appears, and the timing is explained by the convergence rather than by anyone’s moral awakening.

Israel is paradigmatic. The Knesset was for decades the instrument of an Ashkenazi secular establishment that saw no reason to delegate anything to judges. By the mid-1980s that establishment was losing the electoral arithmetic to peripheral constituencies, above all religious and traditional Mizrahi voters. At the same moment Israel was executing a turn toward global neoliberal integration. A cross-party coalition of secular and market interests, with judicial and economic backing, produced Basic Law: Human Dignity and Liberty and Basic Law: Freedom of Occupation in 1992. Three years later, in the Mizrahi Bank decision, the Supreme Court held that the Basic Laws bind the legislature and that it could strike statutes conflicting with them. Aharon Barak named this the constitutional revolution.

The second half of Hirschl’s book is the part almost nobody quotes and it is the half that matters here. He examined what the four new bills of rights produced. Between eighty and ninety percent of high court rights litigation in his four countries concerned negative liberties and procedural fairness, and that litigation succeeded. Distributive claims barely moved. In Canada and Israel, rights litigation was used to strip protections unions had won through politics. Inequality rose in Israel, Canada, and New Zealand after their bills of rights, and held roughly steady in post-apartheid South Africa. His conclusion is that the new constitutionalism delivered a great deal to people who already had things and close to nothing to people who did not, and that the deeper trend it belongs to is the removal of distributive questions from electoral politics into professional bodies: courts, central banks, trade tribunals.

Two decades later, a computer scientist in Jerusalem produced the same first proposition from inside the fight, and then his side won.

Moshe Koppel, founding chairman of the Kohelet Policy Forum, finished an essay on October 6, 2023 that he later published as “The Long Road to Democracy.” Its central section names what he calls the Barbarians at the Gates axiom: that the wrong people won the election, will keep winning elections, and will destroy everything unless the court guarantees that the right people continue to control the state’s unelected institutions and that those institutions retain disproportionate power. He arrived at it, he says, by sitting through a year of meetings with opponents of judicial reform whose arguments made no sense to him until he found the premise that made all of them cohere.

He divides the country into Blues, the heirs of the secular Labor Zionist establishment who still hold the army’s upper ranks, the licensed media, the justice system, the state-funded universities, the professional guilds and the public unions, and Oranges, the amalgam of Revisionists, traditional Sephardim, national religious and Haredim who keep winning at the ballot box. His summary is that the Blues control the institutions and the Oranges win the elections.

That is hegemonic preservation restated by a participant. Koppel does not cite Hirschl and gives no sign of having read him. The two men are making claims about different moments.

Hirschl’s claim is about the founding. He says a threatened elite entrenched its preferences in a court in 1992 because it foresaw electoral defeat. Koppel’s claim is about the defense. He says the same elite fought in 2023 to keep the court’s powers because it fears the electorate. Those are separate propositions and the second does not follow from the first. A group can build an institution for self-interested reasons and later defend it out of conviction. A group can also inherit an institution it did not build and defend it for reasons that have nothing to do with why it exists. Koppel needs the 2023 claim and Hirschl supplies only the 1992 one, and the gap between them is thirty-one years and two generations of personnel.

What Hirschl supplies that Koppel has never used is the consequences chapter. If you want evidence that the Israeli court’s constituency was narrow, it is Hirschl’s finding that the new constitutionalism moved negative liberties and left distribution alone, that inequality rose after 1992, and that rights litigation was turned against organized labor. The Mizrahi and Haredi periphery whose electoral ascent Hirschl says triggered the entrenchment got a court that protected speech and due process and did not touch the wage structure or the housing market or the transfer payments.

Koppel cannot fire this ammunition. Hirschl’s critique of juristocracy is a critique from the left. Its complaint is that the court served capital. Kohelet’s economics is deregulatory and market-liberal, which is to say it is the program of the second member of Hirschl’s three-part coalition. On Hirschl’s account the economic elite was the court’s silent partner, and Kohelet’s answer to the periphery’s grievance is more of what the periphery already received. This is the internal tension the thesis produces when you apply it to its own paradigm case and then ask who inherits the argument.

So: does the Israeli case confirm Hirschl, and does 2026 confirm or embarrass him?

Take confirmation first.

Hirschl’s mechanism carries an expiry date. If a court’s authority rests on a bargain struck by a declining group, then the authority is only as durable as the group’s residual influence over appointments, prestige, and the professional culture of the bar. Complete the decline and the bargain has no political base. On that reading, everything from January 2023 forward is its terminal phase.

Moshe Cohen-Eliya made this argument in Law & Liberty in March 2024, applying Hirschl to the January 1, 2024 reasonableness judgment. He noted that two of the eight justices in the majority, Chief Justice Esther Hayut (b. 1953) and Justice Anat Baron, had retired from the court in October 2023 and issued their opinions within the statutory window for post-retirement rulings, so that by the time the judgment appeared the majority that produced it no longer sat. Requests from moderates in politics and the academy to hold the ruling until after the war were refused. Cohen-Eliya calls this “a declining hegemony willing to maintain power by constraining a future majority.” It is Hirschl’s mechanism operating at the level of a single docket entry.

The judgment struck down an amendment to a Basic Law for the first time in Israeli history, eight to seven, with twelve of fifteen justices affirming the power in principle. A court that asserts authority over the constitutional layer is a court that has run out of ordinary tools, and Hirschl’s model predicts this trajectory: as the founding coalition’s electoral position deteriorates, the institution it endowed has to reach further to hold the line, and each reach costs it legitimacy with the majority it is holding off.

Now the embarrassments, which are larger.

The first is old and was aimed at the book on publication. Carlos Closa, reviewing Towards Juristocracy in this journal’s pages in 2006, pointed out that Hirschl offers “no empirical evidence for what motivates the relevant actors.” Nobody was asked. There are no interviews, no minutes, no letters, no contemporaneous texts in which anyone says they are entrenching to preserve hegemony. Motive is inferred from group membership and from outcome, which is a structural argument wearing intentionalist clothes.

Basic Law: Human Dignity and Liberty passed with thirty-two votes in favor and twenty-one against, in a chamber of one hundred and twenty. Roughly a quarter of the Knesset was present to enact what the court would later treat as a bill of rights, and Barak Medina reports that there were doubts during the legislative process about whether the thing would constrain the legislature at all. Support was bipartisan, which is what Hirschl’s cross-party elite coalition predicts. But a thirty-two vote enactment in a near-empty chamber is also what inadvertence looks like, and the court’s 1995 decision to treat it as constitutional was made by judges, not by the coalition Hirschl describes. On the available record you cannot distinguish an elite strategy from a legislative accident that an ambitious court subsequently converted. Hirschl asks us to read intention off a result, and the result had two authors.

The second embarrassment is the scale of the 2023 defense. Hegemonic preservation is a theory about elites. It predicts that when a court built by a declining group is attacked, that group’s remnant will defend it: the bar, the retired generals, the tech sector, the newspapers. All of that happened, and Koppel documents it with feeling, including a retired defense minister who told him there would be blood in the streets and a retired chief education officer of the army who posted a call for his elimination.

But Medina’s figures describe something the model has no slot for. Demonstrations ran for more than forty consecutive weeks. Up to two hundred and fifty thousand people gathered in central Tel Aviv on a given Saturday night with comparable numbers elsewhere, between three and seven and a half percent of Israel’s adult population in the street each week, and about twenty-two percent reporting attendance at least once during 2023. Scaled to the United States that is roughly twenty million people weekly for forty weeks. Whatever else it was, it was not a professional guild protecting its franchise. Either the theory needs an account of how an elite institution acquires a mass constituency, or the court had one for reasons the theory does not model.

The third embarrassment is the shape of the reversal, and it is the one that should interest people who work on judicial power comparatively, because it runs the mechanism backward and gets a result no strategic model predicts.

On July 15, 2026, in the final days of a Knesset facing automatic dispersal ahead of an October 27 election, the coalition passed the law ending the binding force of the attorney general’s legal opinions. The government may now declare that a written opinion of the attorney general does not state the law, may take its own position in court, and may hire private counsel if the attorney general declines to argue it. Yariv Levin and Simcha Rothman carried it through nearly seventy committee sessions and an eleven-hour filibuster. Benjamin Netanyahu was absent from the vote. Item one on the January 2023 list, enacted alone, three and a half years late.

Every strategic account of judicial power, Hirschl’s included and Tom Ginsburg’s insurance model even more so, says that a group facing electoral uncertainty wants strong courts, because courts are what protect you when you lose. This coalition, weeks from an election polling badly, weakened the courts. That is the opposite of insurance and the opposite of preservation. It admits three readings. The coalition expected to win. The coalition was not behaving strategically. Or the drivers were particularistic in ways the models exclude.

The third is almost certainly right, and it is the finding.

The law came bundled. The Haredi parties supplied the votes for it after the coalition delivered, on the Monday, Basic Law: Torah Study, sixty-three to fifty-two, and on the Tuesday, a suspension of arrest, investigation and prosecution of yeshiva draft evaders through November 30, 2026, fifty-eight to fifty-four. The Wednesday vote nearly collapsed when Moshe Gafni held out over funding for ultra-Orthodox kindergarten teachers. Behind all of it sits a prime minister under criminal indictment whose conflict-of-interest arrangement kept him out of the chamber.

Hirschl’s model has no place for one man’s trial and no place for a sectoral side-payment on conscription. It models classes of actors pursuing durable positional interests. What moved Israeli constitutional law in July 2026 was a defendant, a draft exemption, and a kindergarten subsidy. A structural theory that cannot see those things is not wrong about 1992. It is silent about 2026, which is a different and more serious problem for a theory that claims to explain timing.

There is a fourth objection, and it comes from the strongest recent defense of the court, which is also the reason this argument is live.

Barak Medina of the Hebrew University published an article in the International Journal of Constitutional Law on May 21, 2026 that rejects the hegemonic preservation reading of the Israeli court by name, citing Hirschl and Ruth Gavison (1945-2020) in a footnote. His argument is doctrinal.

Medina says the Basic Laws bind the Knesset not because the Knesset enacted them but because of a cumulative effect: partial procedural legitimacy from the Constituent Assembly’s delegation, plus content that entrenches an unwritten constitution Israel already had. On that reading the court’s power is calibrated by a sliding scale. An amendment consistent with the unwritten constitution passes on a bare majority. An amendment contradicting it needs overwhelming and bipartisan support. And an amendment enacted for the current government’s own benefit is an abuse of the constituent power, void whatever its majority.

The indicia the court uses to detect that abuse were written before July 2026. Does the amendment address a specific interest of the sitting government? Does it take immediate effect? Was it supported only by coalition members? Was it enacted through an expedited process? The court applied these to a 2016 two-year budget amendment, to the 2021 Shafir case, to the January 2024 incapacity amendment that had been passed to shield Netanyahu from a conflict-of-interest petition, and to the reasonableness amendment.

Medina’s answer to Hirschl follows from this. The court requires that a change in constitutional identity be made explicitly, in writing, with substantial and persistent popular support, and his contention is that the 2023 program sought to change that identity without ever amending the documents that state it, because the votes for an honest amendment were not there. He also argues, sourcing it to the coalition’s own published policy guidelines, that the reform’s underlying purpose included enabling the disqualification of Arab parties, discrimination against women and the Arab minority, and the enforcement of religious norms in public.

Whether this answers Hirschl depends on a question the two men would answer differently. Medina’s refutation consists of the court’s own doctrine about the limits of its own power. Hirschl’s method treats such statements as the output to be explained rather than as evidence. Neither position can be settled by the doctrine, and neither can be settled by the outcome, which is why the argument is still running twenty-two years after the book.

The attorney general law does not take effect until January 1, 2027, more than two months after the election. Hirschl’s model has been applied almost exclusively to the creation of judicial power. Run it on dismantling and the delayed effective date becomes the interesting datum, because it reads in opposite directions in the two available frameworks.

Under Medina’s indicia, delayed effect is evidence of legitimate purpose. A government entrenching a rule for its own immediate benefit wants the rule now.

Under a Hirschlian reading, delayed effect is what a group does when it has priced its own mortality. A coalition that expects to be a minority does not need the executive freed from the attorney general during its own remaining weeks. It needs the rule to be in force for whoever governs next, and it needs to reduce the chance that a court freezes it while the coalition that passed it still holds office. That is entrenchment against a future majority, executed by an outgoing one, through legislation rather than through a court. Which is hegemonic preservation with the instrument swapped.

I do not know which reading is right and neither does anyone else, because the evidence that would settle it is evidence about what a dozen people intended, and that is the evidence Closa noted was missing from Hirschl’s book in 2006 and is still missing now. The same fact, one clause about commencement dates, is exculpatory in one framework and incriminating in the other, and the frameworks cannot adjudicate between themselves.

So the accounting.

Hirschl’s thesis survives the Israeli case on the way up. The timing fits, the coalition he describes existed, the 1992 enactments and the 1995 decision came at the moment the old establishment’s electoral arithmetic failed, and his consequences chapter explains, better than any defense of the court has explained, why the periphery never developed loyalty to an institution that protected liberties it could not use.

It does not survive on the way down. It cannot account for a fifth of an adult population defending the court in the street. It cannot account for a coalition weakening courts on the eve of an election it might lose. It has no variable for a criminal defendant at the head of a government or for a religious party trading its votes on judicial power against its sons’ conscription. And its central evidentiary weakness, the absence of any direct evidence about motive, is not a gap that thirty years of subsequent events has filled. It has widened, because we now have two moments to explain and the same inferential shortcut has to serve both.

Koppel’s victory therefore neither confirms nor refutes Hirschl. It marks where the theory stops. Hegemonic preservation is a good account of why elites build courts and a poor account of why anyone dismantles them.

The useful research question is whether the mechanism runs in reverse. Do declining majorities de-constitutionalize on the same logic by which declining minorities constitutionalized? If they do, the January 1, 2027 commencement date is the tell, and comparative constitutional law has a symmetry it has not documented. If they do not, then the events of July 2026 belong to a literature about personal legal jeopardy and coalition side-payments that has yet to be written, and the reason the Israeli case looks so much like theory on the way up and so little like theory on the way down is that the way up was structural and the way down was not.

Notes

The frame. Ran Hirschl, Towards Juristocracy: The Origins and Consequences of the New Constitutionalism (Harvard University Press, 2004; paperback 2007). Page references for what I used, taken from Daniel Moure’s book note (below) and checked against the reviews: the critique of evolutionist and functionalist accounts at 31-38; the three-elite mechanism at 43-48; the Israeli case at 50-60; the 1992 enactments at 54-55; negative rights litigation as eighty to ninety percent of the caseload at 105-08; the impact on procedural justice and negative liberties at 117-19; the erosion of union protections in Canada and Israel at 139-46; the absence of distributive effect and the inequality data at 148 and 156-62; the judicialization of the “who is a Jew” and “Jewish and democratic” questions at 172-99; the wider transfer to central banks and trade bodies at 215-16. Publisher’s summary at Harvard University Press. If you can get one thing beyond the summary, get the consequences chapters; the origins argument is the famous half and the weaker one.

Reviews and the state of the argument. Daniel Moure’s book note, 50 McGill L.J. 459 (2005), is the most useful short summary in English and is free; it also draws the comparison to Charles Beard (1874-1948) on the American founding and to the radical institutionalism of Thorstein Veblen (1857-1929) and C. Wright Mills (1916-1962). Carlos Closa’s review, 4 Int’l J. Const. L. 581 (2006), is the source of the motive-evidence objection and is here. Leslie Friedman Goldstein, “From Democracy to Juristocracy,” 38 Law & Soc’y Rev. 611 (2004), reviews Hirschl alongside Stone Sweet and Guarneri and Pederzoli. For the objection that judicial review is too unpredictable to be a rational elite strategy, see this response on the CALJ blog, which is undergraduate-level.

The rival model.Tom Ginsburg’s insurance thesis, in Judicial Review in New Democracies (Cambridge, 2003), holds that parties facing electoral uncertainty create strong courts as insurance against losing. Hirschl’s hegemonic preservation is the limiting case where the elite already knows it is losing. I raised the contrast in one paragraph and did not develop it. If you want the comparison done properly, Ginsburg and Mila Versteeg have tested insurance theory empirically and that is where the argument now lives.

Hirschl applied to the current fight. Moshe Cohen-Eliya, “Israel’s Juristocracy,” Law & Liberty, March 2024, already applies hegemonic preservation to the January 2024 reasonableness ruling and is the source of the point about Hayut and Baron issuing opinions after retirement. Cohen-Eliya writes from a position sympathetic to reform. See also Udi Sommer, Rachel Colson and Nathaniel Schmidt, “Reconceptualizing the 1990s judicial revolution in Israel and its implications for 2023-25,” Frontiers in Political Science (2025), which argues that Barak reflected a global 1990s democratization wave rather than a local elite strategy. That is a direct challenge to hegemonic preservation and I could have engaged it at more length.

The strongest defense of the court. Barak Medina, “On constitutional identity, democratic legitimacy, and judicial review in times of democratic backsliding: The case of Israel,” Int’l J. Const. L., moag052, published May 21, 2026. Open access. This is the article to read before writing anything further on the subject. Footnote 4 is where he rejects the hegemonic-preservation reading, citing Hirschl and Gavison. His sliding-scale doctrine, his four indicia of abuse of constituent power, and the protest figures I quote (forty-plus weeks, up to 250,000 in Tel Aviv, three to seven and a half percent of adults weekly, twenty-two percent at least once) are all his. So is the thirty-two to twenty-one vote on Basic Law: Human Dignity and Liberty and the note that its constraining effect was doubted during the legislative process. His characterization of the 2023 program’s underlying aims is his argument, sourced to the coalition’s published policy guidelines, and I flagged it as such in the text.

Cases. CA 6821/93 United Mizrahi Bank v. Migdal (1995); HCJ 73/53 Kol Ha’am v. Minister of Interior (1953); HCJ 98/69 Bergman v. Minister of Finance (1969); HCJ 4676/94 Meatrael v. Knesset (1996); HCJ 5555/18 Hasson v. Knesset (2021) on the Nation-State Law; HCJ 8260/16 Academic Center of Law and Business v. Knesset (2017) on the two-year budget; HCJ 5969/20 Shafir v. Knesset (2021); HCJ 5658/23 Movement for Quality Government v. Knesset (January 1, 2024) on reasonableness; HCJ 2412/23 Movement for Quality Government v. Knesset (January 3, 2024) on incapacity. English translations of most of these are free at the Cardozo Israeli Supreme Court Project, which is the single best English resource on Israeli constitutional law and is underused by people writing about this fight. The reasonableness ruling: twelve of fifteen affirmed the power to review Basic Laws, eight of fifteen voided the amendment. Commentary from IDI and Verfassungsblog.

July 2026. Basic Law: Torah Study, 63-52 on July 13 (Jerusalem Post). Defense Service Bill (Amendment No. 28) suspending arrests through November 30, 2026, 58-54 on July 14. The attorney general law on July 15 (Ynet for the substance, the coalition package, the Gafni holdout, the January 1, 2027 commencement and the thirty-day appointment clause; Times of Israel for the filibuster). Ynet reported the tally as 61-51 and the Times of Israel as 65-51; I did not choose between them and neither should you without the Knesset record. Election set for October 27, 2026.

Koppel. “The Long Road to Democracy,” Kohelet, March 30, 2025, originally in Hebrew in Hashiloach, is where the Barbarians at the Gates axiom, the Blues and Oranges, and the account of the meetings appear. His June 2026 Tablet essay restates the argument for a general audience. His account of those meetings is single-sourced and he is a participant with a case to make; I used it as testimony and marked it as such.

Where this essay is weak. Three places. First, I am relying on reviews and a book note for page-level citations to Towards Juristocracy rather than on the book in front of me; verify the pin cites before publishing. Second, Hirschl has published continuously since 2004, including Constitutional Theocracy (2010), which is directly relevant to the yeshiva conscription fight as a study of how courts contain religion, and I did not use it, partly to stay inside the frame and partly because I have not read it recently enough to cite responsibly. If he has written on the 2023-2026 events himself, that would supersede a good deal of what I have argued here and I did not find it. Third, my claim that structural theories cannot accommodate a prime minister’s criminal jeopardy is a claim about a literature I have sampled rather than surveyed. Someone who knows the comparative judicial politics literature properly may be able to point to a model that handles it, and if so the argument in my last three sections needs revising rather than defending.

Trust in Numbers: Moshe Koppel and the Politics of Measurement

Theodore Porter (b. 1953) published a book in 1995 that inverted the usual story about why the modern world counts things. The received view held that quantification spread outward from physics and astronomy, where it worked, into economics and psychology and public administration, where people wanted it to work. Porter looked at the archives and found the arrow pointing the other way. Trust in Numbers: The Pursuit of Objectivity in Science and Public Life argues that quantitative rigor is a political product before it is an epistemic one, that it arrives under pressure from outside, and that its function is to make judgment invisible.

His demonstration rests on a comparison of two engineering elites in the same century doing similar work.

The French Corps des Ponts et Chaussées built roads, bridges, canals and railways, and its members were among the most prestigious officials in France. They calculated constantly. They did not believe the calculations decided anything. Colson taught a generation of them that traffic forecasting could never be made rigorous, because applying a formula required weighing the social and economic condition of the population served, which was the hard part. In 1967 one of them described the corps’s role as verifying the legitimacy of other men’s calculations, weighing the consequences if those calculations diverged from reality, and determining how much could be left to chance. These were men who trusted themselves to decide. Inside the corps, informal discussion among people who knew each other was enough to reach agreement, and Porter’s judgment is that they saw no reason to perform the ritual of formal quantitative procedure unless attacked from outside.

The United States Army Corps of Engineers was attacked from outside constantly. It fought Congress, rival agencies, the Bureau of Reclamation, private power companies, farm lobbies and railroads, and the historians describe the environment as disunity and savage infighting. Out of that came cost-benefit analysis, a system of rules elaborated across thousands of pages, which by the 1960s had become the American standard of public rationality. Porter’s finding is that the regime of calculation was imposed not by all-powerful experts but by weak and divided ones, in a setting of pervasive public distrust.

The British actuaries make the third case and the sharpest. Facing proposals that the state impose uniform premium rules, they answered that uniform rules would produce uniform errors and that a mathematically gifted graduate with no practical experience would be unqualified to price a life. Porter’s summary of their defense is six words: “In place of precision they offered a profession.” The regulators had enough power to prevail in some jurisdictions and not others, and the pattern is not about who was right.
Wikipedia

From this Porter draws his central distinction. Disciplinary objectivity means trusting a trained community to exercise discretion; its guarantee is membership and formation. Mechanical objectivity means following rules that constrain discretion; its guarantee is that the rules produce the same answer regardless of who applies them. The first is available only to groups secure enough that outsiders will accept their word. The second is what you build when they will not.

Hence the phrase that carries the book. Quantification is a technology of distance. Numbers minimize the need for intimate knowledge and personal trust, they travel past the boundary of a community, and what they exclude is judgment. On page eight Porter puts the political consequence in one sentence: “Quantification is a way of making decisions without seeming to decide.” Objectivity, he adds, supplies authority to officials who have little of their own.
OUP Academic

Now consider a career.

Moshe Koppel took a doctorate in mathematics under Martin Davis at NYU in 1979 and spent his working life at Bar-Ilan University building instruments that identify hidden authors. The design principle of his best-known work is Porter’s principle stated as an engineering specification.

What Koppel and his collaborators discovered is that the useful signals in prose are the ones the writer cannot govern. Vocabulary is a poor guide, because a historian writing about Napoleon and a physicist writing about superconductivity differ in their nouns for reasons that have nothing to do with either man. The reliable markers are function words: pronouns, articles, prepositions, conjunctions, prefixes and suffixes, punctuation, the frequency of grammatical forms. A writer can decide to drop a favorite adjective. Nobody can hold three hundred small grammatical habits under conscious supervision across ten thousand words.

The whole point is that the signature survives intention. His unmasking method takes this further: train a classifier to separate two bodies of text, strip out the features doing the most work, retrain, and repeat, on the theory that superficial differences vanish quickly while deep ones persist. The output is a number describing how far down a difference goes.

This is mechanical objectivity as a research program. The instrument is designed to produce the same answer whoever runs it, on material the subject could not have arranged, and its authority comes from the fact that no human discernment intervenes between the text and the verdict. In 2003, when the group reported that the same statistics could predict a writer’s sex, the Boston Globe and The New York Times both covered it within a month of each other, which is what happens when a number arrives in a place where people had been arguing from impression.

Porter’s question is not whether such instruments work. It is where they get built. And Koppel’s record answers with unusual clarity: he built them at the precise points where a community’s own procedures had failed to settle a question.

Take the first serious application. Yosef Hayim of Baghdad (1835-1909), the leading halakhic authority of Iraqi Jewry, published in 1878 a collection of 524 responsa called Torah Lishmah, presenting it as an old manuscript he had come across rather than as his own work. Historians had long since concluded that he wrote it. The internal evidence was suggestive, the motive was intelligible, and the consensus was stable without being demonstrable.

Koppel and Jonathan Schler compared the disputed collection against 509 responsa Yosef Hayim acknowledged and against collections by four contemporaries. The comparison authors dropped away. The distance between the disputed text and the acknowledged text behaved the way same-author distances behave. What arrived was a measurement where there had been a scholarly judgment, on a question the rabbinic world had transmitted for a century without resolving through its own authorities.

The Kherson Genizah is the harder case, because there the community had an answer.

A trove of letters surfaced in Ukraine around 1917, presented as copies of correspondence among the Baal Shem Tov, Shneur Zalman of Liadi and other early Hasidic masters. Scholars judged them fraudulent. Material analysis of the paper and ink pointed to a date too late for the attributions. But the material evidence had been absorbed: as Koppel described the counter-argument in a 2021 discussion, the response available inside the tradition was that the documents were later copies of authentic originals, which explains the paper and leaves the letters standing.

That is the exact condition Porter describes. Intimate knowledge and personal authority could not close the dispute, because the parties disagreed about what would count as closing it. Koppel’s stylometric analysis compared the letters attributed to different masters and found the supposed distinctions among the writers behaved like the work of one hand rather than several. He published it in Hebrew, in a rabbinic journal, in 2010, aimed at readers inside the argument.

A technology of distance, deployed at a distance of one community’s width.

Then the Bible, where Porter’s engine ought to have run away with the operator, and did not.

In 2011 Koppel published a paper with Navot Akiva, Idan Dershowitz and Nachum Dershowitz on splitting a document into stylistic components with no labeled examples. The method builds about two hundred synonym sets for biblical Hebrew, generated automatically by finding roots the King James renders with a common English root, and clusters passages by which synonym they choose.

Before running it on anything contested they built a test set, which is mechanical objectivity in its ideal state. Take Jeremiah and Ezekiel, two books by two authors, shuffle them, and see whether the method unshuffles them. Seventeen percent of verses were unclassifiable for want of any distinguishing word. Of roughly 2,200 that were classified, twenty-six landed wrong. An error rate of 1.2 percent, published, checkable, and independent of anyone’s discernment.

Then they ran it on the Torah, split into two families, and the output matched the standard critical division between Priestly and non-Priestly material for about ninety percent of the verses where the standard authorities agree with each other.

The wire services wrote that an algorithm had shed light on the authorship of the Bible. By the time the story reached Haaretz, readers learned that the project’s head had announced proof that the Torah had multiple human authors.

He then did something Porter’s book records almost nobody doing. He wrote three thousand words on the Seforim Blog, explaining that the number of families was an input he had chosen, that the method could not have discovered it, that asking it for two families and receiving two says nothing about how many authors a text had, that requesting three degraded the correspondence and four or more made the critics’ J and E vanish, and that none of it could speak to divine authorship because nobody possesses a theory of what a divinely written text should look like. He noted the objection that this makes the claim unfalsifiable and declined to resolve it there.

A man who had built the instrument told the public that the instrument could not carry the weight the public wanted to put on it. Porter’s chapters on cost-benefit analysis are largely a record of the opposite: economists who took a rough bureaucratic convention and turned it into a license to price lives, days of illness, and the net benefit of polio vaccination.

Koppel also holds the other half of Porter’s pair, and holds it with conviction.

Meta-Halakhah, published in 1996, argues that Jewish law works like a natural language. A grammar describes a language without generating every judgment a competent speaker makes. Native speakers know a sentence is wrong before they can name the rule it violates. Accomplished halakhists recognize that an argument does not fit the tradition before they can identify the doctrinal reason, and the intuition is not sentiment but competence built through immersion in cases. Codes are indispensable, and excessive codification flattens the intuitions that give the codes their sense.

Read that alongside the actuaries and the sentences are interchangeable. Uniform rules imposed from outside would produce uniform errors. A brilliant formalist with no practical formation is unqualified to price the risk. In place of precision, a profession. Koppel makes the actuaries’ argument about the rabbis, and makes it in a book, and means it.

So one man builds mechanical objectivity for the Ben Ish Chai and the Kherson letters, and defends disciplinary objectivity for the halakhic decisor, and the question Porter’s frame poses is what determines which he reaches for.

Attributing a nineteenth-century responsum collection and deciding a novel question of Sabbath law are both hard in the same general way, both underdetermined by the written record, both dependent on accumulated tacit competence.

It is the boundary of a community he belongs to. Inside it, he argues that the expert’s discernment exceeds any rule that could be written for it. Outside it, or at its contested edges, he builds the rule that dispenses with the expert.

Dicta, the center Koppel founded to bring computational tools to Hebrew, is the same instinct turned to infrastructure, and Porter’s account of standardization explains what the project is for.

The technical work is elegant in a way that shows the house style. Finding every parallel passage in the Babylonian Talmud by brute force would need something like 1.6 trillion edit-distance calculations, which the authors estimated at roughly twenty years on a single machine. Instead they represented each word by its two rarest letters, which works because the commonest Hebrew letters are yod, aleph and vav, the vowel markers and the common prefixes, so discarding them discards most of the spelling noise. Then they matched short strings differing by at most one word and clustered the results. The run found 4,602 parallel pairs across 1.8 million words in eleven seconds.

The language models followed: BEREL in 2022 for rabbinic Hebrew, on the premise that rabbinic Hebrew is a domain in its own right and not defective modern Hebrew; DictaBERT in 2023; DictaLM and its successors, up to an open-weight family released in December 2025 trained on roughly a hundred billion Hebrew tokens.

The phrase in the technical report is sovereign Hebrew, and Porter supplies the reading. Standardization is what allows knowledge to travel past a locality, and the question in any standardization project is whose instruments do the measuring. A society whose inheritance sits in a low-resource language and whose reading is increasingly mediated by models trained elsewhere is a society measured by other people’s rules. Koppel’s answer was to build the datasets, the benchmarks, the morphology and the models, and to release enough that others can check them. A technology of distance, constructed to keep the local from being flattened by somebody else’s.

Now Kohelet.

Consider what the Israeli legal establishment was defending. The reasonableness doctrine allows a court to void an administrative decision when the decision-maker failed to give all relevant considerations the right weight. Aharon Barak expanded it in 1981 from the narrow standard of a decision no rational person could have reached to a general power to substitute the court’s weighting for a minister’s, and the resulting test has a name that admits what it is: the zone of reasonableness. Its content is what a trained lawyer says it is. There is no formula, no threshold, no count.

The binding legal opinion of the attorney general is the same claim made about a person rather than a standard. The advice governs because of who gives it and how that person was formed, and the argument for the arrangement is that a professional legal service, insulated from ministers, will exercise better judgment than ministers would.

Both are disciplinary objectivity in doctrinal dress. Trust us; we are trained; the discretion cannot be reduced to a rule without producing uniform errors. And in Porter’s terms both are available only to a group secure enough that outsiders accept its word.

Koppel’s argument against them is a set of counts.

Look at the structure of his case in “The Long Road to Democracy.” In no western democracy is the attorney general’s advice binding. In no western democracy may the government’s lawyer argue against the client in court. No western democracy has a bar for unreasonableness as low as Israel’s; the comparison is to Wednesbury, and the point is a distance between two thresholds. In no western democracy are laws at the top of the legal hierarchy subject to substantive review absent an explicit constitutional provision. In no western democracy do sitting justices hold a veto over appointments to a court exercising constitutional review. Israel’s judicial selection committee has nine members, three of them sitting justices and two from the bar association.

Every item is a tally of jurisdictions or a count of seats. This is what Porter’s Army Corps produced, and for the reason he identifies. The Blues held the institutions and argued from formation and discernment. The Oranges held the electorate and had no standing inside the professions, so they argued from benchmarks, because a benchmark travels to a person who does not accept your credentials and a discernment does not.

Kohelet’s organizational form follows from the same condition. Israeli parties and members of Knesset have thin research staffs. What Koppel supplied was not essays but statutory text, comparative memoranda, international tables, expert witnesses for committee, and lawyers who stayed with a bill through amendment. Porter’s Corps of Engineers built cost-benefit analysis because it had no electoral mandate and needed its choices to look like arithmetic. Kohelet had the mandate and needed the arithmetic to acquire standing. Same instrument, opposite deficiency, and in both cases the instrument’s job is to make a preference look like a finding.

Koppel described the payoff himself in April 2024, in his house in Efrat, and his formulation is close to Porter’s. A think tank wins, he said, when it converts a nonissue into an issue. Judicial reform had been nothing at all until somebody produced the numbers that made it a subject.

Porter’s warning about what happens next is specific, and it has now been tested.

His observation on page eighty-five is that quantification imposes order by licensing you to ignore or reconfigure whatever is difficult or obscure, and that when it succeeds it almost always costs something in subtlety. His observation about the technocrat and the quantifier is that the quantifier accommodates himself to legislatures by concealing, and sometimes denying, his own authority as a man of discernment. And his conclusion on page two hundred and fifteen is that the pursuit of mechanical objectivity cannot settle public issues under conditions of pervasive distrust.

On July 15, 2026, after nearly seventy committee sessions run by Simcha Rothman (b. 1980) and an eleven-hour filibuster, the Knesset ended the binding force of the attorney general’s legal opinions. The government may now declare that a written opinion of Gali Baharav-Miara (b. 1959) does not state the law. It may take its own position in court. If she declines to argue that position, it may hire private counsel, and she may not appear without the government’s permission. The government must report each declaration to a Knesset committee. Yariv Levin (b. 1969) called it a restoration of governing ability. Benjamin Netanyahu (b. 1949) was absent from the vote.

Set that against what it replaced. A court exercising the reasonableness standard had to publish reasons, hear an adversary, and write in a form that would bind it in the next case. The professional whose judgment was at issue could be named and criticized. Under the new arrangement a cabinet resolves that an opinion does not state the law, and notifies a committee. No opinion, no adversary, no precedent, no requirement of consistency next year.

The rule displaced the professional, and the discretion moved to a room with no obligation to explain. Porter’s point about mechanical objectivity was never that it fails to remove judgment; it was that it relocates judgment and then conceals the relocation.

The same week supplies the other half of his prediction. On Monday, July 13 the Knesset passed Basic Law: Torah Study, sixty-three to fifty-two. On Tuesday it suspended arrest, investigation and prosecution of yeshiva draft evaders through November 30, 2026, fifty-eight to fifty-four. On Wednesday the attorney general law carried, and it carried because the Haredi parties supplied votes after those two measures, and it nearly failed anyway when Moshe Gafni (b. 1952) held out over funding for ultra-Orthodox kindergarten teachers, so Rothman stretched a speech at the podium while Netanyahu put the kindergarten item on the cabinet agenda.

Not one clause of that was reached by a benchmark. It was pure unquantified bargaining among people who trusted each other enough to trade, which is Porter’s description of the French corps at work, conducted by the coalition that had spent a decade insisting that judgment be replaced by rules. A court whose power had rested on trained discretion was cut back by a majority assembled through undocumented discretion, in a week.

One further application, and it is the ugliest in Porter’s book and the least deniable.

Quantification is asymmetric. It is demanded of others. Porter’s actuaries and accountants spent decades resisting rules imposed from outside while writing standards for people below them, and his general point is that the audit is something you legislate for someone else.

Koppel drafted the law, passed by the Knesset in February 2011, requiring Israeli nonprofits to disclose funding received from foreign governments. The argument was transparency about foreign state intervention in Israeli politics. Opponents said it targeted European-funded human rights organizations while leaving private foreign donations to the right in the dark. Kohelet, over the following decade, generally declined to name its private donors, and maintained that private philanthropy differs from money supplied by a foreign state. That distinction has legal content. It does not answer the point about who is required to be counted.

Then the counting arrived anyway, through the Registrar of Associations. Income of thirty million shekels in 2022, twenty-four million in 2023, 13.6 million in 2024, a drop of 43.2 percent in a single year after Arthur Dantchik stopped giving. Of roughly four million dollars in 2024, about three and a half million came through the Central Fund of Israel. In January 2025 the Knesset Finance Committee left the organization off a list of 138 nonprofits approved for donor tax credits while the Tax Authority examined claims of partisan political activity; the claims were found baseless and the status was restored in March, with ten coalition members turning up to make sure. Kohelet said more than ten thousand Israelis had recently given it money.

Every one of those figures is a technology of distance operating on the organization that built technologies of distance for a living, and none of them was volunteered. Porter would not call this hypocrisy. He would call it the ordinary politics of accounting, in which whether you appear as a judgment or as a number depends on whether you are strong enough to insist on the first.

So where does the man quantify, and where does he trust?

He defers to the halakhic decisor, whose competence he says exceeds any code that could be written for it. He deferred to the biblical scholars, and went out of his way to tell the press that his algorithm had not settled the question they wanted settled. He built Dicta so that the people who read rabbinic Hebrew would have instruments made by people who read rabbinic Hebrew.

He quantified against the Hasidic collectors who would not concede the Kherson letters, against a Baghdadi pseudonym the tradition had carried unresolved, and against an Israeli legal establishment whose central doctrine is that its own discernment cannot be reduced to a rule.

The line does not run between hard questions and easy ones, or between science and politics. It runs along the edge of the communities in which he holds membership. Inside, the profession; outside, the number. And this is not a discovery about one man’s character. It is Porter’s thesis with a biography attached: quantification is what arrives when trust runs out, so a career’s worth of instruments maps where their builder’s trust ended.

The Seforim Blog essay of July 2011 reads nothing like the constitutional writing, and the reason is available in the frame rather than in Koppel’s temperament. In the Bible case he stood among people whose judgment he wanted to leave intact, and he told them what his machine could not do. In the constitutional case he stood outside a profession that had never accepted his side’s standing, and there was nothing on his side of the boundary that his instrument might damage.

Porter’s last chapters argue that objectivity is thickest in communities with insecure borders and persistent boundary problems, and thinnest where a group is confident enough that its word is taken. By that measure the most rigorously objective institution in this story is the think tank Koppel built to argue with people who did not trust him, and the least objective is the Gerrer shtiebel on the Upper West Side where he learned what he was arguing for.

Notes

The frame. Theodore M. Porter, Trust in Numbers: The Pursuit of Objectivity in Science and Public Life (Princeton University Press, 1995; new edition with a fresh preface, 2020). Page references for what I used: the preface at ix for quantification as a technology of distance, the minimizing of intimate knowledge and personal trust, and objectivity as a set of strategies for dealing with distance and distrust; page 8 for the appeal of numbers to officials lacking an electoral mandate, decisions made by the numbers appearing fair and impersonal, and making decisions without seeming to decide; page 44 on measures being gamed once managers are judged by them, with the Forest Service growth-rate example; pages 85-86 on quantification imposing order by licensing you to ignore what is obscure, and on the cost in subtlety; page 91 on expertise as discernment against mechanical objectivity; pages 108-113 on the British actuaries, Charles Ansell’s testimony, and the line about offering a profession in place of precision; pages 137 and 141-142 on the Ponts et Chaussées, Colson, and J. Mante’s 1967 remark; page 146 on the difference between technocrats and quantifiers; page 149 on the Army Corps of Engineers and the regime of calculation imposed by the weak and divided; pages 187-189 on the capture of cost-benefit analysis by economists and its transformation into a universal standard under conditions of public distrust; page 200 on statistical inference taken up first by weaker disciplines; page 213 on quantifiers having made the world statistical; page 215 on the limits of mechanical objectivity under pervasive distrust; page 230 on the trappings of science appearing most in fields with insecure borders. A useful set of excerpts with these page numbers is here; publisher’s page at Princeton. The disciplinary/mechanical distinction is introduced in the opening chapter, “Cultures of Objectivity,” abstract here. If you want the argument in twenty pages rather than three hundred, Porter’s 1992 article “Quantification and the Accounting Ideal in Science,” Social Studies of Science 22, covers the same ground.

Koppel’s science. Koppel, Schler and Argamon, “Computational Methods in Authorship Attribution,” JASIST 60 (2009). Koppel, Schler and Bonchek-Dokow, “Measuring Differentiability: Unmasking Pseudonymous Authors,” JMLR 8 (2007), which is where the Torah Lishmah case appears with the 524 and 509 document counts. Argamon, Koppel, Pennebaker and Schler, “Automatically Profiling the Author of an Anonymous Text,” CACM 52 (2009). The 2003 press cycle: Charles McGrath, “Sexed Texts,” New York Times, August 10, 2003; Clive Thompson, “He and She: What’s the Real Difference?“, Boston Globe, July 6, 2003.

Kherson. Koppel’s Hebrew paper is in Yeshurun 23 (2010), 559-566. His summary of the counter-argument about the paper and ink and the letters being later copies comes from his own account in a 2021 public discussion, and I have characterized it as his paraphrase rather than as the position of any named authority, which is what it is. The forgery judgment long predates him; his contribution was quantitative evidence, not the conclusion.

The Bible episode. Koppel, “Attribution and Misattribution: On Computational Linguistics, Heresy and Journalism,” Seforim Blog, July 11, 2011. Every figure I cite is his: about two hundred synonym sets, seventeen percent unclassifiable, twenty-six of roughly 2,200 misassigned in the Jeremiah and Ezekiel test, about ninety percent agreement with the Driver and Friedman consensus, and the collapse of the correspondence at three or more families. The underlying paper is Koppel, Akiva, I. Dershowitz and N. Dershowitz, “Unsupervised Decomposition of a Document Into Authorial Components,” ACL 2011.

Halakhah. Meta-Halakhah: Logic, Intuition, and the Unfolding of Jewish Law (Jason Aronson, 1996). Also Seder Kinnim, and the journal Higayon, co-edited with Ely Merzbach.

Dicta. Shmidman, Koppel and Porat, “Identification of Parallel Passages Across a Large Hebrew/Aramaic Corpus,” J. Data Mining & Digital Humanities (2018): 4,602 pairs, 1.8 million words, eleven seconds in the revised text, against an estimated twenty years for exhaustive comparison. BEREL, DictaBERT, and the Dicta-LM series at dicta.org.il. The coincidence of surnames between Ely Porat and Theodore Porter is a coincidence.

Kohelet and the comparative argument. Koppel’s list of jurisdictional comparisons is in “The Long Road to Democracy,” Kohelet, March 30, 2025, originally in Hebrew in Hashiloach. His remark about a think tank winning by turning a nonissue into an issue is from Eliav Breuer’s interview, “Kohelet’s Moshe Koppel reflects on the judicial reforms a year later,” Jerusalem Post, April 13, 2024. On the reasonableness doctrine and the zone of reasonableness, the case law is at the Cardozo Israeli Supreme Court Project, which is free and better than any secondary summary.

July 2026. Basic Law: Torah Study, 63-52 on July 13 (Jerusalem Post); the Defense Service amendment suspending arrests through November 30, 2026, 58-54 on July 14; the attorney general law on July 15 (Ynet for the substance, the notification-to-committee requirement, the coalition package and the Gafni holdout; Times of Israel for the filibuster). Ynet reported the tally as 61-51, the Times of Israel as 65-51; I avoided the number and you should check the Knesset record.

The money. Times of Israel on Dantchik, August 4, 2023; Shomrim, July 21, 2025 for the shekel figures, the 43.2 percent fall, and the Central Fund of Israel share; Times of Israel, January 20, 2025 and JNS, March 3, 2025 on the Section 46 fight and the ten thousand donors. The 2011 foreign-government funding disclosure law is documented in Koppel’s Wikipedia biography with a link to the sponsor’s archive.

Where this is weak. Three places, and the first is the largest. Porter’s thesis is about institutions under external pressure, and I have applied it to one man’s intellectual biography, which requires assuming that a scholar’s choice of method responds to the same pressures a bureau’s does. That is plausible, and a hostile reader should press there first. Second, my central claim is about where Koppel’s trust ends, and I inferred it from what he built rather than from anything he has said about it; he has never described his own work in these terms and might reject the description. Third, nobody appears to have applied Porter to computational philology or to Israeli policy institutes, which is why the argument has room, and also means there is no literature to correct me. If you want the frame stress-tested before publishing, the obvious counter-case is Dicta: a project that quantifies inside a community Koppel belongs to, which either complicates the membership rule or shows it applies to disputes rather than to tools. I lean toward the second and did not argue it.

A Little Man Looking at a Good Map

In July 1787 a French naval expedition under the comte de La Pérouse (1741-1788) anchored off a coast in the North Pacific. Louis XVI (1754-1793) had sent him to bring back a better chart, and one live question was whether the place he was standing on, Sakhalin, was an island or a peninsula. European geographers had produced contradictory maps and were arguing.

La Pérouse asked the local fishermen. An older man stood up and drew the island in the sand, at scale, with the detail required. A younger man saw that the rising tide was about to erase it, took one of La Pérouse’s notebooks, and drew it again in pencil.

Bruno Latour (1947-2022) built an entire theory of knowledge out of that afternoon. His point is that nothing in the encounter shows a European mind working differently from a Chinese one. The fishermen understood projection, scale and navigation, and they knew the coast better than any Frenchman ever would. What separated the two parties was a difference in trajectory. The islanders were born there and would die there, so they could generate a map whenever they needed one and had no reason to keep it. La Pérouse was leaving in the morning and would die far away, and the whole purpose of his voyage was to arrive somewhere else carrying something that would settle an argument among people who had never been to Sakhalin, decide who owned which part of the ocean, and route the next fleet.

So the sand drawing was worthless to him and the pencil drawing was the object of his mission. What made the difference was not the quality of the knowledge. It was the invention of a form of the knowledge that could survive the journey.

Latour called such things immutable mobiles, which is an ugly term for a simple idea: a thing that travels without changing. A chart, a printed table, a photographic plate, a specimen in gelatin, a drawing to scale. And he called the places where these things pile up centres of calculation, which is another ugly term for another simple idea: a room where the paper accumulates. Versailles. A laboratory. An office.

The operation, once you see it, is always the same. Go out. Convert whatever is there into marks that will not degrade in transit. Come back. Spread the marks on a table where they can be compared with marks from a hundred other places. Combine them, extract a smaller set of marks from the pile, and send instructions back out along the same route. Do this for two centuries and the room can act on places nobody in it has visited.

Latour’s list of what marks on paper let you do, from a 1986 essay called “Visualisation and Cognition,” is worth having in front of you, because it reads like a specification document. They move. They arrive unaltered. They are flat, and a flat surface a few feet across can be dominated by one pair of eyes, with nothing folded or hidden. Their scale can be changed at will without changing their internal proportions, so a galaxy and a chromosome end up the same size on the desk. They can be copied cheaply, so every place and time can be gathered in one place and time. They can be shuffled and recombined. Images from unrelated sources can be laid over one another. They can be dropped into a written text with a little cleaning up, so that an article carries its evidence inside itself. And they merge with arithmetic and geometry, which is the largest advantage of all. You cannot put a ruler on the sun. You can put one on a photograph of the sun, and the centimeters you read there will travel to any other object you like.

Latour’s summary of what all this buys is one sentence, and it is the least mystical account of power anybody has written. “The ‘great man’ is a little man looking at a good map.”

Now consider Moshe Koppel, who has spent forty years in two rooms where paper accumulates.

Start with the authorship work, because it is La Pérouse’s problem restated.

Koppel took his doctorate in mathematics under Martin Davis in 1979 and became known for identifying hidden authors from statistical traces. The method rests on the finding that the useful signals are the ones a writer does not govern: pronouns, articles, prepositions, conjunctions, prefixes, suffixes, punctuation, the frequency of grammatical forms. Vocabulary betrays subject matter. Function words betray the hand.

His first serious case was a rabbinic dispute. Yosef Hayim of Baghdad, the leading halakhic authority of Iraqi Jewry, published in 1878 a collection of 524 responsa presented as an old manuscript he had come across. Historians had long concluded he wrote it himself. The conclusion was stable and it was not demonstrable, and it had been transmitted for a century as an educated opinion.

Notice what the Baghdadi world had and lacked, in La Pérouse’s terms. Its scholars knew Yosef Hayim’s prose the way the fishermen knew the coastline, from immersion, from a lifetime of reading, in a form no classifier will ever match. What they did not have was a version of that knowledge that could leave Baghdad, arrive at a machine learning conference, and come back with a citation attached.

Koppel and Jonathan Schler built one. Compare the disputed collection against 509 responsa Yosef Hayim acknowledged, and against collections by four contemporaries. Convert every text into frequencies. Frequencies are flat, they arrive unaltered, they rescale, they superimpose. Then run the unmasking procedure, which trains a classifier to separate two bodies of writing, deletes whatever features did most of the separating, retrains, and repeats, on the theory that surface differences collapse fast between texts by one hand and deep differences persist between texts by two. The output is a single number describing the depth of a difference.

Six collections, all of them mutually incomparable as reading experiences, laid out on one surface and reduced to a distance. That is the pencil drawing in the notebook.

The Kherson Genizah is the harder case.

A trove of letters surfaced in Ukraine around 1917, presented as correspondence among the Baal Shem Tov, Shneur Zalman of Liadi and other early Hasidic masters. Scholars judged them forged, and material analysis of the paper and ink pointed to a date too late for the attributions.

The material evidence had already been absorbed. As Koppel described the counter-argument in a 2021 discussion, the answer available inside the tradition was that these were later copies of authentic originals, which accounts for the paper and leaves the letters standing.

Latour has a proposition for this situation. Any single inscription can be reinterpreted, and a dissenter can always decline to look. Which is why, in his account, so much effort goes into producing one more trace, one more contrast, one more redrawing, and the reason it works is not that the trace compels belief. The reason is that each additional inscription raises the price of continuing to disagree.

Koppel’s analysis compared the letters attributed to the various masters against one another and found the supposed differences behaving like the work of a single hand. He published it in Hebrew, in a rabbinic journal, in 2010, aimed at readers whose dissent it was designed to make more expensive.

The parallel-passages paper is the purest instance of the specification list in the whole career, and it is worth walking through slowly because the engineering is the argument.

The Babylonian Talmud repeats itself constantly, with variation: a passage in one tractate reappears three hundred pages later, rephrased, respelled, with a word dropped. Finding every such pair by direct comparison would require something like 1.6 trillion edit-distance calculations, which Koppel, Avi Shmidman and Ely Porat estimated at roughly twenty years on one machine.

Their solution was to represent every word by its two rarest letters. This works because the commonest letters in Hebrew are yod, aleph and vav, which happen to be the vowel markers and the common prefixes, so throwing them away throws away most of the spelling instability that made the comparison hard. Then match short strings differing by at most one word, and cluster the matches. The run produced 4,602 pairs across 1.8 million words in eleven seconds.

Every property on Latour’s list is doing work there. The transformation makes unstable Hebrew orthography into comparable strings, which is arrival without alteration. The output is flat. It rescales, so 1.8 million words fit on a screen. And two passages three hundred pages apart, which no reader could hold in one glance, are now inches apart, which is Latour’s description of what a well-kept file does: distant things become adjacent, folded things become flat, and thousands of instances become visible at once.

Which brings the frame to its most interesting encounter with its subject, and to the one place where it fails.

Latour’s essay ends on what he calls a strange anthropological puzzle. Our culture trains people to array marks in cascades, each summary summarized again, and then to believe the last item in the series more than any evidence to the contrary: more than common sense, more than the other senses, more than local authority, more than tradition, more than scripture. Galileo believed his triangular diagram against the evidence of his eyes. When in doubt, believe the inscription. Latour thinks that habit, and not any change in the human mind, is what our societies are made of.

In 2011 Koppel published a paper with Navot Akiva, Idan Dershowitz and Nachum Dershowitz that built such a cascade. Roughly two hundred synonym sets for biblical Hebrew, generated automatically from the King James renderings. Passages scored by which synonym they choose. A clustering step. A set of core passages identified as reliably assigned. A model trained on those. Per-verse classification from the model. Six or seven layers, each one a simplification of the last.

They tested it the way you test a ruler. Take Jeremiah and Ezekiel, shuffle them, and see whether the machine unshuffles them. Seventeen percent of verses could not be classified for want of any distinguishing word. Of roughly 2,200 that were classified, twenty-six landed wrong. An error rate of 1.2 percent.

Then they ran it on the Torah, split into two families, and the output matched the standard critical division between Priestly and non-Priestly material for about ninety percent of the verses where the leading authorities agree with each other.

The wire services wrote that an algorithm had shed light on the authorship of the Bible. By the time the story reached Haaretz, readers learned that the project’s head had announced proof of multiple human authorship.

He then declined to believe his own last inscription, in public, at length. The Seforim Blog essay of July 11, 2011 explains that the number of families was an input he had chosen, that the method has no capacity to discover it, that asking for two and receiving two says nothing about how many authors a text had, that requesting three degraded the correspondence and four or more made the critics’ J and E disappear entirely, and that none of it could touch divine authorship because nobody possesses a theory of what a divinely written text should look like. He raised the objection that this renders the claim unfalsifiable and left it standing.

Latour’s frame accounts for why that essay is unusual. It cannot account for why it exists. In his vocabulary a refusal to press an advantage is a strategic choice inside an argument, and there is no place in the apparatus for a man giving up an inscription because it does not support the claim. Hold that thought too.

Dicta, the center Koppel founded to bring computational tools to Hebrew, is the part of the career that Latour explains best and that nobody else has explained at all.

His term for the relevant work is metrology, by which he means the whole invisible apparatus of standards without which no measurement travels: the calibration, the reference constants, the agreed units, the bureaus that maintain them. His figure is that metrology costs up to three times the budget of all research and development, and that this is only the first link in the chain. Universality, in his account, is not a property of science. It is a practical achievement, paid for continuously, and full of holes.

Hebrew’s texts resist the operation at every level. Ordinary writing drops the vowels, so one written string supports several pronunciations and several meanings. Prefixes carrying conjunctions, articles and prepositions attach to the front of words, and suffixes carry possession and objects, so a string that looks like one word contains four grammatical units. Rabbinic material adds Aramaic, dense abbreviation, and spelling that never stabilized. Biblical, rabbinic, medieval and modern Hebrew differ in vocabulary and syntax. None of it will hold still long enough to be counted.

Dicta’s tools are the plumbing that makes counting possible: automatic vocalization, abbreviation expansion, morphological analysis and prefix segmentation, source identification, parallel-text discovery. Then the models, BEREL in 2022 trained on rabbinic Hebrew as a domain in its own right, DictaBERT in 2023, and the DictaLM series through an open-weight family released in December 2025.

The technical report calls the goal sovereign Hebrew, and read through Latour the phrase is a claim about standards rather than about pride. A society whose inheritance sits in a low-resource language, and whose reading is increasingly mediated by models trained elsewhere, has had another centre’s measuring apparatus extended across its texts. Building your own is the only alternative to being an entry in somebody else’s table.

Latour has a passage nearby on scripts that resist being arrayed in cascades, and he applies it to Chinese ideograms in a way that has aged badly and was contested when he wrote it. I would not lean on that passage. The underlying observation, that some writing systems require more preparatory work than others before they can be counted, is Dicta’s engineering problem and what the vocalizer is for.

Now the second room, and the reason this frame earns its place.

Latour has a sentence about bureaucracy that reads, three decades later, like a description of a specific building in Givat Shaul. A bureau, he writes, is a small laboratory in which many elements can be connected because their scale and nature have been averaged out: legal texts, specifications, standards, payrolls, maps, surveys. The rationality attributed since Hegel and Weber to the minds of Prussian officials is in the files. Distant domains become inches apart, hidden ones become flat, thousands of cases can be seen at once, and then the files can be summarized into files of files until a few men hold millions in their hands.

The Kohelet Policy Forum, founded in February 2012 at Am Ve’olamo 8, is that operation performed on Israeli policy, and the reason it worked is a conversion problem identical to La Pérouse’s.

The Israeli right’s convictions existed in abundance and did not travel. A resentment does not survive a trip into a Knesset committee. Neither does a sermon, a column, an election result, or an argument made at a kitchen table. What survives that trip is a document. A draft bill can be photocopied, tabled, amended, laid alongside a rival draft, quoted in a protocol, cited in a judgment, and carried into the next Knesset by someone who was not in the room the first time. Israeli parties have thin research staffs, so a body that arrives with finished statutory language, comparative memoranda, prepared responses to objections and a lawyer who stays through the amendments has converted belief into a form that does not degrade.

The comparative tables are the clearest case, and they are the part of Koppel’s writing most often read as rhetoric when it is closer to cartography. Look at the structure of the argument in “The Long Road to Democracy.” In no western democracy is the attorney general’s advice binding on the government. In no western democracy may the government’s own lawyer argue against the client in court. No western democracy sets the bar for unreasonableness as low as Israel does, and the comparison is to the Wednesbury standard, which is a distance between two thresholds. In no western democracy are the laws at the top of the legal hierarchy subject to substantive review absent an explicit constitutional provision. In no western democracy do sitting justices hold a veto over appointments to a court with constitutional review, and Israel’s selection committee has nine members, three of them sitting justices and two from the bar association.

Somebody had to go out to twenty jurisdictions, flatten their constitutional arrangements into commensurable form, and bring the result back so that it could sit on one page in front of a member of Knesset who has never read a foreign constitution. That is the cycle Latour describes, applied to constitutions instead of coastlines, and its product is the ordinary source of a centre’s advantage: the room can see many places at once, and each of those places can see only itself.

Latour’s account of who wins an argument is that it goes to whoever can assemble the largest number of well-aligned allies on the spot. A table of twenty countries is twenty allies present in a room that none of them has entered.

Koppel described the payoff himself in April 2024. A think tank, he said, does not measure victory only in laws passed. You write papers, you place them in newspapers, you talk to politicians, and the achievement is that an issue now exists which did not exist before. Judicial reform had been nothing at all, and Kohelet made it a subject. That is the operation stated by the operator.

Latour’s other proposition is that a claim’s strength is the network holding it up, which means the network can thin, and the Israeli record supplies both directions inside three years.

Thinning first. Arthur Dantchik stopped funding the organization on August 4, 2023 after protests outside his house in Philadelphia. Income fell from thirty million shekels in 2022 to twenty-four million in 2023 to 13.6 million in 2024. Dozens of employees went, research was cut, support for allied groups ended. Of roughly four million dollars in 2024, about three and a half million came through a single channel, the Central Fund of Israel, and the organization began advertising a tax refund to ten thousand small domestic donors after a fight over its Section 46 status that it lost in January 2025 and won in March.

Then enrolment. On Monday, July 13, 2026 the Knesset passed Basic Law: Torah Study, sixty-three to fifty-two. On Tuesday it suspended arrest, investigation and prosecution of yeshiva draft evaders through November 30, 2026, fifty-eight to fifty-four. On Wednesday, after nearly seventy committee sessions run by Simcha Rothman (b. 1980) and an eleven-hour filibuster, it ended the binding force of the attorney general’s legal opinions, allowing the government to declare that a written opinion of Gali Baharav-Miara (b. 1959) does not state the law, to take its own position in court, and to hire private counsel if she declines to argue it. The vote nearly failed when Moshe Gafni (b. 1952) held out over funding for ultra-Orthodox kindergarten teachers, so Rothman filled time at the podium while Benjamin Netanyahu (b. 1949), absent from the chamber, put the kindergarten item on the cabinet agenda.

Latour’s name for that trade is translation. It is the process. Item one on Yariv Levin’s (b. 1969) list of January 4, 2023 became law because a draft existed, and because two other communities’ interests were made to run through it. Nothing in the comparative tables produced that outcome. The tables produced a document, and the document could be traded.

His first methodological rule is to study these things before the lid closes. Once the law takes effect on January 1, 2027, it becomes a settled object that later actors will use without reopening the arguments that made it, the way the 2018 Nation-State Law is already used. The 2011 Bible essay looks, from here, like the same rule applied against the author’s own interest: he pried open a box the newspapers had already shut.

So the frame does what nobody else’s does. It describes Dicta and Kohelet in one vocabulary, and it turns out that the man who built machines for converting a dispersed textual inheritance into portable marks also built an organization for converting dispersed political conviction into portable documents, and that the two operations have the same structure, the same advantages, and the same relationship to distance. That is a real unification and it is not available from any other angle I know.

It is also where the objection sits, and the objection is fatal to using the frame for anything more than description.

Latour’s vocabulary cannot tell the two rooms apart. A stylometric classifier that misassigns twenty-six of 2,200 verses and a table of twenty constitutions assembled by people who want a particular result are both marks on paper, both flat, both rescalable, both superimposable, both traveling to a room where they combine. Nothing in the nine advantages distinguishes them. Nothing in the account of network strength distinguishes them either, since both derive their force from the allies they bring.

The difference is that one of them has data held back on purpose. The Jeremiah and Ezekiel experiment can go against you, and a 1.2 percent error rate is a number the author did not get to choose. A comparative table cannot go against anybody. There is no held-out set of constitutions.

Latour spent his career declining to grade inscriptions by their truth, on the ground that truth is what gets settled at the end rather than what explains the settling, and by 2004 he was publicly uneasy about it, asking in an essay whether the critical apparatus he had built had armed people he never intended to arm. That is the right note to end on, because Koppel is the same problem seen from the other side. He built instruments for making hidden structure visible, and twice he faced the question of whether to believe his own final number. In the philology he said no, in writing, against the newspapers, at cost to himself. In the constitutional work the question has not arisen, and the reason it has not is that a table of jurisdictions has no error rate with which to say no.

Both rooms are in Israel and both are full of paper. Latour’s line holds for each of them, which is his point and his limitation. A little man looking at a good map.

Notes

The frame. Bruno Latour, Science in Action: How to Follow Scientists and Engineers Through Society (Harvard University Press, 1987). Chapter 6, “Centres of Calculation,” opens with the La Pérouse episode under the heading “The Domestication of the Savage Mind” and proceeds through “Action at a Distance,” cycles of accumulation, “The Mobilization of Worlds,” and metrology; the relevant range is roughly pages 179-257. The fuller and more useful statement of the same argument, including the numbered list of nine advantages of inscriptions and the passages on bureaus, metrology and money, is Latour, “Visualisation and Cognition: Drawing Things Together,” in H. Kuklick (ed.), Knowledge and Society vol. 6 (Jai Press, 1986), 1-40, which Latour posted himself and which is free here. Read that essay rather than the book chapter if you only read one thing: the La Pérouse telling is better, the nine advantages are laid out in order, and the sentence about the great man and the good map is at the end of section V. The earlier source for inscription devices is Latour and Steve Woolgar (b. 1950), Laboratory Life (1979).

Latour’s own examples I drew on. Ivins on perspective and optical consistency. Svetlana Alpers on Dutch visual culture. Elizabeth Eisenstein (1923-2016) on the printing press, and the reading of Tycho Brahe (1546-1601) as the first astronomer to look at all previous predictions written in one form rather than at the sky. Foucault on clinical records and on files. Booker on engineering drawing. The Pasteur theater of proofs. The estimate that metrology costs up to three times all research and development spending is Latour’s, citing Hunter (1980), and I have not independently checked it; treat it as his figure. His paragraph on Chinese ideograms resisting cascades, drawn from McNeill’s The Pursuit of Power, is the weakest thing in the essay and I flagged it rather than using it.

Koppel’s science. Koppel, Schler and Argamon, “Computational Methods in Authorship Attribution,” JASIST 60 (2009). Koppel, Schler and Bonchek-Dokow, “Measuring Differentiability: Unmasking Pseudonymous Authors,” JMLR 8 (2007), where the Torah Lishmah case appears with the 524 and 509 counts and the four comparison authors. Shmidman, Koppel and Porat, “Identification of Parallel Passages Across a Large Hebrew/Aramaic Corpus,” Journal of Data Mining and Digital Humanities (2018): 4,602 pairs, 1.8 million words, eleven seconds in the revised text, against the twenty-year brute-force estimate, with the observation that the commonest Hebrew letters are also the vowel markers and prefixes. The Kherson analysis is in Hebrew, Yeshurun 23 (2010), 559-566; the forgery judgment long predates Koppel and his contribution was quantitative. BEREL, DictaBERT, and the Dicta-LM series at dicta.org.il.

The Bible episode. Koppel, “Attribution and Misattribution: On Computational Linguistics, Heresy and Journalism,” Seforim Blog, July 11, 2011. Every figure is his: about two hundred synonym sets, seventeen percent unclassifiable, twenty-six of roughly 2,200 misassigned, ninety percent agreement with the Driver and Friedman consensus, and the collapse at three or more families. Underlying paper: Koppel, Akiva, I. Dershowitz and N. Dershowitz, “Unsupervised Decomposition of a Document Into Authorial Components,” ACL 2011.

Kohelet. Founding in February 2012, offices at Am Ve’olamo 8, Givat Shaul; principals include Avraham Diskin, Avi Bell and Eugene Kontorovich. The comparative claims and Koppel’s account of the compromise talks are in “The Long Road to Democracy,” Kohelet, March 30, 2025, originally in Hebrew in Hashiloach. The remark about a think tank winning by creating an issue that was not an issue is from Eliav Breuer, “Kohelet’s Moshe Koppel reflects on the judicial reforms a year later,” Jerusalem Post, April 13, 2024. Levin’s five-part program of January 4, 2023 and the citation of Aviad Bakshi are documented across the Times of Israel coverage of that spring.

The money and the votes. Times of Israel on Dantchik, August 4, 2023; Shomrim, July 21, 2025 for the shekel figures and the Central Fund of Israel share; Times of Israel, January 20, 2025 and JNS, March 3, 2025 on Section 46. July 2026: Basic Law: Torah Study 63-52 (Jerusalem Post); the arrest suspension 58-54; the attorney general law on July 15 (Ynet for the substance, the Gafni holdout and the January 1, 2027 commencement; Times of Israel for the filibuster). Ynet gives the tally as 61-51 and the Times of Israel as 65-51; I avoided the number.

Latour’s later doubts. “Why Has Critique Run Out of Steam? From Matters of Fact to Matters of Concern,” Critical Inquiry 30 (2004), is where he worries in print that the tools he built for showing how facts are made were being used to unmake facts he wanted kept. I used it in one paragraph and did not develop it. If you want to build the argument out, that essay plus his later climate work is the material, and it is the same author on the same problem, so it stays inside the frame.

Where this is weak. Four places. First and largest: the frame’s inability to distinguish a tested classifier from an advocate’s table is not incidental to Latour, it is his considered position, so the objection I close on is an objection to the whole enterprise and not a repair I can offer. Anyone who takes Latour seriously will say I have smuggled in a standard he spent his life refusing. They are right, and I think the standard is worth the smuggling. Second, I have run a theory of laboratory practice over a political organization, and the analogy between a Knesset committee and a scientific controversy needs more defense than I gave it; Latour himself did extend the apparatus to bureaus and to money, which licenses the move without securing it. Third, my page ranges for Science in Action come from secondary summaries rather than from the book in hand, and the chapter structure should be verified. Fourth, nobody has applied this frame to computational philology or to policy institutes, which is the opportunity and also means there is no literature to catch my errors. The obvious stress test is Dicta’s release policy: Latour’s account predicts that a centre hoards its inscriptions to keep the advantage, and Dicta gives its tools away. Either that complicates the frame or it shows Koppel building a centre for somebody other than himself, and I did not resolve it.

Selling the Policy: Moshe Koppel and the Insurance Theory of Courts

Tom Ginsburg published a book in 2003 that asked a question nobody had answered well. Why would a government that has just won power voluntarily create an institution capable of overruling it?

Judicial Review in New Democracies studies three constitutional courts built in the wake of authoritarian rule, in Taiwan, Korea and Mongolia, and rejects the two standard answers. It is not culture, because the same East Asian societies said to be constitutionally hostile to judicial power built functioning courts within a decade. It is not a sudden collective commitment to rights, because the strength of the review power varies with the political arithmetic at the moment of drafting rather than with anyone’s professed convictions.

His answer is that judicial review is insurance, and that the people who buy it are the ones who expect to lose.

Work through the logic from the drafter’s chair. You are negotiating a constitution and you do not know who will win the elections it establishes. If you are confident you will dominate the legislature, a strong court is a pure cost: it can only obstruct what you intend to do. If you are confident you will be shut out permanently, a court is worth something but the people writing the rules will not give you one. The interesting case is the middle. When power at the drafting table is dispersed, and no party can be sure of its position after the first election, every party has a reason to want a body that will enforce the bargain against whoever happens to win. Each of them is buying protection against a future in which it is the loser.

So the prediction is specific and it runs in both directions. Fragmentation and uncertainty at the design moment produce strong courts. Concentration and confidence produce weak ones. Ginsburg found the pattern across his three cases and across the design of the review powers within them: the more dispersed the bargaining, the broader the standing rules, the easier the access, the wider the jurisdiction.

Ran Hirschl’s better-known argument, published a year later, is the limiting case of this one. Hirschl says a threatened elite entrenches its preferences in a court because it foresees defeat. Ginsburg’s model contains that as a special case at the far end of the uncertainty distribution, and covers a wider range besides, including the ordinary situation where nobody is sure of anything.

In 2014 Ginsburg and Mila Versteeg tested the thing against its rivals with a dataset covering 204 countries from 1781 to 2011. The measure of political competition standard in this literature, and the one they used, is the gap between the seat shares of the largest and second-largest parties in the legislature: a small gap means high uncertainty. Their finding was that adoption of constitutional review tracks domestic electoral politics, and that they could find no general support for ideological explanations, for federalism, or for the idea that countries adopt review because their neighbors did. Rights provisions diffuse across borders. The structural machinery does not. It responds to who is afraid of losing at home.

Israel is the best available test of this, for a reason nobody uses.

Most countries have a founding moment. A constitution is written once, by people in a room, under an identifiable balance of forces, and the theory gets one observation. Israel has never had that moment, and the absence is deliberate. On June 13, 1950 the Knesset adopted the Harari Resolution, which deferred the writing of a constitution and instructed the Constitution, Law and Justice Committee to prepare it chapter by chapter, each chapter to be a Basic Law, the chapters to be combined into a constitution at some unspecified future date. Seventy-six years later the combination has not happened.

The result is that Israel has constitutionalized in installments, over seven decades, under wildly different political configurations. Which means insurance theory can be run on the same polity a dozen times, with the political arithmetic varying and the country held constant. That is a natural experiment and comparative constitutional lawyers have not exploited it.

Run it on the central case and the fit is uncomfortably good.

The 12th Knesset was elected in November 1988. Likud took 40 seats. The Alignment took 39. On the standard measure of political competition used in this literature, a one-seat gap in a chamber of 120 is close to the theoretical maximum of uncertainty.

Then the arithmetic turned into an experience. In March 1990 the national unity government collapsed in the first successful no-confidence vote in Israeli history, following an attempt by Shimon Peres (1923-2016) to assemble an alternative coalition, an episode Israelis still call the dirty trick. Peres failed to form a government. Yitzhak Shamir (1915-2012) put together a narrow one that governed for two years without ever looking secure. Nobody in that chamber had any basis for confidence about anything.

Now look at what the chamber did in the last month of its life.

On March 3, 1992 it passed Basic Law: Freedom of Occupation. On March 17 it passed Basic Law: Human Dignity and Liberty. On March 18, the following day, it passed a new Basic Law: The Government establishing the direct election of the prime minister.

Three constitutional enactments in fifteen days, by a legislature at the end of its term, under maximum uncertainty about the succession.

Insurance theory does not merely accommodate this. It predicts both halves. The direct-election law was an explicit response to instability, an attempt to insure the office of prime minister against the coalition arithmetic that had just humiliated everyone. The rights laws are the same purchase made against a different risk. The government bill, drafted by a team under Justice Minister Dan Meridor (b. 1947), and the competing private bill from Amnon Rubinstein (1931-2024), both came out of the discussion opened by the unity government, and both would have handed an enforcement power to an institution that neither Likud nor Labor controlled.

Three years later, in United Mizrahi Bank, the Supreme Court held that these Basic Laws bind the Knesset and that it could void statutes inconsistent with them. Aharon Barak (b. 1936) called it the constitutional revolution. Read through Ginsburg, the revolution was a claim on a policy that a frightened Knesset had bought without reading the terms.

Two problems with that reading, both serious, and the second is worse than the first.

Basic Law: Human Dignity and Liberty passed with thirty-two votes in favor and twenty-one against, in a chamber of one hundred and twenty. Roughly a quarter of the members were present. Insurance theory models actors calculating their future electoral prospects and pricing an institution accordingly. A near-empty chamber is not what deliberate risk management looks like, though it is compatible with a cross-party understanding among the people who cared, and the bipartisan character of the support does cut in the theory’s favor.

The larger problem is who wrote the policy. The Knesset enacted two Basic Laws with no express provision for judicial review of legislation, and doubts were voiced during the process about whether they would constrain the legislature at all. The court supplied the enforcement power itself, three years later, by construction. Insurance theory is a theory about what designers do. It has no account of an insurer that writes its own coverage after the premium has been paid, and if that is what happened in 1995, the causal story runs from judicial ambition rather than from legislative fear, and the 1992 timing is a coincidence rather than an explanation.

That is not a small caveat. It is the same evidentiary gap that has dogged this literature since it began: the theory infers intention from configuration and outcome, and never from anything anyone said at the time.

Moshe Koppel (b. 1956), who founded the Kohelet Policy Forum in 2012 and supplied much of the intellectual apparatus for the judicial overhaul announced in January 2023, has made the insurance argument twice without naming it.

The first time is diagnostic. His essay “The Long Road to Democracy,” finished on October 6, 2023, describes a year of meetings with opponents of the reform whose arguments he could not follow until he located the premise underneath them. He calls it the Barbarians at the Gates axiom: the wrong people won the election, they will keep winning, and unless the court and the legal bureaucracy retain broad power, the state’s institutions will pass into unfit hands. He divides the country into Blues, who hold the professions and the institutions, and Oranges, who win the votes.

Stated in Ginsburg’s vocabulary, that is a claim that one camp holds an insurance policy it purchased when it expected to lose, and is now defending the policy rather than the principle.

The second time is prescriptive, and it is more interesting. In a Kohelet essay published in November 2025, and in the closing pages of the longer piece, Koppel argues that constitutional rules should be designed behind a veil of ignorance: without knowing which camp will hold power when the rules are applied, because nobody knows what the coalition and the court will look like in thirty years. He applies the test to his opponents, asking whether they would grant the same powers to unelected institutions if the Oranges staffed them and the Blues formed the government, and says everyone knows the answer.

That is an argument that his own side should behave as though it faced electoral uncertainty. Which is to say: an argument that his side should buy insurance. Ginsburg’s model, recommended as a normative principle, by a participant, in the middle of the fight, without a citation.

Nine months later his side declined to buy it.

On Wednesday, July 15, 2026, the Knesset passed the law ending the binding force of the attorney general’s legal opinions. The government may now declare that a written opinion of Gali Baharav-Miara (b. 1959) does not state the law, may take its own position in court, and may retain private counsel if she will not argue it. Yariv Levin (b. 1969) and Simcha Rothman (b. 1980) carried it through nearly seventy committee sessions and an eleven-hour filibuster. Benjamin Netanyahu (b. 1949) was absent from the vote. It takes effect on January 1, 2027.

Two days later the Knesset dissolved, having become the first since 1988 to serve a full term, a point the coalition whip made a point of underlining as members adjourned. The election is set for October 27, 2026.

Here is what the coalition knew when it voted. A Maariv poll conducted July 8 and 9, published the following Friday, put the opposition bloc at 62 seats and the coalition at 48. Likud was tied for first with Gadi Eisenkot’s (b. 1960) new party at 22 each. Shas stood at seven, its worst showing since the 2022 election. Eighty-three percent of respondents opposed including the ultra-Orthodox parties in the next coalition. A polling average published on July 15, aggregating seven Israeli series, put Likud narrowly ahead of Eisenkot on party size while leaving the coalition well short of sixty-one. Channel 14’s polling was more favorable, as it consistently is, and the blocs are close enough that no aggregate settles the question. What no reading of the numbers supported was confidence.

Under insurance theory this is backward. A coalition that expects to lose should be buying protection, not liquidating it. It has just handed a freed executive to whoever wins on October 27, and it has scheduled delivery for after the handover.

The resolution is that there is no such actor as the coalition, and this is where the essay earns its keep, because insurance theory’s unit of analysis is a constitution-maker, and in a fragmented parliamentary system that entity does not exist. Three sets of hands passed those laws and they face three different risks.

Take the prime minister’s party first. The risk Netanyahu personally faces is not that his policies will be reversed by a successor. It is a criminal trial. Against that risk a strong court supplies nothing, and an attorney general with binding authority is not insurance but exposure. Insurance theory has no variable for a leader whose downside is a conviction rather than an election, and this is not an oversight I am imputing to Ginsburg. It is a gap he identified himself, which I will come back to.

Take the ultra-Orthodox parties second, because they supplied the decisive votes and they are the cleanest case.

Insurance theory assumes the court is a neutral insurer, available to whichever party finds itself out of power. That assumption holds where a court’s composition rotates with the legislature. It fails where a court has a settled ideological character and a forty-year record against one constituency. For Shas and United Torah Judaism the Israeli Supreme Court has been the antagonist in nearly every fight that matters: yeshiva deferments from Ressler in 1988 through the unanimous nine-justice ruling of June 25, 2024 ordering conscription and cutting subsidies, and beyond that conversion, kashrut certification, Sabbath commerce, and funding formulas. A bloc in that position does not hold a policy. It faces a recurring claim against it. Its expected return from judicial power is negative in every state of the world, including the states in which it loses the election, because losing the election does not turn the court into its protector.

So the Haredi parties sold. And notice that they sold while facing high uncertainty, which is the configuration under which insurance theory says buyers appear. Shas at seven seats and four voters in five opposed to their inclusion in the next government is not a picture of confidence. They weakened the court anyway, and under a corrected version of the theory that is rational, because the theory’s prediction depends on the court being an insurer and for them it is not.

And then the third thing, which is that the same bloc did buy insurance that week, in the form insurance theory predicts, at the moment it predicts.

On Monday, July 13, the Knesset passed Basic Law: Torah Study, sixty-three to fifty-two, placing Torah study in the constitutional layer where a future court must reach past a Basic Law to touch it. On Tuesday, July 14, it suspended arrest, investigation and prosecution of qualifying yeshiva students through November 30, 2026, fifty-eight to fifty-four. The Wednesday vote nearly collapsed when Moshe Gafni (b. 1952) held out over funding for ultra-Orthodox kindergarten teachers, and Rothman stretched a speech at the podium while Netanyahu put the kindergarten item on the cabinet agenda.

Constitutional entrenchment of a sectoral interest, executed by a bloc facing an uncertain election, weeks before losing the ability to legislate. That is the insurance thesis operating with textbook precision. The Wednesday law was the premium.

So the same week contains the theory’s confirmation and its refutation, and which one you see depends entirely on whether you model the coalition as one actor or three.

The delayed commencement remains the hardest fact, and I do not think anyone can currently explain it.

If the coalition believes the polls, January 1, 2027 hands a freed executive to Eisenkot or to a Bennett-Lapid formation. If it does not believe the polls, there was no reason to delay. Three readings are available. The coalition genuinely expects to win, and Channel 14’s numbers are the ones its members read. Or it expects that repealing a law is harder than passing one, so the rule survives a term in opposition. Or the members who cared most about the timing were the ones with the least at stake in who wins, because the ultra-Orthodox parties have been in most Israeli governments of the last four decades regardless of which bloc led them, and a rule that frees the executive frees whoever they sit with next.

The third reading is the one I would investigate first. It is also the one insurance theory cannot generate, because it requires an actor whose electoral uncertainty is high and whose governing uncertainty is low, and the model has one variable where two are needed.

The most useful thing I can report is what Ginsburg did when he looked at Israel.

He wrote about it in 2023, in the Israel Law Review, under the title “The Long Hand of Anti-Corruption.” His argument there is that the empowerment of the Israeli legal system has origins in the policing of corruption, that the global anti-corruption movement has reshaped politics in many countries by routing power to prosecutors and courts, and that this history makes control of the Supreme Court unusually valuable to a governing coalition with its own exposure. He also thought the Israeli particulars might work in the court’s favor in the confrontation.

He did not use insurance theory. The author of the insurance thesis, confronted with the case, reached for a different explanation, and the explanation he reached for is precisely the variable I have been describing as missing: what happens when the thing a politician needs protection from is a prosecutor rather than a successor.

That is evidence about the theory’s scope rather than about Israel, and it is the sort of evidence a scholar rarely provides about his own work. Insurance theory explains why courts get built. It was not designed to explain why they get dismantled, and its author, given the chance to extend it, chose not to.

The empirical literature confirms the boundary. The Ginsburg and Versteeg dataset codes adoption of constitutional review across 204 countries and two centuries. There is no comparable dataset on the weakening of review, no coding of the political configurations under which legislatures cut standing, abolish reasonableness, unbind attorneys general, or immunize constitutional amendments. Every empirical claim in this literature is a claim about buying. The Israeli case of the last three years is about selling, and it sits in a gap.

So the accounting.

On the design moments the theory holds, and holds better on Israel than on most places, because Israel supplies a series of them rather than one. A dominant-party Knesset in 1958 wrote Basic Law: The Knesset with no judicial teeth. A Knesset elected 40 to 39, in the last fortnight of its life, having just lived through the collapse of a government, bought rights entrenchment and electoral reform in fifteen days. A confident right-wing majority in 2018 wrote a majoritarian Nation-State Law. A coalition of sixty-four in 2023 moved to cut the court’s reasonableness power. Across seventy years the pattern tracks the arithmetic, and any scholar who wants to test insurance theory properly has, in the Israeli Basic Laws, a dozen observations of the same country under varying conditions that nobody has assembled.

On the dismantling the theory has nothing, and the reason is structural rather than accidental. Demolition in a parliamentary system is done by coalitions, and coalitions contain members whose risk profiles point in opposite directions. One party’s leader needed protection from a prosecutor. Another bloc needed protection from the court and could not get it from the court. A third set of members went along. The aggregate looks like a group liquidating its own insurance while its house burns, and it is nothing of the kind. It is three parties each buying what covers its own risk, and paying each other in the only currency available.

There is a test coming, and it is clean. On January 1, 2027 a government will take office holding an executive freed from its attorney general, and on present polling it will not be the government that passed the law. If insurance theory describes anything about how politicians treat legal constraint, the first move of an incoming coalition facing an uncertain future should be to reinsure: to restore the binding opinion, or to build something else that binds its own successors.

If it does that, the theory survives its extension to demolition. If it keeps the power and uses it, then what the July 2026 law revealed was not one coalition’s calculation about one election but a change in what Israeli politicians believe a legal constraint is for, and no theory built on the arithmetic of seats will explain it.