In 1972, Guido Calabresi and A. Douglas Melamed published a now-celebrated account of how the law protects entitlements. An entitlement, they observed, can be secured in either of two ways: (1) by a property rule, which lets its holder refuse to part with it at any price; or (2) by a liability rule, which lets someone else override it on payment of damages. Either side of a dispute, in turn, may be the party who holds the entitlement. When they arranged those possibilities into a single grid, a fourth combination appeared that courts had almost entirely overlooked: a polluter might be entitled to pollute, and yet a neighbor could shut that pollution down, provided she paid the polluter to stop. “The very statement of these rules in the context of our framework,” they wrote, “suggests that something is missing.”[1]
The missing piece was a fourth rule. What is easy to miss about it is that the rule was not so much invented as revealed: it emerged from the act of arranging familiar remedies into a more orderly structure, until the empty square in the grid became impossible to ignore. The doctrine was the result; the organization was the innovation. That pattern (i.e., progress arriving through arrangement rather than invention) is older, and more consequential, than it first appears.
Consider arguably the most consequential example of all: Roman law, the foundation on which the entire continental European legal tradition would later be built. Roman law’s survival into the modern world was secured by an act of compilation and organization. In the sixth century, the jurists of the emperor Justinian gathered a thousand years of scattered legal writing into the Digest, an anthology that the historian Peter Stein describes as roughly one and a half times the length of the Bible, distilled from some twenty times that much material. As Stein observes, “The whole collection marked the culmination of a millennium of legal development. Without Justinian’s compilation we would know very little about the earlier law.”[2] The act of gathering and ordering was itself the revolution.
The compilation then fell quiet for five centuries, sitting in Italian libraries largely unread, its bulk and difficulty deterring the few who might have used it. What revived it, around the turn of the twelfth century, was once again a matter of method rather than doctrine. The glossators of Bologna worked entirely within Justinian’s text, linking each passage to every other that bore on the same question, laying out the arguments on either side, and reconciling the contradictions that centuries of accumulation had left behind. In time this connective apparatus grew so authoritative in its own right that lawyers would say of the standard gloss, “What the Gloss does not recognize, the Court does not recognize.”[3] A new way of organizing old texts had set the course of European law for the next eight hundred years.
Sometimes the frontier of the law is a genuinely new idea. At least as often, the frontier is a better way to compile, connect, and surface what is already on the books.
This is the conviction that has guided our own work at Filevine, though we did not begin with caselaw. We began with the case file (like the medical records, depositions, and discovery that fill a single litigation matter), where we found that embeddings and semantic search could surface the document or passage a lawyer needed, even when it shared none of the obvious search terms. Having seen what those techniques could do for the evidence inside one case, we began to wonder whether the same methods might be turned outward, onto the far larger and more tangled body of published caselaw.
Over the past year, our machine-learning team has pursued two complementary approaches to that question. The first builds a citation graph (a map of every opinion that cites every other) and walks outward from a given case to find the decisions connected to it. The second works semantically: it renders the paragraphs of each opinion as mathematical representations, or embeddings, and measures how near two of them sit in meaning, so that opinions addressing the same question can be linked even when neither cites the other. The Commerce Clause decisions Wickard v. Filburn, about a farmer’s wheat, and United States v. Lopez, about a gun near a school, make the point. Because Lopez cites Wickard directly, the citation graph already ties them together; but the semantic approach would also recognize that the two engage the same Commerce Clause questions, even if the citation graph did not expressly link the cases. That is exactly what is required for the many opinions that speak to one another without ever citing by name.
That research now lives inside LOIS, Filevine’s tool for legal research. Given a holding that matters to a case, it gathers the opinions that bear on that holding along both pathways.
None of this rewrites the law of precedent, and it does not set out to. It is the humbler kind of advance: a better way of arranging what already exists, so that what matters is no longer lost in the sheer volume of the law.
See it on a holding you rely on. LOIS reads the law the way a careful associate would, and brings back the authority that bears on your case, whether or not anyone thought to cite it. Explore LOIS legal research →
[1] Guido Calabresi & A. Douglas Melamed, Property Rules, Liability Rules, and Inalienability: One View of the Cathedral, 85 Harv. L. Rev. 1089, 1116 (1972).
[2] Peter Stein, Roman Law in European History 33, 35 (1999).
[3] Id. at 49.