The Twelve Bridges

The framework

The framework begins from one claim: legal reasoning is constrained, not creative. A lawyer's "therefore" cannot float free of precedent. It must rest on a recognized mechanism by which courts move from existing authority to a new conclusion, or it is advocacy rather than established law.

The working papers name twelve such mechanisms, called bridges, and group them into four categories. The numbering predates the grouping, which is why the numbers are not consecutive within each category.

Each bridge carries conditions. Bridge 1 requires that the cited language be holding rather than dicta, that the new facts satisfy each element, and that the holding has not been overruled. Bridge 2 requires that a court performed the synthesis, not the advocate or the AI system. Bridge 4 requires that every case sit in the same doctrinal silo. Bridge 8 requires an enumeration that is genuinely exhaustive. Bridge 11 requires that the policy rationale was stated by the court, not inferred by the advocate.

Eight fault patterns complement the bridges. Each mimics a legitimate bridge while violating one of its conditions:

Parallel checking. A validator runs all twelve bridge checks and all eight fault detectors on a proposed synthesis at once, because the two operations are independent. A synthesis can fail every bridge without triggering a fault (novel advocacy, not invalid reasoning), or appear to satisfy a bridge while a fault shows the bridge is illusory. The output rates Legal Authority (binding, persuasive, argumentative), Synthesis Quality (supported, qualified, unsupported), and Advocacy Viability. The central distinction it preserves is between "this is not law" and "this is bad law."

Why twelve. The completeness claim rests on the view that precedent-based reasoning falls into three modes: rule application (Bridges 1, 2, 5, 6, 8, 10), analogical and principled reasoning (Bridges 3, 4, 7, 9, 11), and balancing (Bridge 12). A new bridge, on this view, would specialize an existing mode rather than add a new one.

How to apply it

The working papers' own test case is a synthesis of the kind a language model produces fluently:

Brown v. Board of Education, Gideon v. Wainwright, and Roe v. Wade all establish that government power over individuals is presumptively limited; therefore, regulations affecting equality, justice access, or personal autonomy are presumptively invalid.

Every citation is real and the characterizations are broadly accurate. Run the parallel check:

• Bridge 1 fails: no single holding states the principle.
• Bridge 2 fails: no court has synthesized these three cases.
• Bridges 3 and 4 fail: the cases rest on different provisions (Equal Protection, the Sixth Amendment, Due Process), so there is neither a functional parallel nor a shared silo.
• Bridge 7 fails: no hierarchy orders the cases. Bridge 11 fails: the cases state different rationales. The remaining bridges do not apply.
• Fault 1 fires: the cases are grouped by outcome, not doctrine.

The report reads: Legal Authority, argumentative (0 of 12 bridges); Synthesis Quality, unsupported; Advocacy Viability, possible but requiring substantial doctrinal work. The recommendation is not to delete the idea but to relabel it: do not present it as established law; narrow it to a single doctrine; search for a synthesizing case; or state openly that the argument is novel. (A further problem, which the papers acknowledge elsewhere: after Dobbs, Roe is no longer good law at all.)

Contrast a synthesis that crosses a bridge. In Batson v. Kentucky the Supreme Court barred race-based peremptory strikes because they undermine the impartial jury; in J.E.B. v. Alabama it extended that principle to gender-based strikes. The working papers treat this as Bridge 3 done correctly: the functional reason (identity-based exclusion impairs impartiality) was articulated, applies to the new facts, and the extension was drawn by the Court itself. An associate citing the pair would name the bridge, quote the rationale, and show the match. The discipline in both examples is the same question, asked every time the word "therefore" appears: which bridge am I crossing, and does the case law truly support it?

Known limitations and critiques

Structure is not substance. The framework checks whether an argument follows a recognized pattern, much as a type-checker checks that a program is well formed. A synthesis that crosses a valid bridge can still lose: the analogy may be weak, the rationale distinguishable, the facts different in ways that matter.

No ground truth yet. There is no labeled dataset of valid and invalid legal syntheses. The papers call for expert-annotated benchmarks; until they exist, claims about detection accuracy remain untested.

The completeness claim is contestable. Mapping twelve bridges onto three modes of precedential reasoning is a jurisprudential argument, not a proof. The papers acknowledge that refinement may be needed as new forms of reasoning emerge.

Common-law scope. The taxonomy is built for Anglo-American precedent. Civil-law systems may need a different or supplementary taxonomy.

Internal tensions. Fault 7 treats implicit overruling claims as invalid, while the Institute's Case Treatment Taxonomy includes an "abrogated" category for implicit overruling through incompatible holdings. The two need reconciling. The working papers are also not yet consistent in labels: one relabels the bridges in graph terms, and another cross-references fault numbers beyond the eight listed here. This page follows the names in the framework paper and the synthesis-validation memo.

Hard extraction problems. Separating holding from dicta, identifying functional analogies, and dating when a characterization first appeared are open natural-language problems. The framework specifies the checks; it does not make them easy.

History. The twelve-bridge version grew out of an earlier four-bridge framework, and it should be expected to change again.