Proposition-level extraction
Decomposing judicial opinions into discrete, independently citable legal propositions (holdings, standards, burden allocations, doctrinal constraints), each linked to its source text and stored as its own record, instead of indexing whole documents or text chunks.
First used in The Promise Fulfilled
An attorney does not cite a case; she cites a proposition and offers a case as authority for it. A single opinion routinely contains many propositions that favor different parties, address different stages and have different futures.
Institute papers argue that document chunking is not a coarser version of proposition extraction but a different representational choice. A passage may straddle several propositions, or capture a partial holding that cannot be evaluated without context it does not contain.
Proposition-level records are the unit to which the other requirements for structural reliability attach: typed adversarial metadata, authority-health composites, jurisdiction modeling and rejected-argument indexing. The link to source text is what distinguishes extraction from paraphrase, which matters when the downstream use is legal argument.
Related terms
Related frameworks
The Proposition-Usability Model
A cited proposition is usable only when it is still valid, governs in the forum and procedural posture, helps the represented party, and has not already been rejected in the jurisdiction.
Version Control for Law
Legal interpretation can be governed like a versioned codebase: enacted text is the protected main branch, opinions are signed commits, dissents are preserved branches, and overruling is a force push allowed only through logged, gated conditions.