Structural reliability
A research system is structurally reliable for a class of error when the information needed to catch that error is recorded in its stored representation and consulted by a defined, repeatable check, so that the error is impossible or reliably detectable.
First used in The Promise Fulfilled
The claim is deliberately modest. Structural reliability does not mean a tool never errs. It means the error class is handled by a procedure someone could inspect, rerun and audit.
Its counterpart is opportunistic reliability. The pair converts an unanswerable evaluation question (is this tool accurate?) into an answerable one: for which classes of error does the tool have a defined detection procedure, and for which is it relying on what retrieval happens to surface? Vendors can answer that question, and a refusal to answer is itself information.
Institute papers derive five minimum representation requirements for structural reliability in legal research: proposition-level extraction, typed adversarial metadata, authority-health composites, jurisdiction-hierarchy modeling, and rejected-argument indexing.
Related terms
Related frameworks
The Four Laws of System Design for Computational Law
A legal AI system must never overstate its confidence, must explain every output, must ground every legal proposition in validated authority, and must keep its reasoning components testable and versioned, with each law yielding to the laws above it.
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.