1/2
A two-stage architecture with a hard boundary between AI and computation.
A large language model identifies every 'Structural Legal Element' across the corpus (the contracts, statutes, institutions, entities, etc and relationships that carry legal effect), then decomposes each into its 'Instruments':
and also;
This takes about 120 seconds.
With no AI in the computation, the pipeline analyses what was extracted:
This takes under 5 seconds.
Every output links to the Structural Legal Element and specific provision in the specific Instrument. The full audit trail is preserved. The process is fully traceable for what was extracted, what was discarded, and why. Nothing is summarised, approximated, or inferred without provenance.
Check out this great video
The hard boundary between AI and mathematics is deliberate. AI is effective at reading natural language. It is not reliable for measurement, comparison, or structural computation. The deterministic pipeline is precise, repeatable, and auditable, but it cannot read legal text. By separating the two, the system provides natural language understanding at the front, mathematical precision at the back, and a clear line between them that you can inspect and verify.
The system's architecture of deterministic computation, full audit trail, hard boundary between AI and measurement, human oversight preserved at every stage, was all designed for trustworthiness.
It also means Fabric & Ledger's approach aligns with:
Meeting transparency, traceability, accuracy measurement, and human oversight requirements for high-risk AI systems
The first legally binding international AI treaty, signed by the UK, US, and EU in September 2024.
The MOD's principal policy framework for safe and responsible AI adoption.
Copyright © 2026 Fabric & Ledger - All Rights Reserved.