Contract Intelligence for the LIBOR Transition

A contract is registered and its format classified. A digital document is parsed for free; a scanned image is read by the metered vision stage, which judges how legibly the page scans rather than reproducing it verbatim. The clauses are segmented with page and coordinate provenance preserved, analysed for benchmark references and fallback language by the second metered stage, and reduced to a set of structured fields - benchmark, fallback presence, dates, parties, amendment indicator.

A confidence score, gated to substantive pages, routes the extraction to one of three lanes, and a remediation record captures whether transition is required, at what priority, and whether it is amendment-ready.

Confidence is gated by two floors. Above the auto-accept floor the read is recorded automatically; between the floors it goes to the review queue for a human to confirm; below the review floor it lands in the exception queue. An uncertain read is a manual-check flag, never a fabricated certainty - the system is built to say “I'm not sure, look at this” rather than to guess.

Validated records roll up into the amendment register, ranked by benchmark × materiality so the highest-exposure contracts surface first - turning a pile of paper into a prioritised, auditable remediation queue.