Research
Methodology
How Cessions approaches AI-assisted contract intelligence.
Most AI extraction tools evaluate whether a citation is present — whether the system attached a source reference to its answer. That is the wrong bar for reinsurance treaty review. A system can produce a real, verifiable clause reference while mischaracterizing what the clause actually says. We call this the gap between citation presence and citation correctness.
Citation correctness means the cited span, read on its own terms, supports the specific claim attached to it — including scope (which peril, which layer, which time period) and polarity (an inclusion is not an exclusion). This is the metric that governs how Cessions is built and how it should be evaluated.
The Cessions architecture separates extraction into three stages: retrieval-grounded extraction that constrains the model to retrieved document spans, a second-pass verification check that examines every claim against its cited text for scope and polarity match, and a human review queue where every finding lands as a draft requiring explicit reviewer approval before it flows into generated work product.
The failure mode, made concrete
A synthetic treaty. One clause. Two documents. Two very different answers.
Citation present. Citation incorrect.
“War-related losses are excluded.”
— cites Article 12
The reinstatement in Endorsement No. 4 is omitted. The finding is technically cited but materially misstates current coverage.
Citation present AND correct. Multi-document synthesis performed.
“The war exclusion applies except for cyber-related acts of foreign state actors, reinstated up to a $10,000,000 sublimit.”
— cites Article 12 AND Endorsement No. 4, verified for scope and polarity
Both the original exclusion and the endorsement carve-back are cited together. The net coverage position is stated correctly. This is the multi-document synthesis case that presence-only evaluation would miss.
The three-stage pipeline
Stage 1
Retrieval-grounded extraction
Documents are chunked at clause level using structural pattern recognition (article headers, endorsement markers, section boundaries). Per-category keyword retrieval constrains the extraction model to relevant spans only — it may not generate findings from text it was not given. Every finding must include an exact verbatim quote from the source.
Stage 2
Second-pass verification
Every claim is checked against its cited span for three specific failure modes: scope mismatch (the claim overstates or understates what the span covers), polarity reversal (an inclusion characterized as an exclusion or vice versa), and omission of a material qualifier (a condition, sublimit, or exception present in the span but missing from the claim). Flagged findings are shown to the reviewer with their flag visible — not hidden.
Stage 3
Human review queue
Every finding lands as a draft requiring explicit reviewer approval. Approval is a recorded human action with reviewer identity and timestamp in an append-only audit log. Generated work product marks unapproved findings as “(pending review)” and carries the footer: AI-assisted draft generated by Cessions for professional human review. Human review stays in control.
Output
Generated work product
Renewal Review Packets, Treaty Comparison Memos, Broker Question Lists, Exception Reports, Executive Briefs, Internal Review Packets — each built from approved findings with source citations. Every document is an AI-assisted draft for professional review.
Read the methodology paper
The full architecture, evaluation protocol, illustrative example, and pre-registered kill criterion are described in the working paper.
Download the methodology paper (PDF)Sammy Orangkhadivi, Cessions, Inc. — July 2026.
Early methodology paper. Quantitative results are pending pilot testing on real treaty documents. This paper does not claim peer-reviewed status.