The AI beam sweeps clause by clause, as an NLP contract-review model would: each
clause is turned into a feature vector (ambiguity, obligation strength, liability
exposure) and scored by a small classifier. Flagged clauses are then checked against
a linked-data graph of prior case law for a matching precedent.
Risk score: R(c) = σ( Σ wᵢ·xᵢ(c) ) flag if R(c) > τ
Precedent: match if cos(θ) = (v_c·v_p)/(|v_c||v_p|) > 0.8
Compliance: score = (scanned − flagged) / scanned × 100%
- Scan speed — how fast the AI reads through the document.
- Risk threshold τ — the compliance cut-off; raise it to flag only the riskiest clauses.
- Document size — number of clauses in the contract being reviewed.
- Precedent search — draws a link from a flagged clause to the closest matching case-law node when the similarity check passes.