We tested nine AI systems on a Lebanese waqf matter. Most got the decisive article wrong.

One real Arabic client instruction, a 1947 Lebanese statute, and every citation checked article by article. Justinian V2 scored 88. The highest-scoring general model reached 82.