Arxo Law, enregistré

Cinq conversations enregistrées avec une tranche de droit formalisé. Le modèle lit la question et choisit où chercher ; chaque réponse, chaque refus et chaque empreinte ci-dessous viennent du moteur.

• Hardship, end to end

  A question in plain words becomes an address, the address becomes a rule, the rule names the facts it needs — and two of those facts turn out not to be the model's to supply.

• The engine refuses the model

  The most useful thing this system does is say no. A plausible-sounding legal term, an argument too few — both come back as refusals with an address, not as a confident answer.

• The word only a tribunal may say

  Article 2.1.20 turns on whether a standard term is one the other party "could not reasonably have expected". The model is asked to settle that. It cannot, and the engine says who can.

• The rule is in the model, the number is not

  Article 7.4.9 names how the rate of interest is chosen, not what it is. A model of the instrument that invented a percentage would be worse than one that says where the number has to come from.

• The same answer, twice

  An answer worth citing has to survive being asked again. The model hands back only the question and the hashes it was given; the server runs the law a second time and compares before it will show anyone a chain of reasoning.