A majority is not the whole judgment.

The local parser reads validator, outcome and reasons from each entered line. In the illustrated fixture, two lines say Grant and one says Deny. Bars compare two majority entries with one dissenting entry at one hundred fifty pixels per ruling. The source chooses the most frequent outcome label, not a legally valid or fair judgment. Labels are arbitrary user text; the tool does not authenticate validators, verify a court decision, run independent models or execute GenLayer. Three rulings produce three unordered pairs. In the fixture, two identical reasons normalize to supplier, pay and buyer: three shared terms over three union terms gives a Jaccard score of one. Comparing either with the disjoint refund, required and now set gives zero shared over six union terms, or zero. At four hundred pixels per unit score, bars show four hundred and zero pixels. Token overlap does not verify evidence, reasoning quality, independence or correctness; it measures these source token sets only. Both Grant fixture lines contain the identical sentence Supplier must pay buyer. The source collects sentences containing obligation keywords and deduplicates majority strings for display. Bars compare two collected copies with one displayed sentence at one hundred fifty pixels per count. Minority obligation sentences are shown separately, not enforced. Input text saves locally; reload restores it without rerunning analysis. Reset hides the export and clears the majority and bars but leaves other old text untouched. The JSON records local parsing and explicit limitations, not binding legal obligations.